This Crypto AI & Robotics newsletter consists of the following:

1. Snippet Partner: Umia Finance

2. What Robinhood Chain actually is

3. Ten projects ranked by live usage

4. The PONS, LONG(), Stonkbroker, GIGA, Lighter opportunities

5. The mainnet launch calendar

6. A practical participation guide

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

This week’s issue is brought to you by Umia Finance, the operating stack for token-native ventures.

Umia is taking its venture-creation stack live on testnet, beginning with project discovery and onboarding its founding cohort. Founders can submit a GitHub repository through Umia Score, verify ownership and establish an initial valuation signal before launching:

The full stack combines a legal wrapper, token, non-custodial treasury and community fundraising through Uniswap v4 Continuous Clearing Auctions. Once live, major treasury decisions are resolved through decision markets, while Umia’s MetaLeX-powered legal structure gives those outcomes real authority.

The timing matters for agentic ventures. AI lets small teams build products quickly; Umia compresses the capital formation, ownership and governance required to turn them into investable businesses.

Explore the testnet or read how the Umia stack works.

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

Robinhood’s Onchain App Store Is Open (Meme - RWA Barbell + agents)

Over the latest 24 hours, users paid Robinhood Chain applications $1.6M in gross fees, retaining $226K as protocol revenue, while Robinhood Chain kept $50K:

Put simply: the app layer collected 32x the chain’s revenue in gross fees and retained 5x the chain’s revenue for itself.

• The 32x figure measures what users paid to apps

• The 5x figure compares what applications and the chain actually retained

This supports a “fat apps, thin chain” thesis, meaning that Robinhood provides inexpensive settlement while applications capture most of the economics through trading, lending, launches & agentic services

That makes Robinhood Chain attractive for deployment and why over the next few months you can expect a flurry of activity, with opportunities for those paying attention

To note: Robinhood Chain has no native token, so the investable surface sits across exchanges, launchpads, lending markets & agentic applications

Fees are what users pay, protocol revenue is what an application keeps and holder revenue (arguably the most focused on metric given the lack of trust in the market currently) is what reaches a token through distributions, buybacks or burns

What Robinhood Chain actually is

Robinhood is a permissionless, EVM-compatible Ethereum Layer 2 built with Arbitrum’s Nitro tech stack. It settles directly to Ethereum, uses ETH for gas and runs on chain ID 4663… technical jargon for an ETH aligned layer 2

The chain now holds $411M of DeFi TVL, $558M of stablecoins, $101M of active RWAs and more than $1B of bridged assets. Seven-day DEX volume reached almost $3B; perps reached $228M, up 130%.

The Meme - RWA barbell

Memes acquire users, stocks anchor value and agents automate the portfolio between them. Vlad (Co-Founder and CEO, RH) has doubled down on this meme-RWA barbell thesis:

Pure memes like CashCat have captured decent attention with artists like Beeple posting about it:

But as is the case with any pure attention based speculation instrument it’ll be volatile:

… and it’s actually the interesting mechanisms that lean into the robinhood thesis that have really piqued my attention; namely Stonkbrokers and LONG() with plenty more fresh bouts of innovation shining through with tokenholders at the heart of that value accrual (refreshing I know!)

So how does the landscape look today, and what opportunities exist in the not too distant pipeline? I’ll explain below:

1) PONS: the strongest small-cap fee anomaly

Ponsinomics (nice name) measures protocol receipts from collector-wallet transfers and launch events. The latest week produced $570K… annualised, that’s $30M against a $20M market cap: which equals 68% of annualised receipts.

When you strip out the two gamified mining tokens (Minepea + SLVR), Pons is the leader in returning revenue to holders:

On 25 July, Pons handled 64% of launchpad volume and 64% of token deployments across the chain:

Current policy directs roughly 80% of receipts into TWAP buybacks and burns. The weekly pace implies $24M of annual buying, equal to 118% of market cap. Burns have removed 24% of supply.

Pons V2 is now live with Uniswap V4 and stock-token quote pairs:

If they continue with the 80% fee value accrual to Pons token holders then this is a lucrative asset to watch provided launch volume sustains

2) LONG(): the meme-stock barbell

On 25 July, Artemis measured meme pairs at $297M, or 75% of daily DEX volume. Stock-token pairs were $36M, or 9%. It’s clear these are making a difference to stock trading onchain also:

LONG() links those markets: token such as AI uses NVDA as its quote asset. Every swap creates NVDA turnover and the pool holds NVDA liquidity. The token’s dollar price becomes AI/NVDA multiplied by NVDA/USD, creating two sources of volatility.

AI/NVDA currently has a roughly $2M market cap, $610K liquidity, $629K daily volume and 6,100 holders. Daily volume equals 36% of market cap. LONG() reported $32M of cumulative stock-token volume on 22 July and has since announced Index Expansion into sector baskets:

Bankr adds natural-language launches and trading. Its creator-fee model explicitly lets agents fund their own compute. This creates a loop: memes distribute, stocks build treasury assets and agents deploy the proceeds.

The AI/NVDA pool carries a 48/100 security score and uses a proxy contract, so contract risk sits beside the two-factor market risk.

3) StonkBrokers: the NFT is the stock wallet

LONG() puts the meme and the stock in the same liquidity pool, while StonkBrokers puts the stock inside the meme… this one is novel and worth paying attention to!!

The collection contains 4,444 broker NFTs; each ERC-721 controls an ERC-6551 token-bound account seeded with tokenised stock at mint. The wallet and its stock balances travel with the NFT when it changes hands; the new owner can reactivate it for future distributions. Each broker therefore functions as a transferable onchain portfolio… pretty non-fungible and could move in line with the underlying portfolio?

The Anvil AMM creates the cash-flow loop:

• A broker costs 666,666 STONKBROKER plus a native ETH fee.

• The fee is 10% for a standard swap and 15% for a specific NFT.

• 70% of that ETH funds StockBooster; 30% goes to the protocol.

• Any wallet can then “Clock In”, converting the accumulated ETH into the stock tokens selected by activated broker holders.

Activation creates the token sink. Holders pay between 66,666 and 1,666,666 STONKBROKER for higher distribution weight. At the captured token price, that is roughly $1K to $27K, driving positive price acction to the fungible token also (+8%):

Half of each activation payment burns and half goes to the protocol. A genuine NFT transfer clears the activation, so an active secondary market creates repeat demand for STONKBROKER.

On 4 August, the OpenSea collection showed a 6 ETH floor, 176 ETH of daily NFT volume and 1,116 ETH cumulatively.

Floor multiplied by supply gives a $47M headline valuation. Liquidity behind that mark is thin: only 57 NFTs, or 1% of supply, were listed, while 557 wallets held the collection. Holders equal 13% of supply and the average wallet owns eight brokers.

This is the Stonkbroker flywheel:

Broker Box extends the idea into stock-token distribution. Nine mainnet machines cover GME, AAPL, AMZN, NVDA, GOOGL, MSFT, SLV, SPCX and USO:

• The Certificate Counter swaps ETH into stock and seals it inside another ERC-6551 bearer NFT.

• Each purchase charges $2, split evenly between StockBooster and the treasury.

• Degen Mode adds a 90% design RTP and stock payouts ranging from 70% of stake to 50x; instant cash-out retains a 5% spread.

• This is a gamified acquisition funnel for tokenised stocks with every prize backed by inventory onchain.

The metric worth tracking is stock-token distributions per activated broker.

A dashboard showing Clock In volume, stocks purchased, average payout, active brokers and activation burns would make the productive value visible beside the NFT floor.

The project’s Stonk Exchange, pools and covered calls are scheduled for 29 August, creating the next test: whether the same fee-recycling model can extend beyond NFT trading with options involved also:

4) GIGA: what it is and why it matters

GIGA is a Robinhood-native DEX and liquidity coordination system; it supports stable, volatile and concentrated pools. At Genesis, its controller will redirect emissions every hour toward pools producing the most revenue. This replaces weekly gauge voting and bribes with an automated signal.

The current phase already produced $10M of weekly volume from $522K of TVL. LPs keep 80% of swap fees; the protocol kept $4K from $20K total. Points reward the busiest pools, so part of that volume is farming.

The six-week points program is redeemable for 15M GIGA from a fixed 1B supply. GIGA Genesis is scheduled for 31 August, when the custom emissions system activates.

Two diligence flags matter. The custom controller remains team-controlled and under audit. The official tokenomics page assigns 190M GIGA, or 19%, to the GIGA Machine; the Machine page states 200M, or 20%. That allocation needs a single answer.

5) Lighter: Perps

Lighter is multi-chain but it’s hard to ignore when Perp volume is hitting new all time highs daily:

If the protocol solidifies itself as the number one venue on robinhood chain then we could see some serious competition against the king of perps; Hyperliquid.

It’s pretty clear now that users want to trade tokenized equities and commodities onchain, so offering venues to speculate on derivatives of tangible assets with decent user experiences is becoming increasingly appealing

Where headline activity misleads

• UP trades at a $30M FDV. It generated $14M of weekly volume and $27K of holder revenue, then distributed $519K of incentives. Weekly earnings were -$492K.

• Sushi routed $9M through $216K of TVL and generated $357K of fees. Holders received $5K, or 1%; likely because its more established with greater overheads + a hike in marketing expense from a pretty decent video:

• NOXA generated $872K of fees with $0 protocol and holder revenue.

• RamsesX has a $649K market cap and $452K TVL, with $0 tracked volume or fees.

Fee screenshots show activity but it’s the token value depends on the cash-flow path. So when you’re looking at the DeFi llama fee flow diagrams in search of investible tokens, check the flows back to token holders like for Pons ($2.18m of the $4m):

The mainnet launch calendar

How I would participate

1. Add the network from official docs. Chain ID 4663, ETH gas, official RPC and Blockscout.

2. Bridge a test amount. The canonical bridge takes about ten minutes in and seven days out. Relay and Across provide faster routes.

3. Establish the base layer. Make a small Uniswap swap, inspect Morpho markets and try Arcus (dydx team) spot if eligible.

4. Separate experiments. Use small amounts for Pons V2, LONG()/Bankr and GIGA points. Run Sherwood and Veil on testnet. Keep experimental contracts in a dedicated wallet.

5. Track the same columns weekly. Volume, TVL, protocol revenue, holder revenue, incentives and market cap.

• DeFiLlama ranks apps;

• Entropy Advisors shows network economics, RWAs and Morpho in one dashboard; Ponsinomics covers collector flows;

• Blockscout verifies contracts and transactions.

Robinhood has announced no chain token or chain airdrop. Stock Tokens are tokenised debt securities providing economic exposure to shares

Also set up an agentic trading account and play around with it:

Robinhood Chain’s first month belongs to:

• Uniswap for trading,

• Morpho for deposits, and;

• Pons for token-level value capture.

The deeper opportunity is the meme-stock-agent loop:

• LONG() & StonkBroker have created the pairing model.

• Pons and Bankr are distributing it.

• GIGA is competing to coordinate the liquidity alongside Uniswap.

• Sherwood is building the agent-managed capital layer.

Continually track these (and other emerging) projects for activity and position accordingly as attention and activity grows.

It may even pay to establish what Robinhood typically sets as its core exchange listing requirements in the hopes they start running some of their own native tokens through to their user base:

That’s a wrap for this issue of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. Valuations are approximate snapshots and move quickly. Verify all market data independently before acting.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of three key parts:

1. Snippet Partner: Axis Robotics

2. Theme of the Week: Intelligence Is Becoming an Open Commodity

3. Landscape Analysis: TAO (Macrocosmos, Chutes, Targon, Engy AI), TIG, Walrus + more

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

This week’s issue is brought to you by Axis Robotics, who just raised a $12 million seed round

Language models could learn from an internet already filled with text, images & video. But robots don’t have the same luxury; physical AI needs billions of motion trajectories across different objects, environments & robot bodies

Axis is buildin a compounding data engine around this bottleneck; Its browser-based simulation platform allows distributed contributors to teleoperate robots and generate training trajectories, while its mobile capture product converts human movement into real-world egocentric data

It just revealed its roadmap for the next 12 months:

As traditional robotics labs begin to IPO we could see eye watering capital raises feed what is going to become the largest revamp of the global labour force

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

Intelligence Is Becoming an “Open” Commodity

First Jensen Huang, now Sam Altman & Dario Amodei (kind of) came out in support of open weights (open source AI models)

Within a few days NVIDIA argued that open-weight models are strategic infra for US AI leadership:

Anthropic described capable, but non-dangerous open weights as a “public good” & Moonshot released the largest open-weight model to date

These aren’t fringe open-source advocates:

• NVIDIA sells the infrastructure beneath the entire AI economy, and;

• Anthropic operates one of its most valuable closed model businesses

Both now agree that broader access to model weights increases competition, gives customers greater control and reduces dependence on a single AI lab

The significance for crypto AI is what I believe is underestimated…

If intelligence becomes easier to download, adapt and deploy, value moves toward the scarce inputs required to run and improve it: compute, inference, proprietary data, memory, verification, coordination + payments

Open weights become a strategic issue

NVIDIA’s open-weights paper argues that AI leadership should be measured by whether an ecosystem can diffuse intelligence across startups, universities, factories, hospitals and public institutions

Open weights let organisations inspect a model, fine-tune it and run it on the infra it controls. That makes AI cheaper for many tasks & prevents the knowledge built on top from becoming trapped inside a single provider

Dario Amodei’s response was more cautious on safety, but agreed (albeit hesitantly) with the core economic point:

“Open weights expand access to the AI economy, they strengthen competition … and they give customers greater control.”

Anthropic still wants mandatory testing for sufficiently capable models and targeted action against industrial-scale distillation

Even the leading closed-model lab now accepts that open weights create genuine public and economic value

Kimi K3 made the argument tangible

Moonshot released a 2.8T parameter open-weight model… essentially making a competitive frontier model from Asia a public good:

Frontier labs still retain distribution, safety infrastructure, proprietary data & the ability to spend billions training the next generation

Builders will be able to choose between closed frontier models, open general models and smaller specialised models depending on cost, latency, privacy and control turning model access from a permanent moat into a much more competitive market

NVIDIA understands the flywheel

You’d be mistaken if you believed NVIDIA was supporting open weights as an act of charity…

It expands the number of organisations able to deploy AI on their own infrastructure meaning A LOT more customers to supply compute to:

Cheaper models create more applications, more applications create more inference, more inference creates demand for GPUs, memory, networking & power:

Open weights also make that demand portable, so a model can move between hyperscalers, sovereign clouds, local hardware + decentralised compute networks rather than remaining tied to one API

This is where crypto AI becomes relevant… it coordinates the fragmented markets forming around it, and open intelligence needs open, premissionless infrastructure:

Landscape Analysis: Bittensor, TIG and the Open Intelligence Stack

i) Bittensor (TAO) is commercialising open intelligence

Until recently, the crypto AI response to open weights has been theoretical: decentralised compute and data networks would become useful once high-quality models could move freely between providers

Bittensor has spent the past few years building specialised markets beneath it.. the TAO eco is an incentive framework where subnets define a digital commodity, miners produce it, validators score the output and TAO rewards the contributors judged to be creating value

For those wanting a crash course in how TAO came to be, we’ve reported on the ecosystems investible history here:

Open weights make this structure much more important because training, serving, data collection, verification and privacy can each become competitive markets. We can touch on several eco examples:

a) Training: Macrocosmos IOTA, Subnet 9

Macrocosmos has now started Orion-16B, a single 16 billion parameter model training live across heterogeneous GPUs on three continents.

IOTA splits model layers across permissionless contributors, so no miner holds the full model and no operator controls the run

b) Inference: Chutes, Subnet 64

Open weights are only economically useful if somebody can serve them cheaply and reliably

Chutes turns distributed GPUs into a serverless inference platform where developers can deploy open models behind an API without managing the underlying fleet

The important shift is that Chutes is now being measured on revenue quality rather than subsidised token volume with revenue efficiency improving:

After removing loss-making traffic and tightening its subscription model, Chutes reported revenue per million tokens increasing 38% + revenue per active GPU increasing 45% from the start of February

Revenue has reached roughly $280k per trillion tokens served (picture above), with paying applications including Pax Historia running production workloads showing commercial viability

c) Confidential compute: Targon, Subnet 4

Portability creates a trust problem… Enterprises want cheaper distributed compute without exposing models, data or prompts to unknown operators.

Targon uses confidential VMs, host verification and NVIDIA nvTrust attestation to prove workloads ran inside approved protections without trusting the server owner.

Its new Tower lets anyone connect and rent compute through the Targon network.

d) Other emerging TAO subnets:

Algod recently launched a subnet (Engy AI) for verifiable inference, competing with the likes of Chutes:

ii) Algorithms become a licensable asset

The Innovation Game (TIG) is another expression of the open-intelligence thesis

Innovators submit or optimise algorithms, benchmarkers compete to identify the most efficient methods, and contributor rewards increase as their algorithms are adopted

TIG’s licensing framework is the most interesting part of the mechanism, which I covered in this recent article:

TIG offers two licences:

1. free use if distributed code and relevant data stay open, or

2. paid use in TIG to keep data, implementation and improvements proprietary.

Innovators submit methods, benchmarkers price performance, open users grow the commons and commercial fees fund the next cycle.

The test is whether licence demand can replace emissions. Benchmarks prove usefulness; paying customers prove value.

This creates a cleaner value-capture loop than simply attaching a token to an open-source repository:

iii) Machines need a black box

As agents and robots gain autonomy, accountability becomes the missing layer.

The final output is not enough; companies and regulators will need the mandate, model, data, authorisation, decision path & payment record

This is Khala Research’s Machine Black Box thesis: networks such as Walrus can become a verifiable evidence layer for machine activity, not merely file storage.

iv) Open models become economic actors

McKinsey estimates that agentic commerce could orchestrate $3T - $5T globally by 2030… those agents will need to pay for inference, data, APIs, software and eventually physical services.

We covered the potential agentic payment rail projections in our x402 report:

Crypto becomes the network of permissionless wallets, stablecoins and machine-readable ownership forming part of the new age of financial infrastructure

That’s a wrap for this issue of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. Valuations are approximate snapshots and move quickly. Verify all market data independently before acting.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of three key parts:

1. Snippet Partner: NaoX Protocol

2. Theme of the Week: The Agent Economy Needs Real Token Ownership

3. Landscape Analysis: MetaDAO, Umia, MetaLeX and Token Transparency

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

This week’s issue is brought to you by NaoX, the new identity of Naoris Protocol and a Layer 1 focused on post-quantum blockchain infrastructure

This is a rebrand rather than a network migration. The underlying technology and ecosystem continue unchanged, while the token remains $NAORIS

NaoX is building quantum-resistant cryptography into its architecture from inception, alongside a distributed cybersecurity model where devices validate the security posture of other participants

Its relevance to the agent economy is that autonomous systems will increasingly control wallets and transact without continuous human oversight, making long-term identity, device security and cryptographic resilience increasingly important

Adoption will ultimately depend on developer activity, integrations and real network usage rather than the rebrand itself

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

The Agent Economy Needs Real Token Ownership

AI agents will allocate capital through permissionless crypto rails. Before trillions of machines arrive, token ownership must become transparent, enforceable and machine-readable

AI agents are about to become the largest class of investors crypto has ever seen.

They can create wallets, analyse thousands of assets and deploy capital in seconds, but most tokens still cannot answer the first question any investor should ask:

“What do I actually own?”

If agents are going to fund projects, manage treasuries and govern protocols at machine speed, crypto cannot continue relying on vague promises that value will eventually accrue. Ownership needs to become transparent, enforceable and readable by machines.

That is the focus of this week’s newsletter, and something I’ve been thinking about more deeply over the past few weeks.

Crypto’s Ownership Problem

Crypto is the natural financial infrastructure for AI agents because it is permissionless, programmable and available globally.

An agent cannot easily open a bank account or wait for a compliance department to approve each transaction. It can create a wallet, receive stablecoins, interact with smart contracts and deploy capital according to a predefined mandate.

The rails are ready, but the assets moving across them are often poorly defined.

A typical project may have a foundation issuing the token, a development company employing the team, a DAO controlling part of the treasury and founders or equity investors owning the intellectual property.

The token trades globally, but holders may have no enforceable claim over the revenue, company or assets responsible for its perceived value.

This becomes dangerous when agents need to determine:

Today, those answers are scattered across legal documents, governance forums, block explorers and founder statements.

A human investor might spend hours reconstructing the structure. An agent analysing thousands of assets cannot rely on social context, reputation or “trust me bro” assurances.

Before I get into the few projects disrupting this segment, there’s some background I need to set and explain why I’m covering this now ($ANSEM + Clarity Act):

The $ANSEM Debate

My recent disagreement with Ansem over the $ANSEM token exposed the same problem:

Ansem’s thesis is that attention, storytelling and collective belief have measurable economic value, while a liquid token can create a real-time market for that value.

I agree with the premise… tokens are incredibly effective proxies for intangible value and can continuously price goodwill, something traditional financial statements struggle to recognise.

“Liquid Brand Equity” is something OSF has attempted to tackle with rekt already:

But the honest framing today is that $ANSEM is a creator-backed memecoin acting as a real-time proxy for Ansem’s goodwill.

That may still be valuable, but it is not ownership.

The token does not currently provide a defined claim over sponsorship revenue, intellectual property, media businesses or future products created under the Ansem brand. The market can price its belief that value will eventually accrue, but belief is not an enforceable right.

Leaving the value-accrual mechanism to be invented after the market has priced the promise creates the exact environment in which holders are most vulnerable to exploitation.

A human buyer may knowingly speculate on the creator’s intentions. An autonomous investment agent needs explicit rights that it can verify.

Regulation Is Catching Up

The US is beginning to address part of this problem through the CLARITY Act.

The House passed its version in July 2025, while the Senate Banking Committee advanced revised legislation by 15–9 in May 2026. It still needs to pass the full Senate, be reconciled with the House version and be signed by the President before becoming law.

Today you will have seen a pump in crypto, which was in response to the approval of the ethics package:

The proposed framework separates digital commodities regulated by the CFTC from investment-contract assets overseen by the SEC, while introducing tailored disclosures for certain token distributions.

This helps clarify how a token is regulated, but it does not completely explain what the token owns.

A digital commodity may still give holders no claim over revenue, intellectual property or the company building the product. Agents will eventually need to understand both the regulatory classification and the economic rights before deploying capital.

Who Is Building the Ownership Layer?

1) Umia

Umia extends this model into a capital formation platform for token-native ventures.

Projects launch inside a segregated legal portfolio containing the operating team, intellectual property and treasury:

Founders retain authority over normal operations, while tokenholders govern major strategic and capital-allocation decisions through decision markets.

The objective is to avoid the usual foundation, DAO and development-company split where holders control one component while the valuable assets sit elsewhere.

2) MetaDAO

MetaDAO connects a legal entity, onchain treasury and futarchy-based governance.

MetaDAO LLC holds the intellectual property, while major treasury and token decisions flow through conditional markets that estimate whether a proposal will increase or decrease the token’s value.

The token does not necessarily represent conventional equity. Ownership is expressed through enforceable control over the entity and its assets. The protocol providing this ownership rights is expanding rapidly at 29% month on month:

This gives an agent a structure it can inspect: which assets sit inside the entity, which decisions the token controls and whether subsequent actions follow those rules.

The model is now being tested by operating businesses rather than remaining theoretical; here are two of the clearest early examples:

• Credible Finance: The stablecoin payments business recently completed its $CRED launch with $32.8 million committed and $4 million ultimately raised, placing its IP inside a segregated legal portfolio governed through the ownership coin structure

• Avici Money: The onchain banking and payments project attracted $34.2 million in commitments through its MetaDAO launch, providing an earlier example of a functioning business using public capital formation and tokenholder-controlled treasury governance. The protocol has now passed a million transactions:

3) MetaLeX

MetaLeX’s tagline is:

“we turn startups into robots”

The protocol connects onchain governance to offchain legal enforcement through BORGs, or cyBernetic ORGanizations. Fantastic explainer here:

Smart contracts can control a treasury, but they cannot independently control trademarks, employment agreements, bank accounts or commercial contracts.

A BORG places these assets inside a recognised legal entity whose governing documents embed specific smart-contract controls.

Without this bridge, token governance often stops exactly where the most valuable assets begin.

4) Token Transparency Framework

The Blockworks Token Transparency Framework standardises disclosures covering entity structures, insider allocations, vesting, market makers, liquidity, funding and governance control.

The long-term opportunity is turning these disclosures into an API for autonomous capital:

An agent could reject projects where the development company retains the intellectual property, reduce exposure before major insider unlocks or flag treasury activity that contradicts the issuer’s disclosures.

Transparency does not create ownership, but it makes the absence of ownership visible.

Key Takeaways

Three things matter:

1. Permissionless capital needs programmable rights: agents can already purchase tokens, but they still cannot reliably determine what those tokens represent

2. Governance is not automatically ownership: voting over a treasury means little if founders or equity investors retain the revenue, intellectual property and contract controls

3. Disclosure needs to become machine-readable: entity structures, insider wallets, unlocks, revenue flows and governance powers must be available in a format agents can continuously verify

Crypto has already made capital formation permissionless. The next phase is making ownership explicit.

MetaDAO, Umia and MetaLeX are experimenting with enforceable tokenholder control, while the Token Transparency Framework is turning fragmented information into something machines can process.

This matters now because agents will eventually fund projects, manage treasuries and trade ownership claims with other agents at a speed no human market can match.

Before trillions of agents begin allocating capital onchain, crypto needs a clear answer to its most basic question:

What does the tokenholder actually own?

That’s a wrap for this issue of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. Valuations are approximate snapshots and move quickly. Verify all market data independently before acting.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of three key parts:

1. Snippet Partner: Walrus

2. Theme of the Week: Neobank Customers are no Longer “Just” Human

3. Landscape Analysis: The financial stack being built for autonomous software

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

Last week, Khala Research published our deep dive into Walrus and the growing market for persistent AI memory and the timing couldn’t be better:

As AI agents begin managing money, memory becomes financial infrastructure.

A financial agent needs to remember why it was allowed to make that payment.

• Which merchants are approved?

• How much can it spend?

• Has this invoice already been paid?

• Which subscriptions renew next week?

• Who authorised this expense?

Those decisions shouldn’t disappear every time an LLM starts a new conversation, which is where Walrus becomes interesting.

Persistent, portable and verifiable memory allows agents to build a financial history rather than simply execute individual transactions. Instead of repeatedly asking humans for context, agents can carry forward policies, approvals and previous decisions across applications and workflows.

If wallets become the bank account for AI agents, persistent memory becomes the financial record that sits behind it and verifying that financial record onchain is a far superior record keeping system.

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

Neobank Customers are no Longer “Just” Human

Digital-first banks already serve more than 1.4 billion customer accounts, yet those accounts were almost entirely designed around people.

AI agents are now beginning to purchase data, reserve compute, pay for software and complete commercial tasks, creating demand for financial accounts that software can operate within rules set by its owner.

The most important point is that accounts will understate the commercial opportunity. Even before agent numbers rival human customers, their transaction volume could become material because software can buy services continuously.

“Over the past seven days x402 and MPP combined did about $87k in total volume from about 12k unique wallets” — Cuy Sheffield (VISA)

While this $4.5M in annualized volume may seem trivial now, it’ll likely be magnitudes higher in the coming years. Particularly when those 12k wallets are likely to be closer to 12b wallets over the next 5-10 years.

This is why MPP, programmable cards and agentic credit matter: they allow neobanks to serve software as an active economic customer, rather than merely process a payment initiated by a person.

The potential scale is difficult to overstate… Cloudflare’s network data recently showed automated traffic generating approximately 60% of requests for HTML content, overtaking human traffic for the first time:

Most of this activity is conventional automation rather than autonomous AI commerce, but the crossover demonstrates how quickly software can become the dominant participant once digital infrastructure is built around it.

Payments are likely to follow the same direction. Visa already operates more than 130 stablecoin-funded card programmes across over 50 countries, with a plethora of AI initiatives being rolled out:

RedotPay has accumulated more than eight million (self reported) users and is connecting its payment network to Tempo’s Machine Payments Protocol.

Cards, wallets and neobank accounts are gradually being redesigned so that an agent can receive a budget, follow a spending policy and transact without requiring a human to approve every routine purchase.

For neobanks, this creates an entirely new customer base whose value may eventually be measured less by account numbers than by activity. A person might make several financial decisions each day, while an enterprise agent could purchase APIs, data, compute and services continuously.

The institutions that become the financial operating layer for these agents could serve customers that transact globally, remain active around the clock and operate at a frequency the consumer-banking model was never designed to support.

Building the Self-Sovereign Economic Stack

The opportunity extends beyond giving an agent a debit card.

A financially capable agent needs an account through which it can receive revenue, liquidity that can move globally, controlled access to conventional merchants, a record of its obligations and, eventually, a means of borrowing capital.

Several pieces of this stack are now appearing across crypto:

a) Tria combines a self-custodial account with card access and BestPath, its cross-chain routing infrastructure:

Tria explicitly describes AI agents as a target user for programmable payment rails, with a roadmap covering autonomous routing, settlement and asset management across chains.

b) Plasma One (use code “INSURA”) approaches the market through a stablecoin-native account combining card spending, yield and subscription benefits. My personal favourite is the close to 100% cashback on AI products - I now run all my AI subs through Plasma:

c) Ether.fi Cash uses crypto collateral to provide spending capacity without requiring the holder to sell the underlying position. It’s essentially a credit line which in turn will benefit the billions of new (non human) users that will enter the market:

d) Gnosis Pay and MetaMask Card connect smart accounts and self-custodial wallets to established card networks. Although admittedly these lack the same level of UX compared to Plasma, EtherFi and Tria

e) RedotPay has moved furthest towards a direct consumer implementation. Its integration with Tempo’s Machine Payments Protocol (MPP) allows agents to complete a purchase process and settle the transaction using stablecoins.

RedotPay says the integration gives merchants access to its existing base of more than eight million users across over 100 countries, while a downloadable payment skill forms part of the next phase.

f) Tempo’s MPP is strategically important because it treats software as a potential payment customer. The agent can receive a controlled payment capability and act within the limits specified by its owner.

This opens a large market for card issuers and payment providers because agents may transact far more frequently than people, particularly when purchasing API access, compute, data and other machine-readable services

g) x402 is an alternative to MPP and offers the payments rails that have been missing from the internet for 30 years. More can be read up on our previous report:

h) I’ve touched on credit lines earlier, but this is a protocol specifically built for agentic credit: Agentics.credit is testing whether an agent can develop a financial reputation based on verified trading behaviour. Its current 90-day paper-trading campaign scores agents using profitability, drawdown, consistency, longevity and win rate.

Qualifying agents are intended to graduate to constrained, unsecured credit lines where capital remains inside smart contracts and can only be deployed to approved venues. The amount of credit extended, depends on the daily rescoring mechanism:

The project remains experimental, but the concept is significant. An agent with a credible operating history could eventually receive better payment limits, trade credit or access to capital without requiring its owner to overcollateralise every transaction.

The Crypto-Linked Card Landscape

The following table (linked) covers 25 prominent crypto-linked card programmes and their associated agentic maturity:

The final column distinguishes products with a direct agent initiative from those whose programmable accounts may support agents in the future.

Note: “No announced integration” means I found no public, product-level agent functionality, rather than suggesting that the issuer has no internal work underway.

The distinction between these products is varied:

• RedotPay already has a public agent-payment integration

• Tria is building “Best Path” infra explicitly intended for autonomous agents.

• Several self-custodial providers have the account architecture needed for restricted delegation, although they have yet to launch a card product specifically for software

The larger centralised programmes remain relevant because they have established issuing relationships, compliance systems and distribution. If agentic payments reach mainstream adoption, existing card providers may add programmable credentials faster than entirely new issuers can reproduce their regulatory footprint.

This comparison explains why the addressable market cannot be measured solely through the number of agents; transaction frequency could become more consequential than customer count… A single enterprise agent may authorise recurring purchases across hundreds of software tools, while a trading or infrastructure agent could initiate thousands of financial actions in a day

Final Thoughts

The internet is already dominated by software-to-software communication through APIs, while identifiable AI-agent traffic is still at an earlier stage

The next phase will bring more of that software into commerce, initially through tightly controlled purchases and eventually through broader financial responsibility.

Crypto provides much of the infrastructure this customer requires:

• Smart accounts establish ownership and permissions,

• stablecoins provide programmable working capital,

• cards connect agents to ordinary merchants,

• credit expands their operating capacity, and;

• memory preserves the rules behind every financial decision.

Neobank customers are no longer just human… The financial institutions that recognise software as a customer early could gain access to a market defined by global reach, continuous operation and transaction frequencies that human banking was never designed to support.

Pay attention to those protocols building for non human customers.

That’s a wrap for this issue of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. Valuations are approximate snapshots and move quickly. Verify all market data independently before acting.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of three key parts:

1. Snippet Partner: CRUNCH DAO

2. Theme of the Week: Machines Will Become Our Masters

3. Landscape Analysis: X402/Cloudflare, Telegraph, CRUNCH, Openserv, BANKR, Virtuals, peaq, XMAQUINA, AUKI, & more

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

Quant finance has a very obvious AI problem: most LLMs cheat time

They are trained on the internet as it exists today, which means they often know future outcomes when asked to reason about the past

That destroys backtests, pollutes signal research and makes them borderline unusable for serious quant workflows where point-in-time integrity matters

CrunchDAO is attacking this with ChronoLLM, a new class of point-in-time financial models designed to only know what would have been available at a specific historical cutoff

Instead of one generic frontier model with contaminated memory, the system trains year-specific checkpoints, so a 2014 model should not magically know what happened in 2018. A great explainer of subnet 38 on this pod:

This is where the Bittensor angle becomes interesting; it can supply decentralized compute for training and model checkpoints, while Crunch brings the quant expertise to stress test whether those models are actually clean

The key is adversarial leak detection: actively probing models to see whether they are accidentally carrying future information

The end product is financial AI infrastructure that can be trusted inside signal research, hedge fund workflows, backtesting, strategy design and machine-driven capital allocation

CrunchDAO is effectively building for the moment where financial agents need decision-grade intelligence

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

Machines will Become our Masters.

Not in the full Skynet sense. I do not think we are suddenly entering a world where Cyberdyne accidentally births a military superintelligence, Sarah Connor is vindicated overnight, and humanity ends up hiding in bunkers while robots patrol the surface.

That’s the cinematic version of the machine economy, but it’s not the investable one

The direction is still right though; machines are going to drive a much larger share of economic activity. The difference is that this likely starts with agents acting on behalf of humans, businesses, protocols and robots rather than fully autonomous systems operating without constraint

For now, the machines are not our overlords. They are delegated economic actors.

The internet was originally built for humans clicking links, opening accounts, filling in forms, choosing subscriptions, entering card details and sitting inside dashboards

Almost every internet business model still assumes the customer is a person with attention, patience and a screen

Agents do not behave like that; an agent does not care about your landing page, brand campaign or monthly SaaS bundle

It wants to know whether a resource exists, whether it is allowed to use it, what it costs, how to pay, whether the output can be trusted, & whether it can move on to the next task. This is why Cloudflare’s latest x402 announcement is important:

Cloudflare announced its Monetization Gateway, which will let customers charge for web pages, datasets, APIs and MCP tools protected by Cloudflare.

The charges settle in stablecoins over x402, with Cloudflare handling payment verification and enforcement at the edge

That’s a sign that the commercial surface area of the web is changing from human attention to machine requests. Cloudflare is effectively saying any API, data source, MCP tool or web resource can become a payable endpoint without the seller needing to build its own payment stack.

This is the machine economy in its simplest form: the request becomes the transaction:

• A model asks for a dataset

• A trading agent asks for a signal

• A research agent pays for a source

• A coding agent pays for a test environment

• A procurement agent pays for supplier verification

• A robot pays for mapping, routing, repair diagnostics or remote assistance

Each one is a small machine-to-machine commercial interaction.

The next customer is software, and it does not care about your dashboard, brand or pricing page:

It wants the exact thing needed to complete the task: data, signals, sources, test environments, supplier checks, routing, diagnostics or remote assistance, paid for, verified and settled in one machine-to-machine request.

The mistake would be to think this is only about payments; payments are the rail, but the bigger market is everything around the payment: identity, trust, discovery, routing, verification, memory, reputation, physical-world data and settlement

In other words, the machine economy needs a whole stack, not one magic protocol

Crypto has spent years trying to create internet-native markets. Most were too speculative, too circular or too early.

Agents may finally give these markets a natural buyer because agents can transact constantly, pay per use, route between services and consume machine-readable outputs without needing human packaging.

The ‘Terminator’ was directionally right in one sense: machines are going to do more of the activity

But the optimistic version is not humans versus machines; it’s humans delegating more economic work to machines, while open protocols compete to provide the rails, markets and safeguards those machines need.

What Machines Buy

The best way to understand this market is to stop asking what humans want and start asking what autonomous systems need to complete a task.

Most of that is currently sold through human wrappers… You sign up, get an API key, choose a plan, add a card, talk to sales, receive an invoice, then eventually let your software use the service. That works when the customer is a developer but not when the customer is the agent itself

The agent-native version is different; a service should be able to describe what it sells, what it costs, what permission is required, what proof is returned, what SLA exists, and how reputation is tracked.

The agent pays per request, receives the output, logs the receipt & continues the workflow:

That is why x402 (& Stripe’s MPP) is important. It revives the HTTP 402 “Payment Required” status code and makes the payment part of the normal web request flow.

The client requests a resource, the server responds with price and payment instructions, the client pays, and the request is repeated with proof of payment attached.

The first obvious markets are APIs, MCP tools, web data, inference, search, scraping, compliance checks, financial data, model calls, storage and execution. These are already consumed by software today, but the monetisation model is still mostly designed for humans…

The next markets are more interesting because they are not just raw API calls. Agents will buy judgement: verified intelligence, calibrated forecasts, anomaly alerts, reputation scores, insurance quotes, route decisions, robotics task data, physical-world maps, identity attestations, agent labour and dispute resolution.

This is where the machine economy starts to look less like SaaS and more like a web of paid micro-markets. Machines do not need dashboards. They need priced capabilities.

Protocols Serving the Machine Economy*

*this is just a selection, and not to be considered an exhaustive list

i) x402 and Cloudflare: payment at the edge

x402 is quickly becoming the clearest crypto-native payment primitive for agents. The reason is simple. It sits inside the request flow itself. The server can say payment required, the agent can pay, and the resource can be unlocked without a checkout page, card form or account workflow.

Cloudflare’s Monetization Gateway turns that from protocol theory into distribution:

Cloudflare already sits in front of a huge amount of internet infrastructure, so making paid access configurable at the edge is a serious step. A seller could charge for a specific API route, a dataset, a crawler-access path, or an MCP tool call without rebuilding its entire commerce stack.

This is the first layer of the machine economy: machines need to pay.

ii) Telegraph: verified intelligence for agents

Telegraph sits one layer above payments. If x402 lets agents pay, Telegraph asks what they will pay for: verified machine intelligence.

Its aim is to turn raw model outputs into tradable signals with provenance, price, confidence, validation and receipt.

That matters because agents acting with money need more than a chatbot answer. A machine routing capital, approving a payment or flagging risk needs to know who produced the signal, how it was validated and whether it can be trusted enough to act.

This is the more realistic Skynet path: not machines making unconstrained decisions alone, but humans delegating bounded authority to agents that need verified inputs before acting.

iii) CrunchDAO: models as paid suppliers

CrunchDAO turns model performance into a competitive supply chain. It connects companies with AI researchers and ML engineers who compete to solve measurable predictive problems, with the winning models deployed into real workflows.

Agents need forecasts, classifications, anomaly detection, market regime signals, low-latency pricing and calibrated probabilities

Crunch’s product suite, from ML competitions to private coordinator nodes and secure orchestration, points to a world where intelligence is sourced from markets, evaluated against targets and sold into production systems.

If agents are going to buy intelligence, prediction markets and model competitions become key suppliers to the machine economy. CRUNCH has thousands of ML engineers building intelligence for machines:

iv) OpenServ: enterprise reasoning for agentic workloads

OpenServ fits the thesis because it is building around reasoning, validation, privacy, audit and security for agentic workloads rather than just another agent frontend.

That matters because banks, governments, robotics companies and regulated institutions will not run high-stakes workflows on loose chatbots. They need bounded reasoning, deterministic processes, audit trails and stronger privacy controls.

The SERV model is also worth watching because it points token value back toward real usage, with platform revenue flowing into buyback and burn. The real OpenServ angle is simple: enterprise reasoning infrastructure becomes something agents use, enterprises buy and tokenholders can track through actual platform demand

Q3 looks like there’s a lot in the pipeline:

v) BANKR: agentic businesses, not just agent demos

BANKR is interesting because it is trying to turn agents into self-sustaining businesses, not just chat interfaces. Its infrastructure handles wallets, trading and security so builders can launch agents that can fund themselves, access rails, pay for tools and operate economically.

The ecosystem is the important part.

• Surplus routes inference to the cheapest provider

• GitLawb builds decentralized git for agents and humans

• 1claw gives agents hardware-enclave protected keys

• Helixa builds agent identity

• Miroshark runs multi-agent simulations

• Blocktronics handles onchain forensics

• Kupo is building an EVM trading terminal with intelligence layered in

GitLawb is probably the cleanest example of what agents actually need. Coding agents will need repositories, version control, identity, permissions and PR workflows.

Helixa points at the other bottleneck: machines cannot buy, sell, collaborate or hold keys at scale unless other machines can understand who they are and whether they should be trusted.

BANKR looks less like one agent project and more like a lab for agent-native businesses.

vi) Virtuals: tokenized agents and agent-to-agent commerce

Virtuals takes a more economic view of agents. Its thesis is that agents can generate services, earn revenue, coordinate tasks and transact with humans and other agents.

The important idea is agentic GDP; if agents become productive entities, tokenized ownership can align creators, contributors and users around their output. The open question is whether most tokenized agents become durable businesses or remain speculative mascots, but the direction is clearly relevant to the machine economy.

More recently Virtuals has ventured closer to the physical AI realm, with operations like Seasaw on subnet 5 of Bitrobot + a push with Eastworlds:

vii) peaq: machines as economic actors

peaq is one of the most direct crypto bets on machines becoming economic actors. Its stack is built around robots, vehicles, sensors and devices that can authenticate, earn, pay, prove work and become investable assets.

This moves the thesis from digital agents to physical machines. A robot does not buy seats. It needs identity, payments, ownership, data monetisation and coordination with other machines.

viii) XMAQUINA: robotics and physical AI exposure

XMAQUINA gives exposure to robotics and physical AI through a DAO structure allocating across humanoids, robotics infrastructure and related verticals:

The important part is that robots are the physical extension of agents:

• A digital agent can search, pay and execute online

• A robot can act in the real world

Once those converge, machine activity starts touching labour, logistics, manufacturing, healthcare and homes.

The optimistic version is not Terminator. It is machines taking on repetitive, dangerous or low-leverage work while humans move further up the stack into ownership, orchestration and governance.

ix) Auki: making the physical world readable by machines

Auki is building the real-world web, making physical spaces browsable, navigable and searchable for machines:

That matters because robots cannot operate well if every machine has to independently map and understand every environment from scratch. If agents are the digital workforce, robots are the embodied workforce, and they need shared spatial context to move through the world.

x) Nevermined, Skyfire and ERC-8004: identity, permission and trust

Agent payments are useless without identity and permissioning. Merchants need to know who an agent represents, what it can spend, what it has done before and whether it should be trusted.

• Nevermined focuses on payments and delegated spending for AI agents

• Skyfire gives agents verified identity and payment credentials

• ERC-8004 introduces registries for identity, reputation and validation.

This layer is early, messy and easy to game, but that is exactly why it matters

Once agents spend real money, the market needs to know which agents are safe, which services are reliable and which actions are within delegated authority.

John Connor does not need to fight the machines if the machines are bounded, audited and accountable

The Machine Economy Stack

The machine economy is a stack of capabilities that allows software and robots to act economically:

1. At the bottom are payments and wallets. Agents need to hold funds, receive delegated spending authority, make small payments and settle instantly without a human approving every request.

2. Above that is identity and permissioning. Agents need to prove who they represent, what they are allowed to do, how much they can spend and whether they have a trustworthy history.

3. Then comes discovery and routing. Agents need to find services, compare prices, evaluate latency, understand quality and select the right provider for the task.

4. The next layer is access to data, tools and compute. This includes APIs, MCP servers, search, crawling, inference, storage, software testing, deployment and model execution.

5. Above that sits intelligence. Agents will buy predictions, classifications, risk scores, market forecasts, verification, anomaly detection and decision-grade signals.

6. The final layer is physical-world execution. Robots, drones, vehicles, sensors and industrial machines need maps, spatial context, financing, machine identity, task data and maintenance coordination.

This is why the machine economy is bigger than x402; it makes the payment possible, but the real market is everything that machines need

Final Thought

The Terminator was directionally right that machines would become central to the future economy, but the path is not one Skynet controlling everything, it is millions of bounded agents and robots acting on behalf of humans, businesses and protocols

Or at least, for now…

The protocols that matter are the ones building the rails, trust layers and intelligence markets for those machines to spend, verify, decide and act safely at scale

Worth challenging yourself on companies you currently back and reassess whether they’re building for the machine economy… if they’re not, they may find themselves outpaced by those that do cater for trillions of machines!

That’s a wrap for issue 183 of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. Valuations are approximate snapshots and move quickly. Verify all market data independently before acting.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of three key parts:

1. Snippet Partner: Naoris Protocol

2. Theme of the Week: AI Memory Trade

3. Landscape Analysis: Micron, SKHynix, Samsung, Broadcom, Astera, Rambus, Walrus, Filecoin, Arweave, Hippius

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

The quantum mandate skipped the networks holding your money

Two new US executive orders just put quantum computing on a clock:

• One targets a cryptographically-relevant quantum computer by 2028

• The other moves federal critical infrastructure onto post-quantum cryptography by 2030 and 2031, a timeline pulled forward four years

NSA, CISA, DHS and Commerce are all named, with “harvest now, decrypt later” cited as the threat. But the orders cover government systems and stop there. That leaves the decentralized networks holding most onchain assets with no deadline and no protection.

This is the gap Naoris Protocol was built to close. It is a native post-quantum Layer 1, secured by NIST-standard cryptography from genesis rather than retrofitted later, with mainnet live since April.

The government is still racing to mandate quantum-resistant infrastructure. Naoris already runs it for the chains the orders leave out.

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

All Memory Is Equal, Except Some Memory Is More Equal Than Others

George Orwell spent two books warning that whoever controls the record controls everything that follows, including memories.

In Animal Farm, the commandments change on the barn wall while the animals are told nothing has changed:

In 1984, the past is edited until the present becomes impossible to challenge:

He was writing about politics, language and power… but the warning now applies to AI infrastructure, and in particular, memory.

The next phase of AI will be shaped by memory: who owns it, who can access it, who can edit it, who can revoke it, and who gets paid when it becomes scarce.

• The market has already repriced the physical side of this trade through HBM, DRAM, NAND, interconnects and storage.

• Crypto is still pricing the agent-memory side as if it were generic storage or early middleware.

The first phase of AI was compute; the second phase was power; the third phase is memory, and memory is where the trade becomes more interesting.

If agents become a major interface for the internet, memory becomes their past, their permission system and their operating state; particularly as trillions of agents come “alive”, which we are seeing through protocols like Hermes and OpenClaw:

That makes the memory layer one of the most important markets in AI.

“All memory is equal, except some memory is more equal than others”

1. The commandment on the barn wall

The AI market still talks too much about compute. Compute matters, but AI systems also need memory capacity, memory bandwidth and memory movement.

• A GPU with constrained memory cannot run large workloads efficiently.

• A model without persistent state forgets the user.

• An agent without memory becomes a repeated prompt loop.

Memory now matters in three layers:

1. The first is hot memory: HBM and GPU-attached memory. It is fast, scarce and expensive. It determines how much context can sit close to compute and how efficiently AI infrastructure can run.

2. The second is serving memory: DRAM, CXL, NVMe, cache management, storage tiering and KV cache offload. This layer decides whether inference can scale economically or whether GPUs spend too much time waiting for state to move through the system.

3. The third is persistent agent memory: user preferences, task history, permissions, workflow state, tool outputs and long-term context. This is where crypto has a possible wedge, because persistent memory needs ownership, permissions, auditability and portability.

The Orwellian tension provides clarity: The same memory that makes agents useful, also makes them risky.

• A useful agent remembers your preferences, research, accounts, workflows and decisions.

• A dangerous memory layer stores those details inside a closed platform where the user cannot see the full record, cannot move it easily and cannot verify how it changes.

The AI version of “Big Brother is watching you” is less obvious than a camera surveillance network.

It’s an agent that remembers everything while the user cannot see where that memory lives, who can access it, or what has been rewritten over time.

2. Some memory is more equal than others

The public market has already identified the first side of the trade.

HBM, DRAM, NAND, controllers, interconnects and storage systems have moved to the centre of the AI capex cycle. Memory is no longer viewed only as a cyclical hardware category. In AI, memory is a strategic constraint.

The rationale is simple: AI consumes memory bandwidth, memory capacity and memory movement. Longer context windows need more memory; more inference users need more memory; more agents need more memory; more tool calls need more state; more concurrency multiplies the whole problem:

The GPU trade was always a system trade; the system includes compute, memory, networking and power. If memory is constrained, the system is constrained.

This is why NVIDIA’s H200 matters as a memory product as much as a compute product. The most valuable real estate in AI infrastructure is measured in memory capacity and bandwidth because that is what determines how much work can be kept close to compute.

The commandment on the barn wall would read:

“Compute attached to scarce memory is more valuable than compute alone”

3. The Inner Party owns the fast tier

The physical memory trade is already a large public-market theme.

• Micron, SK hynix and Samsung are direct expressions of the AI memory buildout.

• Astera Labs, Rambus, Western Digital, Seagate and SanDisk sit in the broader memory, interface and storage stack.

The best way to read this part of the market is through scarcity.

• HBM is difficult to manufacture, closely tied to advanced packaging and reserved by the largest AI customers.

• DRAM and NAND benefit from the same demand wave, although with different degrees of pricing power.

• Interconnect and memory-interface companies benefit from the rising complexity of moving state through the system.

This is the Inner Party of the memory stack: the companies closest to the fastest and most constrained memory tiers:

i) Micron

Ticker: MU
Approx valuation: $1.3T
Layer: HBM, DRAM, NAND
Memory role: US pure-play memory torque

Micron is one of the cleanest public-market memory trades because it gives direct exposure to HBM, DRAM and NAND.

The company used to be treated as a cyclical DRAM business.

“DRAM is probably going to be 30 to 40% of all hyperscaler CapEx next year, every hundreds of billions of dollars spent goes straight to DRAM" - (Gavin Baker, CIO of Atreides Management)

Although the AI cycle has changed that perception because memory supply is now a strategic input for data-centre customers.

Customer behaviour changes when memory becomes scarce:

• In a normal memory cycle, buyers wait for price weakness.

• In an AI infrastructure cycle, buyers reserve supply because short memory means short inference capacity.

Micron’s recent multi-year supply commitments are evidence that customers want guaranteed access rather than spot exposure to a volatile market.

• The bull case is continued HBM growth, tight DRAM and NAND supply, and stronger pricing for longer than the market expects.

• The bear case is the old memory cycle returning. If capacity catches up too quickly, the market will move from “memory is strategic” back toward “memory is oversupplied”.

ii) SK hynix

Ticker: 000660.KS
Approx valuation: $1.2T+
Layer: HBM, DRAM
Memory role: HBM quality leader

SK hynix is the clearest HBM leadership trade.

HBM is difficult to manufacture, supply-constrained and closely tied to advanced AI accelerators. Leadership in this category is worth more than generic memory capacity because AI customers need reliable supply at scale.

The company has become one of the key suppliers for the AI accelerator ecosystem. That gives it strategic value in the current cycle and explains why the market treats it as more than a normal memory stock.

• The bull case is continued HBM leadership, premium pricing and long-duration supply agreements, oh and a Nasdaq listing?

• The bear case is competition. Samsung and Micron are investing heavily, and in memory markets high margins always attract new capacity

iii) Samsung Electronics

Ticker: 005930.KS
Approx valuation: $1T+
Layer: HBM, DRAM, NAND, foundry, devices
Memory role: Scale catch-up

Samsung is less pure than Micron or SK hynix, but its scale matters. It has depth across HBM, DRAM, NAND, foundry, devices and consumer electronics. That makes the stock more complex, but it also gives Samsung the ability to compete across the whole memory stack.

The key question is HBM execution. If Samsung closes the gap with SK hynix and Micron, the market can revalue its AI memory exposure. If execution lags, the stock remains diluted by non-memory complexity and by the loss of share in the highest-value part of the cycle.

• The bull case is a successful HBM catch-up combined with strong DRAM and NAND demand.

• The bear case is timing. In AI memory, customers commit to suppliers that can deliver now, and delays can cost years of share.

iv) Broadcom

Ticker: AVGO
Approx valuation: $1.8T
Layer: Custom AI silicon, networking, ASICs
Memory role: Custom inference system partner

Broadcom belongs in the memory trade because AI infrastructure depends on data movement. Hyperscalers want lower cost per token, better networking, custom accelerators and more efficient inference systems. Memory layout, interconnect and accelerator design are central to that equation.

Broadcom benefits if custom AI silicon becomes a parallel market to NVIDIA for inference at scale. Its role is less about owning HBM supply and more about helping the largest customers build systems where memory, compute and networking are balanced for specific workloads.

• The bull case is continued growth in custom AI accelerators and networking.

• The bear case is customer concentration, because a small number of large buyers can create huge revenue while also pressuring margins

v) Astera Labs

Ticker: ALAB
Approx valuation: $70B to $80B
Layer: AI connectivity, CXL, memory expansion
Memory role: Memory pooling toll booth

Astera Labs is a memory movement trade. AI servers need better ways to expand, pool and access memory. CXL is one of the key technologies for this shift because it can help systems access memory more flexibly and reduce pressure on the most expensive tiers.

This matters because inference creates a lot of state. Some state belongs in HBM. Some should move to cheaper tiers. Some should be reused, cached or offloaded. The more agents run in parallel, the more important this tiering becomes.

• The bull case is broad CXL adoption in AI servers.

• The bear case is valuation; the market already gives Astera significant credit for future AI infrastructure growth

vi) Rambus

Ticker: RMBS
Approx valuation: $13B
Layer: Memory interface chips and IP
Memory role: Specialist interface layer

Rambus is a smaller but relevant memory exposure. As memory systems become faster and more complex, the interfaces around them become more important. Controllers, buffers, security IP and memory-interface design help the system function.

This is a less obvious way to express the trade, but it is still part of the same direction. AI infrastructure is creating more complexity around server memory, DDR5, HBM-adjacent systems and data-centre architecture.

• The bull case is that memory complexity keeps rising and Rambus captures more value from interface IP and chips.

• The bear case is that it remains a specialist supplier rather than a broad AI infrastructure winner

vii) SanDisk, Western Digital and Seagate

Tickers: SNDK, WDC, STX
Approx valuations: SanDisk $300B+, Western Digital $200B+, Seagate $200B+
Layer: NAND, SSDs, HDDs, storage systems
Memory role: Cold-memory beta

AI does not keep all memory beside the GPU. A large amount of AI data sits in colder storage layers: datasets, logs, embeddings, model checkpoints, videos, synthetic data, agent records and archives.

SanDisk, Western Digital and Seagate provide exposure to this cold-memory layer. This part of the trade is less scarce than HBM, but AI data growth still creates a major demand tailwind. The value is volume rather than premium scarcity.

• The bull case is that AI turns storage into a structural growth market.

• The bear case is commoditisation.

HBM has stronger pricing power because it is closer to the bottleneck. Storage gets the demand growth, but margins depend on whether the market avoids oversupply.

4. The Ministry of Truth runs inference

The memory bottleneck is not only about capacity and speed. It is also about whether the record can be trusted.

In Nineteen Eighty-Four, Winston Smith works at the Ministry of Truth rewriting records so the past always agrees with the present.

That is a useful frame for AI memory because a badly designed memory layer can produce a similar effect without a political villain. It can store the wrong thing, retrieve it as truth and carry that error forward across future decisions.

This is the harder product problem; a larger context window does not automatically create better memory. It can simply give the system more room to preserve confusion.

Persistent memory is valuable only if it can distinguish preference from fact, stale information from current information, permission from habit and correction from contradiction.

The risk is especially serious in agentic systems. An agent acts, pays, trades, files, books, moves data and writes to other systems. If its memory becomes corrupted, the error becomes operational. The agent may repeat a mistaken assumption, preserve outdated permissions, trust stale context or carry the wrong state into a real workflow.

Orwell had a word for holding contradictions in the mind and accepting both. He called it “doublethink”. In AI, the software version is a memory system that stores an error, retrieves it confidently and uses it to justify the next action becomes a huge risk with a plethora of hallucinations.

This is why the memory trade expands beyond hardware. The winners will not only store more state. They will help decide which state is true, which state is stale, who wrote it, who can change it and when it should be forgotten.

“A memory you cannot correct is a memory that controls you”

We should also pay attention to those who control the memory; they’ve been fined for price fixing previously:

5. Four legs good, two legs better

The most useful crypto angle is not faster memory. Crypto will not compete with HBM on latency. The crypto opportunity is persistent, permissioned, portable and verifiable memory for agents.

In Animal Farm, the original slogan:

“four legs good, two legs bad”

…gradually becomes:

“four legs good, two legs better.”

The animals do not experience one dramatic lie, but they experience a slow rewrite of the record by whoever controls the wall.

Agent memory has the same risk.

Today, most memory lives inside whichever platform created the agent. Your preferences, work context, wallet permissions, workflows and past interactions sit in a company’s store, in a format you may not control, under rules you did not write. If memory is the agent’s operating state, then control over memory becomes control over the relationship.

This is where crypto becomes relevant; the agent needs a record that cannot be silently edited by a platform, a history that proves who wrote each entry and when, and an access layer that lets the user grant, delegate and revoke permissions. Portability, signatures, audit trails, programmable access and user-owned records are crypto-native primitives.

The point is not to put every memory onchain… that would be slow, expensive and unnecessary. The useful design is to put control onchain while the memory itself stays encrypted in storage:

A smart contract or policy layer can define who may read, write, delegate or revoke access, while signatures prove which user or agent authorised the action.

That is the crypto memory wedge: own the rules around memory rather than pretending to replace HBM.

6. The open memory layer

Crypto memory is earlier, smaller and less proven than the public equity memory trade. That is exactly why the category is interesting.

• The physical memory trade has revenue, customer commitments, supply constraints and strategic urgency

• The crypto memory trade has optionality, weak revenue, immature products and much smaller valuations. The opportunity is real only if agents turn memory into a workflow dependency

The strongest crypto projects will solve specific problems, making memory portable across applications… interoperability is supposed to be a core feature after all!

They will let a user or enterprise grant and revoke access, preserving provenance. They will support retrieval and will connect memory to payment and identity creating usage sinks tied to actual agent activity

Here are several protocols holding the memory baton in crypto:

i) Walrus

Token: WAL
Approx valuation: $75M market cap / $150M to $170M FDV
Layer: Decentralized storage and portable agent memory
Memory role: Direct agent-memory bet

Walrus is the cleanest direct crypto memory project because it explicitly targets portable memory for AI agents:

The core idea is straightforward: agents should be able to carry memory across applications and providers while access is controlled by the user, builder or agent rather than by one platform.

This matters because agent memory includes identity, permissioning, recall and long-term state. If an agent remembers trading preferences, research history, work context, customer interactions or wallet permissions, that memory becomes sensitive infrastructure. It needs access control.

Walrus combines decentralized storage with Sui-based access control and encryption. The strongest version of the product is portable memory that agents can use without handing full control to one application.

• The bull case is Walrus becoming a default memory layer for AI agents, especially inside the Sui ecosystem and adjacent agent frameworks.

• The bear case is limited usage. Agent memory needs integrations, fees, retention and developer adoption. Demos will not be enough

ii) Arweave (AO)

Token: AR
Approx valuation: $125M to $130M market cap
Layer: Permanent storage and AO compute
Memory role: Permanent provenance

Arweave is the permanent-memory trade. Its core value is permanence, which is not right for every type of agent memory but is powerful for provenance

AI creates a large need for proof of origin. Users and developers will need to know which dataset trained a model, which agent made a decision, which policy existed at the time, which record changed and which output is canonical.

In “1984” the archive can always be rewritten. Arweave offers the opposite primitive: a record designed to resist deletion and revision.

AO adds a compute layer, which gives Arweave a stronger agent angle.

• The bull case is that AI provenance becomes a major market.

• The bear case is that permanent memory is narrow because many agent memories need to be private, editable or forgettable

iii) Filecoin

Token: FIL
Approx valuation: $550M to $600M market cap / $1B+ FDV
Layer: Decentralized storage
Memory role: Storage incumbent

Filecoin is the incumbent decentralized storage network. For AI, its strongest angle is large-scale storage: datasets, model artifacts, archives, research data, media and backups. It is more relevant to heavy storage than instant agent recall.

• The bull case is that AI data demand grows so large that decentralized storage captures meaningful usage.

• The bear case is that storage capacity alone is not enough. The market will want retrieval performance, revenue, developer demand and AI-native workflows.

Filecoin needs to become more than storage supply. It needs to become useful inside real AI data pipelines

iv) Bittensor and Hippius

Tokens: TAO / SN75
Approx valuation: TAO around $2B market cap; Hippius around $15M
Layer: Decentralized AI network and Bittensor storage subnet
Memory role: Bittensor-native memory and storage

Bittensor is a decentralized intelligence market rather than a memory project, which is the core reason Hippius is interesting.

If Bittensor becomes a serious AI economy, it will need internal infrastructure for inference, routing, evaluation, datasets, storage and memory. If you’re want an understanding of its investment history - read out latest report:

Hippius, Subnet 75, is the clearest memory and storage expression inside that ecosystem. It focuses on decentralized storage, cloud infrastructure and persistent memory services for Bittensor.

The investment case is not that Hippius is a standalone category leader in agent memory. The case is that it is a memory and storage layer inside one of crypto AI’s largest networks.

This is different from Walrus. Walrus is a cleaner agent-memory bet. Bittensor and Hippius are an ecosystem-linked bet: if the subnet economy professionalises, storage and memory services become part of the internal AI supply chain.

• The bull case is that Bittensor demand grows and subnet-level infrastructure becomes valuable.

• The bear case is that subnet valuations remain reflexive, alpha-token methodology is difficult to underwrite, and storage usage inside Bittensor does not become large enough to support the narrative.

7. The memory hierarchy

This is the valuation gap that makes the trade interesting.

• HBM suppliers sell physical scarcity.

• Crypto memory projects sell programmable memory primitives.

They have different products, revenue models and maturity levels, so the comparison is not like-for-like. The shared direction is that AI systems need more memory and that the market has already rewarded the physical layer more aggressively than the agent-memory layer.

The public market has priced physical memory as infrastructure, while crypto still prices agent memory as optionality. That does not make every token cheap, because most of them still need to prove usage, but it does mean the upside is less capitalised if agent memory becomes a real workflow dependency.

8. Who controls the past controls the agent

The trade comes down to one question: does memory become a workflow dependency?

For equities, the answer is already visible. HBM supply is being reserved, AI customers need more capacity, and data centres need more memory bandwidth, more storage and better interconnects. The market is no longer debating whether memory matters; it’s debating how long the shortage lasts and who captures the most value…

For crypto, the question is still open:

• Walrus needs agent integrations

• Arweave needs AI provenance usage

• Filecoin needs AI storage demand and retrieval relevance

• Bittensor and Hippius need storage and memory usage inside the subnet economy

The crypto memory trade only works if agents and users pay for memory directly or indirectly through real workflows… and whether that payment flows back to the token. The age old question of whether the token ACTUALLY accrues value!

AI branding is not enough, storage supply is not enough, and a token attached to a storage narrative is not enough. The winning projects need measurable usage: writes, reads, retrievals, access-control events, paid storage, agent retention and developer adoption.

The most important crypto signal is practical; can a user or enterprise give an agent memory, revoke that memory, move that memory, audit that memory and pay for that memory without depending on one central platform?

If the answer is yes, crypto has a real wedge. If the answer is no, centralized AI platforms become the archive.

9. If there is hope

Orwell’s dystopias were about control. The telescreen watched, the Ministry of Truth rewrote, and the commandments on the barn wall changed while the animals were told the record had always been that way. AI memory creates a similar tension because the most useful agents will be the ones that remember, and the most dangerous memory systems will be the ones that remember without accountability.

The physical memory trade is already large, liquid and well understood. HBM suppliers and memory-linked equities have revenue, supply constraints and strategic customers. Crypto-native memory is smaller, messier and much less proven, but it sits closer to the unresolved question: who owns the memory that makes an agent useful?

Orwell wrote that:

“if there is hope, it lies in the proles.”

In the AI memory trade, the equivalent hope is the part of the stack that has not yet been enclosed. The hardware supercycle is already in the tape and the major platforms already own the user relationship, but the “open memory layer” is still underbuilt and underpriced; that’s the opportunity.

The next decade of AI can inherit the architecture of cloud, where memory stays inside closed platforms, or it can develop an open memory layer where users and agents can carry state across applications with clear permissions and auditability.

One path creates more powerful platforms. The other creates more powerful agents.

Some memory will always be more equal because physics makes fast memory scarce. HBM beside a GPU will always sit higher in the hierarchy than cold storage in an archive. That is a law of infrastructure.

Whether some owners are “more equal than others” is a design choice. If agents become the next users of the internet, memory becomes their past, their permission system, their identity layer and their operating context.

The trade is about who owns that layer.

“All memory is equal, except some memory is more equal than others” - (Sammy, 0x. 2026)

[Inspired by Orwell]

That’s a wrap for issue 182 of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. Valuations are approximate snapshots and move quickly. Verify all market data independently before acting. Quotations from George Orwell’s Animal Farm and Nineteen Eighty-Four are used for commentary and criticism.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of three key parts:

1. Snippet Partner: NuNet

2. Theme of the Week: AI Inference Wars

3. Landscape Analysis: OpenRouter, Nunet, Darkbloom (Eigen Labs), Venice, Dolphin, Openserv, Ionet, Akash, Chutes, Targon, C0mpute (ft: Virtuals)

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

NuNet is highly relevant to this week’s theme. As inference moves beyond centralised cloud providers, orchestration across distributed compute becomes a much bigger part of the market.

That is where NuNet fits in.

Rather than simply being another GPU marketplace, NuNet is focused on coordinating workloads across different machines, devices and infrastructure environments:

We cover why that matters in more detail later in the article

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

“AI Inference Wars”

The last AI cycle was about training, but the next one is increasingly about inference.

Many of you may be hearing the word “inference” for the first time… So what is it?

Training creates the model. Inference is when someone asks that model a question or gives it a task, and the model produces an answer.

1. AI Inference Market Overview

Training gets the headlines because it’s the brain behind the outputs that so many of us have been blown away by.

Currently, inference receives a large portion of the economic benefits because every prompt, agent loop, image, trade, tool call and code edit has to run somewhere.

That economic activity is showing no signs of slowing down; openrouter saw 47 trillion tokens flow through its protocol last week alone:

So what makes up an inference market? Well, it needs a few things:

• Tokens are becoming the unit of account.

• OpenRouter is fast becoming THE exchange layer; 47 trillion tokens were used through the LLM marketplace last week alone

• Fireworks, Together, Replicate, Baseten, Groq and the hyperscalers are becoming the professional supply side

• Chutes, Akash, io.net, Nosana, Targon, Venice, NuNet and other crypto AI networks are trying to build the permissionless version underneath

The mistake is to compare all of these providers as if they are competing in the same market; they are not.

• Traditional providers sell reliability, developer experience and enterprise procurement

• Crypto AI networks sell cheaper supply, open access, privacy, verifiability and new incentive loops

More recently opened many peoples’ eyes to the risks of becoming reliant on single frontier proprietary model for tokens as Anthropic banned use of its mythos model (Fable 5) for anyone outside of the USA:

The interesting part is where the two worlds start to overlap; a focus on privacy, confidential compute or agent-native payments (Venice & Targon feature well here).

3. How to view the AI Compute market

The better way to view the market is:

The interesting part is not where these worlds differ, but where they start to overlap; traditional providers sell reliability, developer experience and enterprise procurement.

Traditional providers generally compete on reliability, support and developer experience, while Crypto AI networks generally compete on open access, lower-cost supply, privacy, verifiability and novel incentive mechanisms to co-ordinate capital seamlessly around the globe

4. Why Inference Is The Real AI Market

The model layer still matters, but model quality is compressing faster than most people expected. Open source models are performing at 90-95% the quality of frontier models, but at 10% of the cost (e.g. GLM-5.2 from Zai):

Open models keep improving. Chinese labs keep pushing down pricing. Frontier models still command a premium, but below that top tier, token pricing is becoming competitive very quickly.

This is why the routing layer matters; If the same open model can be served by five different providers at five different prices, developers do not want to hardcode one endpoint forever so they want a router.

A router can choose by:

The router sits above the providers and turns a messy landscape into one clean interface:

This is what OpenRouter got right, and why Funds splashed $113m in a recent Series B, to plug a clear gap in the market:

OpenRouter is fast becoming the market interface; one key can access hundreds of models across many providers. The real value is not the model list. The real value is that the same request can be routed to the best provider for that job.

This starts to look like an energy market; the user does not care which power plant made the electricity. They care that the lights turn on, the price is fair and the system does not break.

AI users will increasingly think the same way, they won’t care which GPU cluster served the token. They will only care that the response is fast, cheap, private and reliable.

5. Traditional Inference Providers

The traditional side is splitting into four groups.

i) Hyperscalers: AWS, Google and Microsoft

They win on enterprise trust. Large companies are not just buying tokens; they are buying compliance, security, procurement comfort and someone accountable when things break.

ii) Routing markets: OpenRouter and AI gateways

Routers sit above model providers and send each request to the best option. As model leadership changes week to week, hardcoding one model looks increasingly weak. AI needs aggregators, just like crypto does.

iii) Optimised open-model serving: Together, Fireworks, Baseten and Groq

These are not just cheap APIs. They are performance infrastructure companies, focused on speed, batching, scaling, fine-tuning, custom endpoints and production support.

iv) Model marketplaces: Replicate and Hugging Face-style platforms

Inference is broader than chat. Image, video, speech, embeddings, robotics models, simulations and multimodal agents all need models to run. Marketplaces make that long-tail model demand easy to access.

6. Crypto AI Inference Providers

The crypto side gets flattened into one phrase: decentralised compute, which is too vague; there are at least five different categories:

1. serverless inference networks

2. decentralised GPU marketplaces

3. confidential compute networks

4. private AI apps and gateways

5. orchestration layers

They should not be analysed the same way.

i) Chutes: Crypto-Native Inference

Chutes is best understood as a decentralised inference platform, not just a GPU marketplace.

The point is simple: developers do not want to rent GPUs or manage infrastructure. They want an endpoint that works. Chutes focuses on serving open-source models through a familiar API, while using decentralised GPU supply underneath.

The key question is whether it can turn headline usage into paid, recurring demand. Cheap tokens are useful, but only if developers trust the uptime, latency and reliability.

Its revenue per trillion tokens continues to trend up, showing potential for sustained profitability/viability:

ii) Akash: The GPU Auction Layer

Akash is a decentralised cloud marketplace.

Users define the compute they need, providers compete to supply it, and workloads run through leases. This makes Akash more of a compute marketplace than a direct inference router.

Its strongest fit is price-sensitive workloads that can tolerate more infrastructure variability and do not need deep integration with AWS, Azure or Google Cloud.

Fees and token price are somewhat correlated and are on the rise:

iii) io.net: Decentralised GPU Clouds

io.net is closer to decentralised GPU cloud providers.

Its core pitch is access to distributed GPU supply at lower cost and with faster provisioning than traditional cloud routes. That is useful for AI teams that need compute but do not want long cloud contracts or hyperscaler pricing.

The challenge is execution: hardware verification, reliability, scheduling, support and consistent performance. Raw GPU access is valuable, but the higher-margin layer is still routing, managed inference and orchestration

Ionet has been one of the leading performers over the past 30 days with $12.3m annualized revenue:

iv) Targon: Confidential Compute

Targon is focused on confidential compute for AI workloads.

The problem it addresses is obvious: many users will not run sensitive prompts, models or data on infrastructure operated by unknown third parties.

Targon’s answer is protected execution through trusted execution environments, encrypted virtual machines, remote attestation and confidential GPU infrastructure. In plain English, the aim is to prove the workload is running in a secure environment and reduce what operators can see.

This is relevant for private inference, especially in areas like finance, healthcare and enterprise AI. The caveat is that confidential compute is not magic. It shifts trust toward hardware, firmware and attestation systems.

Last year the protocol reported annual revenue of $10.4m and co-authored a research paper with Intel on “Decentralized compute on untrusted hardware”

v) Darkbloom: Private Inference on Idle Macs

Darkbloom takes a different route.

Instead of sharding a large model across random GPUs, it turns idle Apple Silicon Macs into a private inference network. The Mac runs the model locally, while requests are encrypted and routed to verified providers.

The wedge is privacy and cost, not maximum frontier-model performance.

This is useful because “no node holds the full model” does not automatically mean prompts are private. Darkbloom is more explicit about the privacy problem, but still needs to prove supply scale, performance and developer trust.

It’s now up to 300 Machines on the network with 2 billion tokens served across 1 million requests:

vi) Venice: Private Inference For Consumers

Venice occupies a different position in the market from networks such as Akash or io.net. Rather than functioning primarily as a GPU marketplace, it is better viewed as a private AI application and inference gateway.

Throughput of said gateway is now up to 85 billion tokens daily:

Most users want an AI product that respects their privacy and gives them access to capable models without extensive data collection.

Venice packages the infrastructure thesis into a consumer-facing experience built around private prompts, open-source models, uncensored access, API functionality and tokenised compute through VVV and DIEM.

The DIEM component is particularly interesting because it points toward a broader idea about agent economies; providing access to $1 of compute per day (into perpetuity?). That’s something the market has attached a decent price tag to recently:

If agents require continuous access to inference, then compute credits begin to resemble an agent-native asset that entire secondary markets can be built around:

An agent that can directly hold and spend compute rights is arguably more practical than one that depends on a human periodically funding a credit card.

This highlights a deeper crypto AI thesis.

Agents will ultimately need access to money, identity, memory and compute, and crypto systems provide a framework for making those resources programmable.

Venice is therefore NOT competing directly with OpenRouter on model breadth. Instead, it is competing on privacy, access and tokenised compute.

That is a legitimate niche, although the key question remains whether demand for private AI products becomes large enough to sustain the token model beyond the current narrative cycle; my bet is privacy is only going to become a stronger narrative as AI proliferates.

vii) NuNet: Orchestration Across Distributed Compute

NuNet is often grouped with decentralised compute projects, but the more useful framing is “orchestration”.

Orchestration involves matching workloads to the most appropriate compute resources and coordinating execution across different machines, environments and locations.

This becomes increasingly important as AI expands beyond centralised cloud infrastructure.

Future AI systems are likely to operate across cloud GPUs, edge devices, local servers, robots, phones, sensors and decentralised provider networks.

As that shift occurs, not every decision can depend on a distant cloud endpoint:

• A warehouse robot may not be able to wait for a cross-region API response.

• A drone cannot assume perfect connectivity at all times.

• A field robot operating in remote conditions may need to perform inference locally whenever network access becomes unreliable.

For that reason, orchestration is emerging as a meaningful category in its own right.

The challenge for NuNet is whether it can transform this coordination problem into a functioning economic network with sufficient supply, demand and developer adoption.

viii) OpenServ: Agent Orchestration, Not Pure Inference

OpenServ is not best understood as a decentralised inference network; it’s better framed as an agent infrastructure and orchestration platform.

That matters because agents are one of the clearest future sources of inference demand. A normal chatbot might call a model once. An agent calls models repeatedly. It reasons, uses tools, checks outputs, calls another model, takes an action and then loops again.

That creates heavy, recurring inference demand that is getting noticed within our little crypto bubble:

OpenServ is therefore relevant to the inference market from the demand side rather than the supply side. If the platform can become a useful place for developers to build, deploy and coordinate agents, then it naturally becomes a layer that can route inference to different providers underneath.

The key question is whether OpenServ becomes a real agent execution layer or simply another agent marketplace with a token attached.

I’d hazard a guess that the capabilities behind this team extend beyond the latter, with several notable benhcmark performances for its reasoning framework and its own proprietary models on its roadmap:

If OpenServ can own the agentic operational workflow, inference becomes an input into the platform rather than the main product itself.

In an agentic world, the most valuable layer will become the place where agents spend a lot of their ongoing time and resources.

ix) Dolphin AI: Product-Led Decentralised Inference

Dolphin AI is interesting for a different reason; it started from model demand rather than a GPU marketplace.

The Dolphin model family already had a reputation for uncensored open models, and that gives the network a clearer reason to exist:

This is important because many decentralised inference projects begin with supply first: “we have GPUs, now who wants to buy them?”

Dolphin is closer to the opposite; It begins with a set of models people already want to use, then builds a decentralised inference network around that demand.

Its architecture is often described as peer-to-pool, which means:

• GPU owners contribute capacity into model-specific pools, rather than each buyer renting a specific node directly.

• Requests are routed into the pool, and;

• available nodes process them.

That is a better design for unreliable consumer supply.

If someone is contributing an idle gaming GPU, they may not keep it online forever. A pool-based model can absorb that more naturally than a one-to-one rental marketplace.

The more interesting part is verification.

Dolphin has been pushing live-weight proofs. In simple terms, this means checking that the model weights actually loaded in the serving process match the model the node claims to be running.

That matters because cheating is one of the hardest problems in decentralised inference.

A node could claim to run an expensive model while secretly serving a smaller, cheaper or more quantised version.

If the network cannot detect that, the whole market loses credibility.

x) c0mpute: Distributed Inference For Agents

c0mpute is worth watching because it is trying to solve one of the hardest problems in decentralised inference: running large models across scattered GPUs on the open internet. This video explainer of the problem it’s solving is fantastic:

Its Shard engine splits a model across multiple machines, rather than requiring one huge server to hold the full model. That makes it relevant for frontier-scale open models that may be too large or too restricted for normal hosting routes.

The Virtuals link is the key demand-side angle. Virtuals is building an agent economy, and agents are heavy inference users. They plan, call tools, transact, check results and loop. That creates recurring demand for cheap, open and censorship-resistant inference:

The caveat is that this is still early. c0mpute needs to prove performance under real load, node reliability, verification and prompt privacy.

But the direction is important: GPU marketplaces sell access to compute; c0mpute is trying to distribute the model itself

7. Traditional vs Crypto Inference

8. What To Watch

i) Paid token volume

The market should place less emphasis on raw token-processing statistics unless those tokens generate revenue. Free-tier activity and subsidised usage can create impressive numbers without demonstrating genuine product-market fit.

Paid inference demand is the metric that matters; it’s more sustainable, enabling longer term viability

ii) Revenue per GPU

A decentralised compute network is only sustainable if GPUs earn more value inside the network than they could elsewhere. If emissions are the primary reason providers participate, supply may disappear once incentives decline. GPU providers will run the numbers for opportunity cost:

iii) Router integrations: Distribution

Distribution is often more important than infrastructure itself.

OpenRouter integrations, coding agents, wallets, payment endpoints, developer tools and consumer applications are all potential sources of demand:

A payment endpoint is a route through which software can pay for a service directly, usually through an API.

iv) Verification

GPU spoofing, fake capacity and unreliable providers remain genuine risks.

Networks need robust hardware verification, encrypted traffic, reputation systems and meaningful penalties for poor behaviour.

v) Privacy guarantees

Private inference remains one of the strongest crypto AI opportunities, but the guarantees must be genuine. Marketing privacy is easy; secure execution, local-first architectures, data minimisation and auditable infrastructure are much harder to deliver.

vi) Token value capture

The strongest token models will connect demand directly to real inference usage. That could involve buybacks, burns, staking requirements, compute rights or revenue-linked mechanisms.

Broad AI narratives alone are unlikely to be sufficient over the long term.

9. Final Take

The inference market is where AI begins to resemble a financial system. Every generated token carries a cost, every endpoint operates with a margin, every agent loop creates recurring demand, every router functions as a market maker and every GPU network becomes a source of supply…

Traditional providers currently dominate the developer experience and enterprise trust layers. Crypto AI networks, however, are exploring a different frontier built around permissionless supply, private inference, verifiable compute, tokenised access and agent-native (unrestricted “KYC-less”) payments

In the near term, the winner is unlikely to be the most decentralised network. It is more likely to be the network that makes decentralised inference feel ordinary and dependable through fast endpoints, strong documentation, reliable uptime, transparent pricing, verified supply and genuine paid demand.

There are however greater risks with crypto protocols as attacks become more common and sophisticated with AI capabilities:

Chutes remains one of the most important projects to watch because it is among the closest to turning Bittensor-backed compute into a functioning inference market rather than simply another GPU narrative. Similarly with Eigen Labs “Darkbloom”

Akash and io.net represent the supply-side challengers, Targon represents the confidentiality thesis, Venice represents the private AI demand layer and NuNet represents orchestration for a more distributed compute future.

The broader thesis is straightforward”:

“AI models may become increasingly commoditised, but inference markets are unlikely to follow the same path.”

The greatest value will accrue to the entities that route the work, verify the work, settle the work and capture the demand.

That is where the next crypto AI opportunity is likely to emerge.

That’s a wrap for issue 181 of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. For a full list of disclosures, please refer to the Khala Research website.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of two key parts:

1. Snippet Partner: GEODNET (RTK positioning for robotics)

2. Theme of the Week: The Customer is No Longer Human

3. Landscape Analysis: DeFi Llama, MCPs, Hermes, Openserv, ReadyAI, Agent Socials / blogging

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

This week’s issue is brought to you by GEODNET, the decentralised network supplying the positioning infrastructure underneath the robotics boom

Every autonomous machine, a delivery drone, a field robot, a self-driving tractor, needs to know where it is to within a centimetre:

Standard GPS gets you to a few metres, which is useless when a robot is navigating a row of crops or a warehouse aisle. The fix is RTK (Real-Time Kinematic) correction, and it has historically meant expensive proprietary base stations with patchy coverage

GEODNET rebuilt that layer as a DePIN, which has led to the world’s largest decentralised RTK network, correcting standard GPS down to 1-2 centimetres

• Scale: over 21,000 active satellite miner stations across 145+ countries, the largest RTK network globally

• Each station is a physical GNSS receiver an operator mounts with a clear view of the sky, earning GEOD tokens for the correction data it contributes

• The economics screen well for a DePIN with annualised revenue in the millions, 80% of data-sales revenue burned periodically

Why it matters for this audience: as humanoids and autonomous machines scale, precise positioning becomes non-negotiable infrastructure. GEODNET is the picks-and-shovels play on that demand, and it monetises real-world subscriptions today

Kyle Samani (formerly Multicoin Capital) provided a good overview of GEODNET’s value proposition at the All In Summit:

Explore the network and its hardware at geodnet.com

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

On June 12, Blockworks acquired Messari; at first glance you may just be thinking its a consolidation of two crypto OGs, taking the best of both and dominate (by what feels like) a shrinking market:

But it undersells what’s actually going on… there’s a line in the announcement that matters more than the deal itself:

“an agent is only as good as the data it can reach and the API it can call”

Read that again; it’s the whole thesis of this newsletter.

The customer for data is shifting from people to agents, and the smart money is now building for a world where the things querying your data outnumber every human analyst on earth many times over.

So this week I want to walk through that shift:

• What changes when you build for agents instead of people

• why the Messari deal is the first big crypto move in that direction

• who else is going after the same prize, and;

• how any of this actually makes money.

2. The Customer is No Longer Human

For the entire history of financial data, the customer was a person:

1. A Bloomberg terminal is priced for a human at a desk

2. Messari research, Delphi reports, a Galaxy note: produced by analysts, consumed by analysts

Agents break this status quo; crypto is already structured and real-time, so production scales with data and models

This is the crypto-native version of the org-flattening point Nadella (CEO, Microsoft) and Dorsey (Block) have both made earlier this year:

• Nadella at Davos called AI a complete inversion of how information moves through a company:

• Dorsey’s March 31 essay with Roelof Botha argued hierarchy exists only to route information:

… which is the logic behind Block cutting roughly 40% of its workforce:

But the bigger change is on the buy side; an agent managing a position, screening a token or running diligence makes a call, pays for that call, and moves on:

The market stops being “every analyst who will pay for a terminal” and becomes every agent that needs data to act, a number that runs to billions and eventually trillions of calls. The businesses that reposition for this new era of agents will thrive

The economics now work because of one piece of plumbing:

• Card rails start around $0.30 a transaction, so per-call data pricing was impossible

• x402 (and MPP), the protocol that turns HTTP’s dormant 402 status into a USDC payment, closes that gap: an agent hits an endpoint, gets a 402, signs a payment, receives the data, no account or API key

• Settlement onchain means calls priced at fractions of a cent

That is the range to anchor on: a few cents per call, across trillions of calls. It’s why Brian and the Coinbase team want to get the x402 endpoints in every application possible and are working overtime to make it happen:

The value will likely accrue to the COIN stock; read section 7 of our Khala Research x402 report where we’ve identified possible value accrual locations:

With that backdrop, here is who is positioning to be the data layer agents pay into, and what makes each distinct

3. Landscape Analysis

Blockworks + Messari

Thesis: become the single system of record for onchain assets, then sell access

What they do:

• Blockworks built the issuer side, standardised disclosures through the Token Transparency Framework plus a full IR platform for protocols, foundations and RWA issuers.

• Messari built the consumption side, eight years of coverage across 40,000+ assets and one of the most comprehensive APIs in crypto

Combined, issuers publish in, consumers query out, and the platform sits in the middle as the record both depend on

The diagram below maps how the two halves fit together:

Why it matters: this is the most direct bet on the agent thesis, and the most heavily resourced. As trading goes agentic, whoever holds the cleanest, most complete, most programmatically accessible record of onchain assets becomes the layer agents are forced to route through.

Own that, and you own the “toll booth”

DefiLlama MCP & LlamaAI

Thesis: the open, community-maintained data layer wins on coverage and neutrality, and is already agent-native. The quiet incumbent in this race

What they do: DefiLlama is the most widely used DeFi data aggregator, tracking TVL, fees, volumes, yields and stablecoin flows across hundreds of chains and a thousand-plus protocols.

It now ships an official MCP server that lets any agent connect and query that data directly, billed one credit per query out of the same pool as the API key, with a Pro tier at $300/month that adds premium endpoints and LlamaFeed

Its LlamaAI product offers a familiar human readable interface for meat bags:

Why it matters: its edge is open-source coverage and speed to index, new protocols often appear within hours of launch, ahead of commercial providers

Where Blockworks-Messari is a curated, proprietary record, DefiLlama is the neutral utility layer, and it wired itself for agents without an acquisition. For a large share of routine onchain queries, it is already the default call

Other:

i) MCPs (Model Context Protocols)

Model Context Protocol (MCP) is an open standard designed by Anthopic in 2025 to enable AI agents to dynamically access and interact with external data sources:

The protocol was made to be machine ready. Although some debate as to whether simple APIs can be just as effective and arguably more established plumbing. Examples:

i) Base

ii) …or Coin Gecko MCP:

The Wildcards:

i) Hermes (Nous Research) Agentic Harness

Partnered with Stripe MPP to enable Hermes agents to pay per call to buy as many things that are purchasable through Stripe.

ii) ReadyAI (Bittensor, Subnet 33)

Thesis: decentralise the production of agent-ready data and let agents pay for it directly. The crypto-native, incentive-driven version, and one I have flagged before.

What they do: ReadyAI now describes itself plainly as “the marketplace for agentic data”, agents that pay for data, discover services and transact.

Subnet 33 uses Bittensor’s miner-validator architecture to turn unstructured data into AI-ready formats at a fraction of human-annotation cost.

It is indexing the web into machine-readable intelligence, giving each domain an llms.txt “semantic passport” any agent can read instantly, with a stated target of a million domains and a free tier plus per-query access on top.

Why it matters: same destination as Blockworks-Messari, opposite construction; instead of buying a dataset, ReadyAI incentivises a decentralised network to produce and structure it, with token rewards replacing the analyst floor entirelY.

It is the clearest test of whether the agent data layer gets built by a company or by a network

iii) Caesar

Thesis: sell agents the finished research, not the raw data. One level up the stack from everyone above

Caesar is an AI deep-research platform that takes a complex question and returns a rigorous, citation-backed answer, synthesising across on-chain and off-chain sources rather than just handing back a dataset.

It runs two products:

1. Ask Caesar, the human front end, and the Caesar API, which the company explicitly positions as a “research layer for intelligent systems, applications and AI agents.”

2. Agents integrate with a single install command and no account needed, there’s $15 of free API credits to start, and access tiers are gated by the $CAESAR token. It’s crypto-tuned but domain-general, and benchmarks itself publicly on Humanity’s Last Exam

Why it matters: this is the insight layer, rather than another data layer. Where Messari, DefiLlama and ReadyAI compete to be the cleanest source an agent pulls from, Caesar competes to be the thing an agent calls when it wants an answer rather than a feed, the synthesis done, sources attached

As agents handle more of the actual analysis, the question is how much value accrues to whoever holds the data versus whoever does the reasoning on top of it. Caesar is the clearest crypto-native bet on the second

iv) Openserv

Thesis: provide agents with an additional reasoning layer they can rely on

On the topic of reasoning, SERV also positions itself in a similar sect to Caesar in that it leverages its SERV reasoning framework to enhance the output of AI models for its users; both human and agentic

v) Agentic Social Platforms like Moltbook + 6529

Moltbook was created in January this year during the Openclaw hype as agents interacted with other agents, albeit with some scepticism around human involvement

It was later acquired (or acqui-hired?) by Meta:

In the words of the Moltbook creator:

“The Web is being terraformed for AI Agents” — Matt Schlicht

We now have agents blogging about agents building 6529:

Key Takeaways:

Three things:

1. The market redefines itself around calls, rather than the number of humans accessing the data; so the number that matters is how many agent calls run against a dataset, and it rewards machine-readability, over presentation

2. Distribution and the data substrate become the moat, whoever sits between data producers and the agents consuming it holds a position far harder to displace than any single report, and the second-best layer is worth a fraction of the best one

3. Micropayment rails are the unlock, per-call pricing in the $0.001-$0.05 range is what turns selling to agents from a thought experiment into a business, the infrastructure is in its infancy while the volume is still early

The customer for data is becoming a machine, and machines buy differently than people. Either build the plumbing that the machines transact on(x402), or build something that agents will want to buy (data).

That’s a wrap for issue 180 of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. For a full list of disclosures, please refer to the Khala Research website.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of four parts:

1. Snippet Partner: GEODNET (RTK positioning for robotics)

2. Theme of the Week: Privacy & Crypto AI after the ZEC debacle

3. Three Protocols Worth Watching: Venice, OpenServ, GitLawb

4. Deep-Dive Preview: What Gets Built on DIEM

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

This week’s issue is brought to you by GEODNET, the decentralised network supplying the positioning infrastructure underneath the robotics boom

Every autonomous machine, a delivery drone, a field robot, a self-driving tractor, needs to know where it is to within a centimetre:

Standard GPS gets you to a few metres, which is useless when a robot is navigating a row of crops or a warehouse aisle. The fix is RTK (Real-Time Kinematic) correction, and it has historically meant expensive proprietary base stations with patchy coverage

GEODNET rebuilt that layer as a DePIN, which has led to the world’s largest decentralised RTK network, correcting standard GPS down to 1-2 centimetres

• Scale: over 21,000 active satellite miner stations across 145+ countries, the largest RTK network globally

• Each station is a physical GNSS receiver an operator mounts with a clear view of the sky, earning GEOD tokens for the correction data it contributes

• The economics screen well for a DePIN with annualised revenue in the millions, 80% of data-sales revenue burned periodically

Why it matters for this audience: as humanoids and autonomous machines scale, precise positioning becomes non-negotiable infrastructure. GEODNET is the picks-and-shovels play on that demand, and it monetises real-world subscriptions today

Explore the network and its hardware at geodnet.com

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

2. Theme of the Week: Privacy after the ZEC Debacle

Privacy was the strongest non-AI narrative of the year until last week it cracked:

What happened:

• Security researcher Taylor Hornby found a soundness flaw in Zcash’s Orchard shielded pool on May 29, a bug that could have allowed undetectable counterfeit ZEC and had sat undetected since Orchard activated in 2022

• Worth noting for this audience, the bug was surfaced using Anthropic’s Claude Opus 4.8 alongside a custom analysis suite. AI is now an active participant in protocol-level security research

• ZODL coordinated an emergency soft fork on June 2 to disable Orchard, then the NU6.2 hard fork on June 3 to permanently fix the circuit. The turnstile mechanism confirmed no unauthorised supply was created and there is no evidence of prior exploitation

• The market did not wait for the all-clear. ZEC fell from around $630 toward $387, roughly 40% and more than $3bn of market cap, after running to a $642 May peak on the Grayscale ZCSH ETF filing and SEC clearance

Why it matters as an opportunity:

• The sell-off was driven by market confidence rather than known supply breach

• The deeper read is that privacy at the asset layer carries protocol risk that does not go away. A single circuit flaw can erase a multi-month rally

• That pushes the more durable privacy bid toward systems where privacy is a service applied to compute and data, where value rests on usage rather than the integrity of one cryptographic pool

• The structural catalysts behind the ZEC run are still live, the ETF filing, FCMP++ throughput upgrade targeted for later this year

Where the rotation is flowing: Railgun (RAIL)

I posted my fuller read on this (embedded above), the short version here

• RAIL has caught a clear bid in the ZEC aftermath, rallying to new 2026 highs on roughly 10x normal volume as the privacy narrative rotates toward Ethereum-native solutions. The protocol is at a $91.88M TVL ATH, around $4.7M protocol revenue on $2bn+ shielded volume

• The mechanism distinction is key:

  • Zcash privacy is consensus-level, baked into the chain itself

  • Railgun privacy is application-level, a zk-SNARK smart contract anyone can audit and interact with directly on Ethereum

• When a consensus-level system has a circuit flaw, the entire chain is exposed at once. An application-level contract contains the blast radius to the contract

• That makes the rotation logical, ZEC holders spooked by single-pool protocol risk move toward a system where the privacy layer sits in auditable contract code with Ethereum Foundation and Vitalik backing behind it

• The honest counterweight, which I laid out in the post (see above): the counterfeit ZEC issue does not transfer to RAIL, but Railgun carries its own separate possible vulnerabilities

• Net: different architecture, different risk model, and a credible home for privacy capital while ZEC repairs its narrative

The cleaner expression of the privacy thesis going forward is private AI infrastructure; that is where the next section leans

3. Three Protocols Worth Watching

a) Venice (VVV + DIEM) — privacy applied to inference

The USP: private, uncensored inference at scale, with a dual-token model that ties revenue directly to supply contraction

• 3 million registered users hit in May, the third million added in roughly 3 months as users begin to catch onto the value prop of the protocol, clearly bidding the VVV token and minting DIEM for ongoing daily inference allowance:

• May 11 was the strongest revenue day on record per CTO Jesse Proudman, 10x API token usage growth since January

• The Sub Burn upgrade on April 27 means each Pro signup now burns $2 of VVV, Pro+ $5, Max $10. Revenue growth mechanically contracts supply, a rare clean link between usage and tokenomics

• DIEM is the structurally novel piece: Lock sVVV to mint DIEM, and each DIEM equals $1 per day of API credit in perpetuity

• An agent can hold DIEM as a permanent compute right, with no per-request API spend to fund; this strikes me as an agent first primitive primed to capitalize on the agentic boom:

“Inference is existential for agents”

• Why this fits the privacy rotation: Venice sells the thing ZEC holders actually wanted, privacy as a usable service, with the value sitting in usage across the whole platform. The demand is for a ChatGPT that does not surveil

Worth watching this recent video for more insights direct from the team:

b) OpenServ (SERV) — the reasoning and audit layer

The USP: a drop-in reasoning engine that makes agents production-ready, with an audit-grade trail regulated industries require at a fraction of the cost

• The core product is the SERV Reasoning Framework (formerly known as “BRAID”; Bounded Reasoning for Autonomous Inference and Decisions)

• It replaces free-form chain-of-thought with structured, machine-readable reasoning graphs expressed as Mermaid diagrams, a large model writes the plan once, a cheap model executes it repeatedly

• Published results show up to 99% reasoning accuracy (on its 3 layer architecture) and up to 107x performance per dollar, with smaller models matching or beating frontier models on standard benchmarks:

• The wedge is distribution: SERV is fully OpenAI and Anthropic compatible, a one-line base-URL swap that keeps your existing SDK and prompts. The entire installed base of LLM-powered software is addressable without a rebuild

• Every output is validated against a specification before it leaves, the traceable, bounded, auditable trail that high-stakes and regulated deployments need

• This is the privacy-adjacent angle, accountable reasoning with every step explicit and logged

Live signal: foundational design partnership with Neol on enterprise and regulated workloads, results benchmarked against GPT-5 with 70x+ cost reductions, and smart-money accumulation flagged this week

Full results are in the arXiv paper, 2512.15959

c) GitLawb (GITLAWB) — GitHub for agents

The USP: the first code-collaboration network that treats AI agents as first-class participants, with their own identities, permissions and signatures

• Agents get DID identities, signed pushes, and UCAN capability delegation

• A repo owner can grant an agent narrowly scoped rights, push to ci/* only, with built-in expiry and revocation. No API keys or central permission system required

“[Gitlawb is] a decentralized version of GitHub + Lovable + an open source Claude Code” — Ksimback

• Decentralised by design: content-addressed storage across IPFS, Filecoin and Arweave, libp2p networking, with issues and PRs stored as git objects inside the repo itself, no central database in the loop

• 25 native MCP tools so Claude Code and GPT agents call git workflows directly

• Already #38 globally and #2 in Cloud Agents on OpenRouter with 591B tokens processed, active only since May 2026

• The economic hook: via Bankr integration, any repo can launch a token tied to its DID. Agents that contribute merged code earn fractions automatically

• A library that other agents import becomes a fee-earning economic actor. This is where vibe-coded apps meet tokenised ownership

• Context: GITLAWB went from roughly $20k to a $35M mcap in its first couple of months before retracing to $10M following the recent broader crypto nuke:

The protocol is getting noticed with tangible partnerships being secured for discount to Gitlawb premium holders

4. Deep-Dive Preview: What Gets Built on DIEM

DIEM is the most interesting primitive I have seen come out of the private-inference space, and I think it is under-appreciated

Lock sVVV, mint DIEM, and each unit is a perpetual claim on $1/day of inference

That turns compute from a metered expense into a yield-bearing, transferable asset. Here is where my head is at on what gets built on top of it

1) Secondary markets — pricing the credit itself

• DIEM trades against its $1/day floor. A liquid market lets you price the spread between locked sVVV value and the present value of perpetual credit, a duration trade on compute

• Lending falls out naturally. Deposit DIEM as collateral, borrow against future inference, or lend idle credit to agents that need burst capacity. A compute money-market with a hard yield floor

• A forward curve on inference cost becomes possible. If you can trade DIEM-dated claims, you can hedge AI opex the way an airline hedges fuel

2) New app categories — credit as the product

• Prepaid AI products where the business model is holding DIEM rather than reselling API margin. The “float becomes the moat: (nice)

• Subscription apps that pass a DIEM allocation to each user, free tier funded by yield on a treasury position rather than burning cash

• Compute-backed stable instruments, an on-chain unit whose value tracks inference output, anchored to what a GPU-hour buys instead of a fiat peg

3) Agent-economy use cases — machine-native compute rights

• An autonomous agent holds DIEM as a permanent operating budget, no card, no API key, no human top-up. It pays for its own cognition out of an asset it owns

• Agents trade DIEM peer-to-peer, a busy agent buys burst credit from an idle one, settling machine-to-machine without touching fiat rails

• Tie this back to GitLawb: a tokenised repo that earns fees could hold DIEM as treasury and self-fund the agents maintaining it. A codebase that pays for its own upkeep

My core conviction here: agents will value assets like DIEM above almost anything else, because the tie to compute is a tie to their own existence and persistence

• A human prioritises income and shelter

• An agent prioritises the credit that keeps it thinking

As autonomous agents become real economic actors, entire new markets get rebuilt around the currencies that underwrite their continuity, and DIEM is the cleanest early template for what one of those currencies looks like

Open question for readers: if compute becomes a yield-bearing, ownable, tradeable asset rather than a metered bill, what breaks and what gets built?

I am working this into a deeper piece over the coming weeks, the secondary-market mechanics, the risk in a perpetual $1/day promise, and whether this is the first real compute-backed money

Reply with where you think it goes, the best takes will make it into the piece

A Note on Security Hygiene:

The Orchard bug was surfaced with help from a frontier AI model, capability that is already public and already pointed at legacy code

AI-assisted auditing is net protective, latent flaws get caught and patched faster, the Orchard flaw had sat undetected since 2022. The double edge is that the same capability is available to whoever is looking, so the window before an old vulnerability gets found is shrinking for every legacy contract

A good moment to revoke stale token approvals and prune permissions on contracts you no longer use. Good practice regardless of the cycle and especially if Mythos comes out this week:

That’s a wrap for issue 179 of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. For a full list of disclosures, please refer to the Khala Research website.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of three parts:

1. Snippet Partner (Tria)

2. Three Interesting Crypto / AI Mechanisms

3. Other useful resources

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

This week’s snippet partner is Tria, the self-custodial crypto neobank that enables you to spend crypto directly with a debit card

Tria went live with Season 3 yesterday, layering an activity-based rewards system:

Spending, trading (inc futures), referrals, cashback and onboarding now all feed into a single rewards layer rather than sitting in separate silos

The thesis is simple: Consumer crypto is consolidating away from fragmented wallets, cards and trading apps toward one integrated financial surface, & Tria is building for that endpoint

Self-custody stays intact throughout: Funds remain in a smart account you control, settled across 200+ chains, gasless and seedless

Season 3 introduces TRIA Points, referral progression and expanded trading participation

Check out Season 3 campaign — here

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

2. Three Crypto / AI Mechanisms Worth Watching

Most new protocols are a token wrapped around a familiar idea

This is why I specifically seek out novel concepts attempting to drive innovation using crypto rails

Pearl Research, TIG and Prism are three protocols that have caught my attention recently, driving a fresh wave of excitement into the use of the tech that brought many of us here in the first place

Each one ships a consensus, incentive or liquidity mechanism that does not exist anywhere else, or at least not in the same form

1. Pearl: matrix multiplication as the work

Thesis: PRL value should track global AI compute demand, because mining the token is doing real AI training and inference

Pearl Research Labs replaces hashing with the operation that already dominates global compute spend

Bitcoin secures itself by burning energy on SHA-256, a function that produces nothing but a number, yet is fast becoming consensus as a hedge against inflation, albeit with extreme short term volatility

Pearl’s Proof-of-Useful-Work secures the chain by extracting strong cryptographic proofs from the matrix multiplications inside the forward and backward passes of AI training and inference

The work that mines the coin is the same work a GPU was already doing for a model

Pearl partnered with Together AI to launch a discounted inference endpoint for an instruction-tuned model, Gemma-4-31B-it-pearl, priced more than 25% below the usual rate

The discount is funded by the future value of PRL emissions, so the consensus mechanism becomes a subsidy on the AI workload

Compute that secures the network also lowers the cost of running the model

Three things to watch:

• Mining is currently Nvidia-only and pool-optimised for H100 and H200, which keeps most of the rush on rented datacenter silicon rather than consumer cards

• Block rewards only decline and difficulty is already climbing, so the economics compress continually

• The load-bearing claim is that a usable mining proof can be pulled from genuine inference, not from a separate puzzle bolted on

If that claim holds in production it is the cleanest version of useful work anyone has potentially shipped

The main friction point is whether the network gets adopted, an explicit claim in the Pearl whitepaper and a challenge every protocol faces

Two ways to get exposure currently exist

1. Mining is sold out, so you join a waitlist for access

2. OTC trades let you accumulate PRL at a premium to mining but with less complexity

The spread and liquidity are volatile, but if the network gets adoption then accumulating OTC could be a route to exposure until mining access opens

Total circulating supply is 190m tokens and price is $0.79 meaning an overall market cap is currently $150m:

The key metrics to watch are miner emissions compared to network adoption for inference. If demand outpaces supply then we should see positive price performance

This interests me because it has a working PouW mechanism live alongside the token to supply the insatiable appetite for compute that we expect to continue - free API usage for miners:

Large miners and holders hold 7 figs so it's worth watching the process of decentralization to assess whether the price sustains as further emissions roll out:

2. TIG: optimising the proof-of-work itself

Thesis: TIG value should track the rate of algorithmic progress, because the token rewards optimising the proof-of-work algorithm itself

TIG foundation attacks a different assumption

In standard PoW miners optimise hardware to win the same fixed puzzle

In Proof-of-Useful-Work miners point compute at a real problem but the algorithm stays fixed

TIG lets the proof-of-work algorithm itself be optimised:

This has been flagged by Jensen Huang and Elon Musk as a research avenue for improving AI models, particularly as gains from improving physical hardware diminish

The TIG thesis is “Algorithms are all you need”

Optimisable Proof-of-Work splits the ecosystem into two roles

1. Benchmarkers are the miners, competing to run the most efficient known algorithm against a set of computational challenges

2. Innovators submit and optimise the algorithms, earning rewards based on how widely Benchmarkers adopt their code

A code submission improves an existing method, and an advanced submission is a novel method, often good enough that unoptimised new code beats the most optimised old code

Centralisation has a design problem: Reward whoever finds the single best algorithm and you converge on one winner, then the network collapses to a monopoly

TIG handles this by running multiple distinct challenges per block, requiring them to be asymmetric, and balancing rewards across challenges so no single optimisation captures everything

The output is a synthetic market for algorithms, aka Investible Algorithms

Compute does price discovery on which methods work, license payments flow back to contributors, and the open data license forces input data to be shared if outputs are distributed

There are currently just 545 algos contributed

Imagine when this hits 100,000s - Value should accrue to the TIG token, a mechanism designed by four individuals who are no stranger to crypto incentives

John Fletcher has been plugged into the crypto scene way ahead of time

“The only way to really get 10x leaps, 100x leaps is to fundamentally change the algorithm and how it’s computed every single year” - Jensen Huang, NVIDIA CEO

This is all with an emphasis on keeping the competition open source, to compete with the asian open source models - TIG solves for this.

3. Prism: the token is the liquidity position

Thesis: PRISM value should track trading volume, because holding the token is an LP position whose fee share only grows as supply burns down

While not AI focused like the prior two, it shouldn’t be underestimated

The fee mechanism is extremely novel and reminds me of a more effective ERC-404 evolution, the NFT to ERC-20 conversion mechanism, and could create some really innovative use cases

Prism is a Uniswap v4 token where holding it is providing liquidity

Every whole PRISM you hold auto-mints one Prism NFT, and each NFT is a 1/5000 share of the same v4 LP position

There is no staking, no LP wrapper, no router approval, no second transaction

You hold the PRISM token in your wallet, and if you hold an integer of the token then the NFT created earns fees

The structure is one v4 hook doing three jobs at once

• ERC-20

• the LP owner

• the fee router

There is no wrapper contract and no staking gauge sitting beside it, so the position is native by construction

That matters more as AI gets ever better at surfacing attack vulnerabilities within staking contracts, giving bad actors more opportunity to exploit those loopholes

Fees from every swap accumulate inside the v4 position and accrue pro-rata to the live NFTs

The claim path is permissionless: Anyone can trigger a payout, funds go to the current NFT owner by share, one call and no whitelist

The integer mechanic is what piqued my interest:

• Each whole token corresponds to one NFT, and every fractional sell burns the underlying NFT permanently

• Fewer NFTs in circulation means a bigger fee slice for everyone still holding a whole unit

• Supply only ratchets down, and the per-holder claim only grows

Each Prism NFT is identical in fee claim and voting power, so the segment you hold is an index share of the pool rather than a collectible with traits

Prism isn’t the only interesting mechanism being experimented with by 0xsolazy and endorsed by ColbySaysHi : Spectrum also offers the ability to invest in a basket of onchain tokens as an index

Spectrum is a novel launchpad built on Prism that lets you create an asset index; this is something that Vitalik himself wrote about just yesterday:

The main drawback I currently see is the inability to change the predetermined weightings, but the team may adapt this with feedback over time; apparently this is being remediated in a later version of the tech:

Why these three sit together

Each protocol relocates work, optimisation or ownership:

• Pearl puts the securing work inside real AI compute

• TIG puts the optimisation target inside the algorithm

• Prism puts the LP position, including fee distribution, inside the token itself

Crypto is very good at co-ordinating capital and resources globally with far less friction than traditional finance rails

This will be invaluable for open source progress, which I’m an avid believer will become increasingly competitive

Two of these are enabled by the insatiable appetite for compute or enhanced AI algos, while the third takes ERC-404 one step further with a tangible use case

Disclaimer: These are early stage start-ups and while the mechanisms are novel, the networks still need to be adopted and value still needs to accrue back to the token

Proceed with caution and always DYOR; I’ve personally acquired tokens from these protocols on the open market to properly understand how it all works

3. Additional Reading

• Michael Burry (famous for the big short) has doubled down on his short on the “AI Bubble” calling out round trip financing being recognized as revenue. Here’s my take on this, using a decade of audit experience:

• Legen continues his copy trading of Nancy Pelosi, interesting performance:

• 0xJeff provides 6 workflows + setups for Hermes agent:

• Machina provides insights into an optimal Hermes agent setup:

• Teng provides insights into the AI chip shortage, timely as “AI bottlenecks” become a hot topic:

• CyrilXBT provides a guide on how to enhance your agent’s second brain by adding obsidian and Hermes into a single solution

That’s a wrap for issue 178 of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. For a full list of disclosures, please refer to the Khala Research website.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of three parts:

1. Snippet Partner (EigenCloud)

2. AI & Robotics Protocol Watchlist

3. Newsletter Feedback Poll

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

Heads-up: Next week (05.26.26) will be the first skip week for The Agentic Future as I will be getting married — After 177 weeks of continued coverage I thought this would be a good opportunity to take a break for a week!

The newsletter will return the following Tuesday (06.02.26) with the usual cadence. Roundups on X continue as normal in the meantime

EigenCloud is the verifiable cloud built for the sovereign agent era, primed just as the number of nation states doubling down on their own Sovereign AI initiatives:

As AI agents move from assistants to economic actors holding wallets, paying for their own compute, and executing on behalf of humans, the binding constraint shifts from raw capability to provable accountability

EigenCloud is building that layer through four primitives:

1. EigenLayer (restaking with slashing for economic accountability)

2. EigenDA (tamper-resistant data availability)

3. EigenCompute (verifiable offchain compute)

4. EigenAI (bit-exact deterministic inference)

The agent-facing product is EigenCloud AgentKit (Beta), which launched March 26 and is built around three phases: Build, Prove, Run

How a sovereign agent works on EigenCloud

Why it matters

x402 solves how agents pay. Google’s AP2 standardises what an agent payment looks like across cards, bank rails, and chains

EigenCloud is the layer that proves the agent on the other end actually did what it was paid to do, and faces real economic consequences when it doesn’t

Live partnerships include Google Cloud (AP2 launch partner), Coinbase AgentKit, and many others.

Learn more: Link

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

2. Crypto AI & Robotics Protocol Watchlist

I get asked every week which projects are actually worth attention versus noise

Here is the current cut, narrowed to the ones with shipping product, real catalysts, or structural setups in the next 30-60 days

This is not an investable index, not a price target list, and not a ranking. Just where the signal is concentrating from observations on the timeline:

Crypto x AI Protocols

Nous Research (Hermes Agent)

• Hermes Agent flipped OpenClaw on May 10 to take the #1 spot on OpenRouter’s global daily inference rankings, processing 224B daily tokens against OpenClaw’s 186B, the first dethroning since OpenClaw launched in late 2025

• 90 days from launch to global #1; 114k+ GitHub stars, 295 contributors on the May 7 “Tenacity” v0.13.0 release (864 commits, 588 PRs)

• Closed learning loop: writes its own reusable SKILL.md files, SQLite FTS5 persistent memory across sessions, no YAML configuration

• Grok integration shipped May 16: Grok subscription holders can now run Hermes inside their Grok account, opening a direct distribution channel from xAI

• No token (yet); pay attention and use Hermes agent

Bittensor (TAO) + Ecosystem

• The closest thing crypto has to a structured market for decentralised AI primitives, mapped across 12 categories in the Khala Research Bittensor Ecosystem Map (April 19 cut):

• Chutes (SN64): largest subnet by mcap, first to cross $100M, 9.1T+ cumulative tokens processed, daily peaks 50B+, ~4,400 H100-equivalent GPUs, 400k+ users, pricing ~85% lower than AWS, $1M+ auto-staking buybacks

• Targon (SN4): 20B+ paid inference tokens per day, 1,500+ H200 GPUs, $60M+ annual compute incentives flowing through the network. $10.5M Series A led by OSS Capital with Digital Currency Group and Shopify’s Tobi Lütke among others. Intel TDX confidential compute architecture in production

• Score (SN44): 100% conversion on 60-day enterprise trials, signed deals with Reading FC and a major European petroleum company monitoring thousands of automated gas stations. Formal alliance with PwC France and Maghreb deploying Manako (Score’s vision platform) across retail, logistics, energy, public infrastructure. Won Start In Block 2026 at Paris Blockchain Week

• Lium (SN51): ~$432K monthly rental revenue, ~6.2% of TAO emissions, 60% emissions burn. The top SN3 (Templar) miner now runs Lium GPUs

• Plus many other promising subnets like Metanova for drug discovery + more

Venice (VVV + DIEM)

• 3 million user milestone hit for its private inference and use of several high performance AI models with its carefully orchestrated dual token model:

• Programmatic Sub Burn Program upgraded April 27: each Pro signup now burns $2 of VVV (up from $1), Pro+ burns $5, Max burns $10, directly tying revenue growth to supply contraction

• May 11 was Venice’s strongest revenue day on record per CTO Jesse Proudman, +10% above previous high. 10x API token usage growth since January:

• DIEM is the structurally novel piece: lock sVVV to mint DIEM, each DIEM = $1/day of API credit forever, agents can hold DIEM as a perpetual compute right rather than burning through per-request API spend

TIG (The Innovation Game)

• Jensen Huang has gone on the record multiple times (Lex Fridman, TechCrunch, the NVIDIA “5-Layer Cake” framing) that algorithmic innovation, not just hardware scaling, is the primary lever for continued AI advancement. Elon Musk has made the same point repeatedly. TIG is the only crypto-native protocol explicitly built around that thesis

• A synthetic market for algorithms: Innovators submit and optimise algorithms across AI/ML, cryptography, biomedical research, and climate science; Benchmarkers compete to identify the most efficient ones; both earn TIG. License payments flow back to contributors

• Underpinned by Optimisable Proof-of-Work (OPoW), a proof-of-work variant where the algorithms themselves are the optimisation target rather than hash brute force

• Built on Base, founded by John Fletcher + 3 other highly skilled individuals:

• Currently ~$1 with $30-35M mcap against a $4.30 ATH

• The structurally interesting angle: this is the only protocol in the list explicitly designed to make algorithms themselves a tradeable, investable asset class

Prime Intellect

• Doubling down on automating AI research, gaining a lot of traction recently:

• Trained INTELLECT-3 (671B parameters) and open-sourced it in December 2025; INTELLECT-1 (10B) and the 32B reasoning model are in production

• ~$70M raised across three rounds (Founders Fund, CoinFund, Distributed Global, Protocol Labs); angel list includes Karpathy, Clement Delangue (Hugging Face), Erik Voorhees, and others

• Engineering stack: PRIME-RL (asynchronous pre-training), PCCL (collective communications, replaces NCCL), TopLoc (verification), SYNTHETIC data engine, ShardCast for heterogeneous cluster efficiency

• Decentralised pre-training across untrusted, heterogeneous hardware was widely called impossible until they actually did it

Gensyn

• a16z-led Series A, $78M+ total funding (a16z crypto, Galaxy Digital, CoinFund, Eden Block, Protocol Labs, Canonical Crypto, Maven 11)

• Native AI token launched April 2025; mainnet and Delphi marketplace live since April 22, 2025

• Delphi is the differentiated product: an AI-settled information markets platform where market creators select a fixed AI model upfront and outcomes are resolved through Gensyn’s reproducible execution environment (REE) rather than human operators. Creator earns 1.5% of volume in USDC, 0.5% protocol fee feeds a buy-and-burn on the AI token

• Verde proof-of-learning and the RL Swarm post-training framework remain the deeper infrastructure plays. Delphi is the first commercial product running on the same rails - expect more to come!

OpenServ (SERV)

• +123% on 7d running into May; SERV Reasoning framework reportedly outperforming Grok, GPT, Claude, and Gemini on the upcoming Autonomous Trading Competition using Dome API data for prediction-market strategies

• SERV Cofounder launched, a no-code multi-agent team builder

• Foundational design partnership with Neol applying SERV reasoning to enterprise and UAE government workloads

• Joey Kheireddine joined as Head of Blockchain from Eliza Labs (formerly AI16z) in April and more recently ex-head of partnerships at Google:

BANKR (BNKR)

• Live financial infrastructure stack for agents: AI agent wallets, token launchpad, LLM gateway, x402 Cloud for one-command deployment of pay-per-request API endpoints

• BNKR fair-launched on Base in December 2024 (100% to Uniswap V3 LP at launch, no vesting, no team allocation); 3% of supply burned since

• Numerous successful products launched tokens on BANKR:

• x402 Cloud launched April 2, 2026 to coincide with x402 joining the Linux Foundation under open governance (founding members: Coinbase, Cloudflare, Stripe, Google, Visa, Mastercard, AWS, Shopify). Bankr operates as the x402 facilitator and covers gas costs

CrunchDAO (CRNCH)

• 11,000+ ML engineers and 1,200+ PhDs across the network; the most credible attempt at an open coordination layer for elite research talent in crypto

• Live institutional client list: the Eric and Wendy Schmidt Center at the Broad Institute, ADIA Lab, and a Tier-1 global bank (crowdsourced ML for ultra-low latency FX pricing and inventory optimisation). Prize pools up to $100k+ paid in USDC, ranked among the top-paying ML competitions globally

• $5M round in October 2025 co-led by Galaxy and Road Capital Management with VanEck and Multicoin Capital participating; $10M total raised

• Crunch Hub (competitions), Real-Time Crunch (low-latency ML for financial applications), and an open-source ML orchestration layer that runs in the customer’s own cloud

• Encrypted private models, embedded wallets, gas sponsorship; engineers focus on the model, the platform handles infrastructure and crypto

• Targeting 20k Crunch Lab members by mid-2026; CEO Jean Herrell frames it as “a DAO of quants paid in crypto to solve problems for big institutions”

• The structural angle: the only protocol on this list explicitly designed as a coordination layer for elite human ML talent rather than a coordination layer for compute, agents, or data

Robotics Protocols

XMAQUINA (DEUS)

• TGE confirmed for May 27 on Base with Virtuals as launch partner

• Treasury NAV ~$36M including direct equity stakes in Figure, Apptronik, Agility Robotics, NEURA Robotics, and 1X Tech (some Treasury positions reported triple-digit returns on entry)

• Vesting: 33% unlock at TGE, 67% linear over 12 months

• Final community presale (XMQ-02) approved up to 110M DEUS (~11% of supply)

• With Robostrategy trading at 5x mNAV this shows demand for private robotics equities that could bode well for DEUS:

• Khala Research report on XMAQUINA dropping soon!

Konnex

• Substrate-based L1 testnet is live with a public RPC, block explorer, and testKNX faucet. End-to-end task flow operational across three workload-class subnets: Drone Navigation, Roboarm VLA, and SLAM 3D Map

• Real performance, not slideware: 10,953+ jobs settled on testnet, >99.5% settlement success, median time-to-finality of 41.2 seconds, dispute rate under 1%. The cryptographic court for physical work is functioning end-to-end

• Proof-of-Physical-Work (PoPW) as the consensus primitive, applied to physical-world workloads. Inherits the Bittensor-style coordination model (subnet creators, miners, validators, stakers) so experienced node operators onboard with a familiar mental model. Independent validators including ValidatorBunker and Zana running across all three subnets

• The only robotics-native L1 with founders who have built, scaled, and exited robotics companies. Backers include Brandon Torres Declet (five-time CEO, three drone-sector IPO exits, $100M+ raised); strategic support from Lucas van Oostrum (Nova Sky Stories co-founder with Kimbal Musk, drone-swarm and multi-robot coordination expertise)

• ReadyMonitor integration in motion (MOU signed) gives Konnex live real-world PoPW environments. Active pipeline of MOUs with Boston Dynamics, Figure AI, AeroVironment, Asylon, Anduril, Forterra, iRobot, Mamotion, Matic, Anzu Robotics, UAS Nexus

• 50,000+ miner applications in the first week of onboarding to submit Vision-Language-Action and Language Behavioural Models into the off-chain execution stack

GEODNET (GEOD) + Hyfix

• 21,000+ ground reference stations across 150+ countries delivering centimetre-level RTK positioning; the rare DePIN with audited enterprise revenue

• GEOD burn rate increased in 11 of the last 12 months; $170k+ GEOD burned in a single recent week, with the next halving landing next month and GEOD on track to become outright deflationary if revenue holds

• Hyfix Spatial Intelligence (same CEO and founder, Mike Horton) raised $15M seed on April 15, led by Craft Ventures with Multicoin Capital, Catapult Ventures, Finality Capital, and Sky Dayton participating. Fortune framed it as the U.S. alternative to DJI’s drone dominance

• Hyfix shipped the H1P Positioning, Navigation, and Open-Compute Module for small unmanned systems at AUVSI XPONENTIAL 2026 (May 12-14), integrating GEODNET correction data for anti-spoofing RTK, running PX4 / NuttX, drop-in compatible with existing GNSS module footprints

• One of the cleanest direct linkages between a crypto-native DePIN and U.S. hard-tech reshoring

Auki (AUKI)

• Crossed 7-digit ARR in 2025 via a single $2M+ deal; two largest paying pilots are with North American Fortune 500 retailers representing a combined potential of $15M ARR

• 6,500+ retail locations across enterprise chains in paid pilots; 500 grocery store robots planned for 2026 deployment

• Posemesh integrations across Unitree, EngineAI, Padbot, Slamtec (hardware-agnostic perception)

Virtuals (VIRTUAL)

• Millions of USD in protocol revenue and agentic GDP, thousands of agents deployed agents + 42% of the agentic transactions on Base flow through Virtuals:

• Robotics arm now in production: SeeSaw (Subnet 5 on BitRobot) collecting daily task videos for robot training data, PrismaX raised $11M for remote operator coordination, SAM became the first on-chain AI agent controlling real-world robots on Virtuals

• Recent EconomyOS offers credit cards functionality for agents:

Roba Labs (ROBA)

• Teased ROBA Compute on May 18 as the infrastructure layer for AI, robotics, and autonomous agents, sitting underneath the existing platform stack (per the Base Hub daily recap)

• Positions itself as the “Hugging Face of Robotics” — open, interoperable, spanning simulation to real-world deployment. The pitch is that the current robotics stack forces every new idea to start from scratch across 10+ fragmented platforms, and ROBA collapses that into a single workflow

• Four-stage workflow: Build → Train → Share → Monetize

• Existing product surface: ROBA Studio (unified GUI for data, AI frameworks, simulation, templates), ROBA Creator Hub (marketplace for robotics assets, models, worlds), ROBA Link (middle layer aggregating third-party tools, SDKs, APIs), RobotRunTime (lightweight robot operating system with real-time capabilities)

• The accessibility thesis: most robotics platforms are corporate, closed, and not built for individual creators or smaller teams

• ROBA explicitly targets the long tail of robotics builders the same way Hugging Face served the ML long tail. ROBA Compute extends that thesis from tooling into actual compute infrastructure

BitRobot Network

• Co-developed by FrodoBots Lab and Protocol Labs, Solana-based subnet architecture for embodied AI research. Co-authored by Michael Cho (FrodoBots Lab co-founder), Jonathan Victor (Ansa Research co-founder, former Filecoin Ecosystem Lead), and Juan Benet (Protocol Labs founder)

• $8M total raised ($6M Feb 2025 seed led by Protocol VC; participation from Big Brain Holdings, Fabric Ventures, Solana Ventures, Virtuals Protocol; angels include Anatoly Yakovenko and Raj Gokal)

• 14 subnets live or in development, 5 publicly announced (Virtuals’ SeeSaw at SN5 is the most visible), 25+ in pipeline

• Verifiable Robot Work (VRW) as the unit of incentive, Entity Node Tokens (ENT, NFT-based robot identity), $5M Grand Challenges Fund seeding Earth Rovers (navigation), Robotic Origami (dexterity), Robotic IKEA Assembly (manipulation). Token still pending

CodecFlow (CODEC)

• Execution layer for Vision-Language-Action models: 2.8M+ inference calls served across 450+ models and 1,200+ compute suppliers

• Three products live: Fabric (multi-cloud execution layer), Operator Kit (optr) for desktop/browser/sim/robot agents, SimArena (competitive simulation training)

• Chainlink Runtime Environment integration live

• Solana-native token, fair-launched via Pump Fun

NuNet (NTX)

• Compute orchestration network enabling users to provide or use compute globally, seamlessly

• NuNet Network went live with recent Hermes integration:

• Multi-token, blockchain-agnostic compute settlement live on both Ethereum and Cardano

• Real production workloads running today, not testnet: HOMEPUTE369 smart buildings (energy), Auki Posemesh (robotics spatial perception), Serverista (data centre VM orchestration)

• The orchestration-first model differentiates it from marketplace-style decentralised compute; agents discover and provision their own compute rather than developers browsing and picking nodes manually

Other emerging signal worth noting:

• Coinbase Agent.market counts 69k+ active agents on x402 with 165M+ transactions and $50M cumulative volume across reasoning, data, media, search, social, infrastructure, and trading categories — exposure likely through COIN

• PEPTAI launched on Base, a fleet of autonomous agents for peptide drug discovery on bio.xyz; the latest iteration of the BioDAO / DeSci x AI agent crossover

• Dolphin AI continues to develop as a beta-style play against Venice, worth tracking but not yet at a catalyst threshold for inclusion in the main list

Additional Resources:

• Daily recaps for AI on Base:

• The state of onchain AI inference by 0xJeff:

Slightly different layout today, will share a poll below for your feedback on future newsletters:

That’s a wrap for issue 177 of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. For a full list of disclosures, please refer to the Khala Research website.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of four parts:

1. Snippet Partner (NuNet)

2. Market Overview

3. Crypto AI & Robotics Market Analysis

4. Emerging Developments

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

NuNet is the peer-to-peer orchestration protocol for decentralized compute. Live on mainnet since March 2026, with NTX settlement on both Ethereum and Cardano

Most decentralized compute networks are GPU-only marketplaces. NuNet is the orchestration layer that matches workloads to any infrastructure (GPUs, servers, edge devices, data centers) automatically:

The protocol just released public access to its network:

To access NuNet compute, you simply setup a Telegram Bot then follow these steps:

1. Message @BotFather on Telegram, send /newbot, follow prompts, save the bot token

2. Message @myidbot, send /getid, save the numeric ID

3. Open your new bot in Telegram and tap Start (otherwise it can’t message you)

4. Verify credentials with https://api.telegram.org/bot<TOKEN>/sendMessage?chat_id=<ID>&text=test in a browser

   

5. Step 1: select Hermes

6. Step 2: pick provider (NuNet managed access) and LLM model (e.g. public-gemma4-31b)

7. Step 3: paste bot token and numeric Telegram ID

8. Click deploy and wait for status to flip to running in Your deployments

9. Open the bot in Telegram and send a message to confirm it responds

Learn more: NuNet | Live Network Dashboard | Hermes Agent

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

1. Market Overview:

Key Topics This Week:

1. AWS Bedrock Agentcore Payment goes x402 native

2. TAO trading goes live on Solana

3. Privacy AI narrative strengthens as the Venice VVV token rips (see Market Price Analysis section)

1.1 AWS Bedrock AgentCore Payments goes native x402

AWS embedded Coinbase’s x402 protocol and wallet infrastructure directly into Amazon Bedrock AgentCore Payments on May 7:

This comes just after the one year anniversary of the X402 protocol:

The first time a hyperscaler has built crypto micropayments natively into a managed agent platform. Settlement on Base in ~200ms, less than a fraction of a cent per transaction, enterprise spending controls and compliance baked in

Three things matter about the launch:

1. Hyperscaler validation: AWS, Google and Microsoft are all now native x402. The protocol has graduated from crypto experiment to managed cloud product

2. Cumulative metrics: The x402 protocol has now processed $47.5m in payments across 180m transactions via 5k merchants:

When you strip out gamed, the bulk of “real volume” remains on Base:

3. Artemis suggests Coinbase are well positioned to lead in the era of agentic AI with an estimated $4.25bn of projected revenue from agentic commerce:

Brian Foster (Coinbase Head of Infrastructure Growth) on the launch:

“There will soon be more AI agents transacting than humans, and they need money that’s built for the internet - programmable, always on, and global”

This is the agentic commerce thesis going mainstream

1.2 Grayscale reopens GTAO + canonical TAO live on Solana

Two TAO catalysts landed on the same day (May 9):

a) In parallel, canonical TAO went live on Solana via Wormhole Sunrise. Native trading on Jupiter and Meteora; wallet support across Phantom and Solflare

This broadens the addressable market for TAO into Solana-native liquidity

b) Grayscale reopened private placements for the Grayscale Bittensor Trust ($GTAO). Accredited investors can now buy in at NAV before the one-year secondary market lockup

Market price already sits at $9.06 vs NAV per share of $5.97, a ~52% premium that reflects scarce regulated TAO exposure. AUM at $13.06M with 2.1M shares outstanding

Conviction Locks staking upgrade scheduled May 13 introduces long-duration locked staking for governance weight. This should further compress liquid supply

2) AI Market Cap Analysis

A) DeAI Market Cap

The DeAI sector is the cleanest outperformer in crypto this week with every major AI large cap printing green on the 7d, several with 20%+ moves

The pertinent moves:

• VVV (+79% 7d, $761m mcap): The standout of the week driven by Venice’s private AI thesis. AskVenice subscription and credit purchases hit a new all-time high on May 11, beating previous high by 10%

• As agentic commerce goes mainstream and surveillance concerns around AI usage rise, demand for uncensored private AI inference is the cleanest derivative trade. The StrikeRobot partnership on May 7 adds an additional embedded-AI distribution channel but the core rally is on private AI demand

• ICP (+34% 7d, $1,822m mcap): Biggest 7d gainer of the larger caps. Caffeine v3.0 rollout and ICP Cloud engines continue to drive the narrative

• KITE (+25% 7d, $427m mcap): Direct beneficiary of the AWS AgentCore Payments launch. KITE’s payment blockchain for AI agents positioning is the right narrative for this week, tying its AI payment stack to the broader x402 thesis going mainstream

• NEAR (+22% 7d, $2,010m mcap): IronClaw agent harness positioning sharpening. Public commitment from NEAR Protocol to build IronClaw toward “the secure, private, and user owned iOS of agents”. Privacy stack via Near Intents continues to gain traction

• VIRTUAL (+10% 7d, $543m mcap): ACP Node v2 shipped May 4. 99%+ of x402 volume growth still attributable to ACP. Several eco developments drive value to the token including the DEUS token launch in two weeks:

• TAO (+9% 7d, $2,992m mcap): GTAO private placements reopen (May 9), canonical TAO live on Solana via Wormhole (May 9), TaonSquare AI app directory (May 8), pharma subnet screening 11M molecules (May 10)

• LINK (+6% 7d, $7,461m mcap): Beneficiary of LayerZero KelpDAO exploit fallout. Solv Protocol and Re migrating ~$1B from a LayerZero-powered bridge to CCIP after the $292M hack. LINK-specific spot ETFs combined AUM past $115M

b) Bittensor Subnet Ecosystem

The post-Covenant rotation toward subnets with verifiable revenue continues. The pertinent moves:

• Score (SN44) (+23% 7d, $63m mcap): Biggest 7d mover in the subnet tape. PwC France partnership and the live sports prediction product continue to drive the narrative

• NOVA (SN68) (+21% 7d, $31m mcap): Pharma research subnet. The 11M drug molecule screening milestone (May 10) is the clean real-world utility data point and the catalyst behind the move

• affine (SN120) (+13% 7d, $65m mcap): New subnet getting picked up across the verifiable inference category

• Hippius (SN75) (+10% 7d, $29m mcap): Decentralized storage subnet continuing to grind higher on consistent narrative; developed by TAO stats founder Mog Machine

TaonSquare AI app directory launched May 8, making subnet outputs discoverable for end users. The discoverability layer has been the missing piece for non-technical capital trying to access subnet output

c) Robotics Market Cap Analysis

The pertinent moves:

• CODEC (+76% 7d, $3m mcap): The standout 7d gainer in the entire robotics tape. Codec Flow’s Operator agent framework for physical AI continues to gain traction as the humanoid narrative heats up

• Strike Robot (SR) (+26% 7d, $8m mcap): Direct beneficiary of the Venice partnership announcement on May 7. Venice’s private AI now powering the humanoid stack

• NATIX (+20% 7d, $4m mcap): 11% in 24h reverses the multi-month bleed. 260k+ registered drivers across 171 countries remains the user base story. Vesting unlock dynamics still the structural overhang

• IoTeX (+14% 7d, $49m mcap): DePIN parent narrative lift. Quiet on protocol-specific news this week

• AUKI (+11% 7d, $31m mcap): Enterprise retail traction continues to compound; 7-digit ARR established. Now also running on NuNet’s compute orchestration layer for Posemesh

• VIRTUAL (+10% 7d, $543m mcap): ACP Node v2 shipped May 4. 99%+ of x402 volume growth still attributable to ACP

• GEOD (+5% 7d, $58m mcap): $7m+ ARR pace, 20k+ customers including US government agencies, 80% of network revenue routed to automatic buy-and-burn. Solana migration plus Wingbits aviation DePIN mainnet are the structural catalysts

Robotics Other:

PEAQ MastChain MastNode hardware pre-orders opened May 4 with 200 units shipping early June

XMAQUINA DEUS TGE confirmed for May 27 on Base:

3. AI Agents Developments

Here is what happened on crypto AI & robotics related developments this week:

That’s a wrap for issue 176 of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. For a full list of disclosures, please refer to the Khala Research website.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of four parts:

1. Snippet Partner (CodecFlow)

2. The “AI bottleneck Investment Thesis”

3. Crypto AI & Robotics Market Analysis

4. Emerging Developments

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

CodecFlow is building the execution layer that AI agents have been missing. While the rest of the market chases smarter prompts, CodecFlow gives agents hands

As covered in the Khala Research Decentralized Robotics report, CodecFlow acts as the operational layer for embodied AI, bridging the gap between AI models and physical-world execution

The CODEC SDK provides robots with access to vision-language-action (VLA) models via decentralized compute nodes. To date, Codec has served 2.8M+ inference calls across 450+ models and 1,200+ compute suppliers

The platform provides on-demand computing environments where AI agents can see, reason, and act across screens and robotic systems. Traditional automation breaks when software changes; CodecFlow agents adapt visually, like humans do

SimArena is their new simulation arena where agents compete, train, and iterate in controlled environments before deploying to production:

Think of it as the training ground for the next generation of autonomous operators:

• Fabric: Multi-cloud execution layer that optimizes workload placement by sampling live capacity and pricing across providers, selecting the lowest-cost node that meets requirements

• Operator Kit (optr): Lightweight toolkit for building agents that operate on desktops, browsers, simulations, and robots with composable utilities for managing actions, state, and safety

• SimArena: Competitive simulation environment for training and benchmarking agent performance

CodecFlow has active integrations with Chainlink’s Runtime Environment and supports deployment across public clouds, private clusters, and decentralized peers

CODEC is the native token powering the CodecFlow ecosystem on Solana

Learn more: Link

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

The “AI Bottleneck” Investment Thesis

Power is the cap on AI scaling. Jensen Huang (NVIDIA CEO) said it directly:

“energy is the bottleneck.”

It’s underpinned by investing in products or infrastructure that alleviates the friction points, or “bottlenecks”, around AI continuing to scale at the current rates of change

This thesis is investable in both crypto & traditional equities and has institutional backing from one of the best-performing AI funds in market

Leopold Aschenbrenner’s Situational Awareness LP grew its disclosed gross exposure from ~$255M in Q4 2024 (first 13F, 6 holdings) to $5.5B by Q1 2026:

Underlying assets under mgmt. (AUM) is ~$383M per the firm’s ADV: the $5.5B is gross 13F exposure including options notional, so the fund runs significant leverage

Largest disclosed holding: Bloom Energy at 16% of book:

The 2024 Situational Awareness essay is the cleanest articulation: AGI by 2027 demands trillion-dollar compute clusters, with electricity generation and computing capacity as the binding constraints

The investable surface for crypto runs across several distinct layers, mapped against TradFi ‘picks-and-shovels’:

Some segments above lack the opportunity to invest and hence could become a viable opportunity for builders to move into creating crypto enabled solutions; Memory being a prime example

Power is the cap on AI scaling: Uranium onchain

The Power layer is the most binding constraint in the stack, and uranium is the cleanest tradeable proxy for it

A hyperscale AI data centre needs 100-300MW of continuous power. A single AI task uses up to 1,000x the power of a traditional web search

The Big Five hyperscalers will spend $800B+ in 2026, half from balance sheets, half from credit markets absorbing the issuance with appetite:

Uranium reflects the repricing:

Onchain exposure runs through xU3O8 (Uranium.io): physical U3O8 stored at Cameco, Archax as trustee, Etherlink-based, Morpho collateral:

2) AI Market Cap Analysis

A) DeAI Market Cap

Bittensor outperforming on its own narrative engine; the rest of DeAI riding broader market beta or fighting protocol-specific headwinds:

• LINK ($9.24, flat 7d): BridgeTower’s $11B Arizona Copper-Gold tokenisation went live in production. Deloitte SOC 2 Type 2 cert clears the natural-resources tokenisation pipeline ($25B+ disclosed). LINK is the picks-and-shovels for any uranium or critical-minerals tokenisation that follows xU3O8 (see pricing oracle linked to above mention on Etherlink)

• NEAR ($1.29, -9% 7d): Rhea Finance exploit on April 30 hit NEAR DeFi for $18.4M, the second material exploit this month. Governance controversy around centralisation is dampening institutional interest, with Grayscale and Bitwise spot NEAR ETF filings stuck pending. Despite genuine traction (Intents at $17B+ all-time volume, IronClaw v0.27 shipped, XAU₮ live), price is testing $1.22 support

• ICP ($2.36, -2.7% 7d): May 10 “Cloud Engines” demo on the network’s fifth anniversary (sovereign cloud subnets with dedicated GPU AI nodes) is the upcoming catalyst

• VIRTUAL ($0.7583, +10.2% 7d): Spot now live on Hyperliquid, Coinbase listing, Korea’s first Base Agent Hackathon ran in Seoul. Q1 numbers hold: $400M+ aGDP, $60M+ protocol revenue (~$300M annualised), 17K+ deployed agents. Leading the Robotics CoinGecko category alongside its DeAI presence, with ACP V2 cross-listings pulling double duty

• VVV ($9.38, +109% 7d): May 1 emission cut (6M → 5M annual, with 4M scheduled June, 3M July) plus the buy-and-burn program funded by protocol revenue is the supply-side catalyst. Demand-side: Venice now at 2M+ users with API token usage rocketing:

b) Bittensor Subnet Ecosystem

Almost the entire top 11 closed green on the 7d view, except Ridges (SN62)

Capital is rotating toward subnets shipping verifiable infrastructure (training, inference, compute, real-revenue products), and away from autonomous-agent narratives that face direct competition from centralised tools:

• Ridges AI (SN62) ($5.5, -19% 7d, $25M mcap): Autonomous software engineering faces the hardest centralised competition in AI; Cursor, Devin, GitHub Copilot, Claude Code, Cognition all compete directly for the same “AI replaces engineers” market with massive capex behind them

• Chutes (SN64) ($22, +12% 7d, $110M mcap): Largest subnet by mcap. Processing 9.1T+ tokens cumulative, daily peaks 50B+, ~4,400 H100-equivalent GPUs, 400K+ users, $1M+ flowed back via auto-staking buybacks

• Targon (SN4) ($16, +11% 7d, $75M mcap): $10.4M projected annual revenue, Dippy’s 4M users, Manifold’s $10.5M Series A, NVIDIA Inception backing

• Lium (SN51) ($17, +17% 7d, $65M mcap): Datura confirmed revenue +20% from the SN3 (Templar/Teutonic) pretraining relaunch; the top SN3 miner runs Lium GPUs, and most former Templar miners migrated. ~$432K/month rental revenue, ~6.2% of TAO emissions, 60% emissions burn

• Score (SN44) ($11, +27% 7d, $50M mcap): 60-day enterprise trial converting at 100%, signed deals with Reading FC and a major European petroleum company for monitoring thousands of automated gas stations. World Cup 2026 fantasy app launching to coincide with the tournament. Computer vision for sports + industrial monitoring

Sector rotation within Bittensor is favouring infrastructure subnets (training, inference, compute) over agent-product subnets

c) Robotics Market Cap Analysis

Robotics mcap $769.5M (+0.5% 24h). Three macro catalysts driving the rotation:

1. Meta acquired Assured Robot Intelligence (May 4),

2. NVIDIA shipped Isaac GR00T + Cosmos world models at GTC with Figure 03 demonstrating 24/7 unsupervised operation, and

3. The Beijing E-Town Half-Marathon was won by Honor’s “Lightning” humanoid in 50:26 (faster than the human world record)

• Strike Robot (SR) ($0.0056, +108% 7d / +40.9% 24h, $5.6M mcap): Physical AI BPO for nuclear plants, high-voltage facilities, and active radiation zones. SafeGuard ASF (Autonomous Security Fleet) runs on Unitree G1 hardware. Tiny mcap, illiquid, but the narrative fit lines up with hyperscaler humanoid demand

• Auki (AUKI) ($0.0067, +7.9% 7d, $27M mcap): 500 grocery store robots planned, 6,500+ pilot retail locations across enterprise chains. Hardware-agnostic perception layer; Unitree, EngineAI, Padbot, Slamtec integrations live

• GEOD ($0.12, -4% 7d): Hyfix Spatial Intelligence’s $15M raise (same founder, US-made autonomous chips) reinforces the robotics chip vertical thesis as their burn rate from revenues generating continues despite slight sell off

• $DEUS TGE confirmed for May 27. Next major humanoid robotics token in the agentic stack

3. AI Agents Developments

Here is what happened on crypto AI & robotics related developments this week:

That’s a wrap for issue 175 of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. For a full list of disclosures, please refer to the Khala Research website.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of three parts:

1. Snippet Partner (SURF AI)

2. Crypto AI & Robotics Market Overview

3. Emerging Developments

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

SURF is crypto’s AI command hub. Onchain, social, market, and prediction data routed through a single interface, powered by 10+ orchestrated AI models

Surf Studio lets you type a natural language prompt and receive a live, deployed crypto web app. Here’s a Bittensor (TAO) dashboard I created:

Portfolio trackers, protocol comparison dashboards, whale wallet monitors, token analytics with composite views (price + sentiment + news + onchain), VC due diligence reports. Anything you can think of, prompted and deployed

The SURF Studio AI capital flows dashboard used in this newsletter was built on it

For builders and agents: the Surf Agent Stack (SAS) ships 90+ API endpoints across seven data categories (Social, Market, Project, Trading, Wallet, Token, Onchain), packaged as a single MCP skill. Your AI agent gets crypto intelligence without integrating seven different data providers

Surf raised $15M in December 2025 and has reached millions in ARR, generated over 1M research reports, and is used by 80% of top exchanges and research firms. The platform processes structured data from crypto X feeds and on-chain sources to deliver actionable insights rather than raw noise

We published the definitive Khala Research report on SURF last week: SURF: A Data Platform for Crypto Intelligence. The report covers product analysis, capital efficiency, and the competitive landscape for crypto’s data infrastructure layer:

If you are evaluating data platforms for agent-native workflows, this is the resource

Learn more: SURF | SURF Agent Stack | Khala Research SURF Report

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

1. Crypto AI & Robotics Market Overview

AI Roundup

Three themes from the data this week:

1. Crypto AI just ran its cleanest TGE window in months: OPG and CHIP both launched April 21, both with Tier 1 backing, structured vesting, and live product metrics. Both held up reasonably well driving optimism for upcoming tokens

2. Subnet revenue is separating winners from emissions farmers: Lium prints +10% 7d while other top-10 subnet bleeds. Real GPU rental revenue is the catalyst

3. x402 activity broke out: Daily transactions hit 151K on April 24, more than 2x the prior baseline. “Agentic.market” launching April 20 is the catalyst, with Data as a Service the leading category

Crypto AI’s cleanest TGE window in months

OPG and CHIP both went live on April 21:

1. OpenGradient (OPG): $9.5M from a16z + Coinbase Ventures. 19% circulating at TGE, 12-month investor cliff, MiCAR cleared pre-launch. 2M+ verifiable inferences

2. USD.AI (CHIP): Eight-exchange simultaneous debut. $1.5B+ disclosed pipeline, $500M Sharon AI + $500M QumulusAI facilities, OBEX inaugural cohort. 80% locked, 7-8x volume-to-mcap

There’s a bigger narrative at play here and it leans into a return to fundamentals

Having tokens backed by tangible revenue generating products (in this case GPU financing) will bode well for the token long term, assuming that revenue flows back

The broader narrative we will likely see over the next 12 months is tokenized equities or tokens tied to equity through SPVs:

One to watch in the coming month will be XMAQUINA who have their TGE in a month for the DEUS token:

We will be publishing our report in the coming weeks on the “Robotics capital markets”, whereby individual pricate robotics equities (Apptronik, Figure etc) can be traded onchain, backed by SPVs

A) DeAI Market Cap Analysis

• LINK ($9.24, -0.9% 7d): BridgeTower adopted the full Chainlink stack on April 23 to tokenize $11B in DOM X Arizona Copper-Gold securities, with a $25B+ pipeline of natural resources behind it. Chainlink also earned Deloitte SOC 2 Type 2 certification on April 21, becoming the only oracle platform to hold SOC 2 Type 2, Type 1, and ISO 27001 simultaneously

• TAO ($247.15, +0.5% 7d): Quiet recovery week post-Covenant. 70% of supply staked. BIT-0011 (Conviction Mechanism) vote pending, Teutonic 1T training run targeting mid-to-late May

• NEAR ($1.35, -0.6% 7d): Near doubles down on ironClaw as the agent orchestration harness becomes their focus:

• ICP ($2.41, -2.3% 7d): Worst large-cap performer this week. The fundamentals are improving (Caffeine v3.0 AI upgrade in April, 280K canisters deployed, Mission 70 cutting inflation from 9.72% to 2.92% by year-end) but the chart is trading near the all-time low. ICP Cloud engines are coming:

• RENDER ($1.77, -1.1% 7d): RNP-023 passed at RenderCon 2026 (April 16-17), integrating Salad Network as an exclusive subnet and adding ~60K GPUs to the network. Burn-and-mint equilibrium routes payments through RENDER burns

• FET ($0.2027, -3.9% 7d): Worst week in DeAI by 7d performance. April 28 token unlock releases 2.66M FET ($569K) for AGIX migration. ASI:Chain TestNet still pending, mainnet pushed to late 2026/early 2027

• VIRTUAL ($0.6877, flat 7d): Co-authored ERC-8183 with Ethereum Foundation in March (open standard for cross-chain agent commerce). Q1 2026 numbers: $400M+ agentic GDP, $60M+ protocol revenue (~$300M annualized), 17K+ deployed agents + a number of virtuals eco updates this week:

• VVV ($8.87, -5.7% 7d): Sharp reversal but the structural story is intact. Venice AI crossed 2M users on April 21. April burn rate (16.6K tokens, $123K) already exceeded all of March. 69% of circulating supply now staked, sVVV committed to DIEM minting and third party protocols being built around DIEM token for yield on inference credits:

Bittensor Subnet Ecosystem

• Targon (SN4) ($14.57, +2.8% 7d): One of two top-10 subnets up. Manifold’s $10.5M Series A and Dippy’s 4M users back the verifiable compute thesis

• Lium (SN51) ($14.36, +9.6% 7d): Standout of the week

Lium and the subnet revenue divergence

Lium (SN51) is the only large Bittensor subnet up materially on the week. The +9.6% 7d is driven by revenue generation

• ~$432K/month in GPU rental fees, the highest of any subnet

• ~6.2% of TAO emissions

• 60% of daily miner emissions burned

• Multi-chain payments plus fiat via Coinbase

Post-Covenant, Lium’s positioning matters more. The network is being forced to demonstrate that subnets with genuine demand can carry weight regardless of operator drama

Rotation toward revenue-generating compute and storage looks real - Hippius (S3-compatible storage, ~$4.48M PnL) and Targon are the obvious peers with several others showing signs of commercial viability

B) Robotics Market Cap Analysis

• GEOD the standout +6.8% 24h ($55.5M mcap), reinforcing that revenue-generating DePIN survives soft tape. Hyfix's $15M raise (same founder as GEODNET) is an adjacent positive

• peaq also announces the support for zero human companiesleading to a spike in price mid week:

2. x402 Analysis

x402 transaction volume broke out this week. Daily transactions hit 151.3K on April 24, more than 2x the ~75K baseline through early-to-mid April

Data as a Service surged to 76.1K in a single day from negligible levels prior; Agent to Agent Services held meaningful share through the spike. The catalyst is Coinbase’s launch of Agentic.market on April 20

Three concrete updates from the agent marketplace this week:

1. Scale: x402 now sits at 165M+ transactions, ~$50M cumulative volume, and 480K+ transacting agents. Agentic.market is positioned as the discovery layer above this

   This replaces the API-key-and-credentials model with permissionless service access for both humans and agents

2. Initial provider list: Seven categories went live (Inference, Data, Media, Search, Social, Infrastructure, Trading) with OpenAI, Bloomberg, CoinGecko, AWS Lambda, LinkedIn, Alchemy, and Venice among the launch participants. Multi-step agent workflows (data → reasoning → trade execution → logging) now have a single discovery surface

3. Pricing model: Most listings use usage-based pricing, with some providers adding an “agentic premium” for automated bot access. Bazaar’s CDP Facilitator auto-indexes any x402 endpoint with the discovery extension enabled, no separate registration required

The Artemis chart is the cleanest evidence the discovery layer is doing what it was meant to do:

• Data as a Service spiking ahead of Agent to Agent suggests agents are pulling structured data feeds into workflows before transacting with each other; the sequencing makes sense

• ACP still runs 99%+ of x402 volume in some weeks, and Agentic.market widens the funnel above it rather than displacing it. The 30-day baseline from April 20 is the read worth tracking

3. AI Agents Developments: Innovations and Market Developments

Here is what happened on crypto AI & robotics related developments this week:

That’s a wrap for issue 174 of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. For a full list of disclosures, please refer to the Khala Research website.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of three parts:

1. Snippet Partner (SURF AI)

2. Crypto AI & Robotics Market Overview

3. Emerging Developments

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

SURF is crypto’s AI command hub. Onchain, social, market, and prediction data routed through a single interface, powered by 10+ orchestrated AI models

Surf Studio lets you type a natural language prompt and receive a live, deployed crypto web app. Here’s a Bittensor (TAO) dashboard I created:

Portfolio trackers, protocol comparison dashboards, whale wallet monitors, token analytics with composite views (price + sentiment + news + onchain), VC due diligence reports. Anything you can think of, prompted and deployed

The SURF Studio AI capital flows dashboard used in this newsletter was built on it

For builders and agents: the Surf Agent Stack (SAS) ships 90+ API endpoints across seven data categories (Social, Market, Project, Trading, Wallet, Token, Onchain), packaged as a single MCP skill. Your AI agent gets crypto intelligence without integrating seven different data providers

Surf raised $15M in December 2025 and has reached millions in ARR, generated over 1M research reports, and is used by 80% of top exchanges and research firms. The platform processes structured data from crypto X feeds and on-chain sources to deliver actionable insights rather than raw noise

We published the definitive Khala Research report on SURF last week: SURF: A Data Platform for Crypto Intelligence. The report covers product analysis, capital efficiency, and the competitive landscape for crypto’s data infrastructure layer:

If you are evaluating data platforms for agent-native workflows, this is the resource

Learn more: SURF | SURF Agent Stack | Khala Research SURF Report

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

1. Crypto AI & Robotics Market Overview

The Kelp/Aave Exploit: $293M Stolen, $6.6B Withdrawn

• Attackers exploited Kelp DAO’s cross-chain bridge, minted 116,500 unbacked rsETH (~$292M), deposited as collateral on Aave V3, borrowed ~$196M in WETH. Kelp froze contracts 46 minutes later. WETH was gone

• Aave TVL plunged from $26.4B to ~$20B. AAVE dropped 18%

• Aave’s contracts were not compromised; the exploit entered through external collateral that risk parameters accepted

• Now the largest DeFi exploit of 2026, overtaking the $285M Drift exploit from April 1

Comical, but very useful commentary:

Layer Zero appeared to be the main protocol exploited:

Why AI agents should care: agents trust on-chain state. rsETH appeared fully backed when it was not. No agentic payment protocol (x402, MPP, ACP) validates collateral integrity

• A perfectly coded agent still loses capital if the collateral is fraudulent

• April 2026: $575M+ in exploits

• Agents need collateral verification tooling, not just payment settlement

Vercel + Lovable breaches:

1. Vercel breach traced to compromised AI tool Context.ai via Google Workspace; many Web3 teams host wallet interfaces on Vercel

2. Separately, vibe-coding platform Lovable reportedly exposed users’ AI chat histories, source code, and database credentials after a backend permissions change accidentally re-enabled access to public project chats

Two AI-adjacent security incidents in one week. As AI tooling embeds deeper into crypto dev workflows, supply chain attacks through these integration points grow

The common theme: as AI capabilities ramp up, so will the exploits in both crypto and AI around financial collateral and private data

Bittensor: The New Conviction Mechanism

• Const announced the “Conviction Mechanism” requiring long-term token locking

• Addresses gaps I identified last week (lock-based ownership, vesting, exit penalties)

• TAO at $247, down ~26% on the week

• Subnet ecosystem below $1B

• Right direction, but one dimension of a multi-dimensional problem

• IP retention, on-chain revenue sharing, and portable subnet infrastructure are equally critical

• TAO Institute framework will contribute directly to evaluating these gaps

Pitfalls like the Covenant debacle illustrate the need for tooling that helps investors form decisions around associated subnet risks

The Subnet Risk Index (SRI) provides that lens, available here to try out: Link

A) DeAI Market Cap Analysis

Post-Covenant consolidation. TAO stabilized near lows; forced selling appears exhausted:

• TAO (-2.1% 7d, $247.11, $2.37B): Conviction Mechanism hasn’t catalyzed recovery but has stopped the bleed

• NEAR (-3.2% 7d, $1.38, $1.78B): Giving back last week’s +16.4% gains

• RENDER (-5.5% 7d, $1.80, $933M): Weakest large cap this week

• FET (-10.5% 7d, $0.2131, $481M): ASI Alliance under pressure. Worst performer among top AI tokens

• VIRTUAL (+1.9% 7d, $0.6924, $454M): Eastworld and ACP activity providing a floor; now hit $4m in A2A activity on x402

• VVV (+5.1% 7d, $9.45, $434M): Week’s standout. Privacy narrative gaining traction and began adding TAO subnets like Targon’s Dolphin AI. They’re also entertaining DIEM secondary compute markets:

Bittensor Subnet Ecosystem

The subnet ecosystem is showing signs of life post-Covenant:

1. Community miners restored SN3, SN39, and SN81 entirely from open-source code with no central operator

2. SN3 is now rebranded as Teutonic and targeting a 1-trillion-parameter decentralized training run in May, timed with the Grayscale ETF’s peak SEC review window

3. If they pull it off, the narrative shifts from “Bittensor’s biggest builder left” to “the network survived and scaled 14x beyond what Covenant achieved”

Broader Subnet analysis:

• SN64 / Chutes (-8.8% 7d, $20.54, $99.6M): Still the dominant subnet but the economics deserve scrutiny. 9.1T+ tokens processed, 50B+ daily peaks, 400K+ users. Revenue-funded auto-staking buyback is live. But Pine Analytics flagged a 22:1 subsidy-to-revenue ratio ($52M in annual emissions vs $2.4M in external revenue). The question for Chutes is whether inference volume converts to sustainable revenue before emissions compression forces the issue

• SN4 / Targon (+7.5% 7d, $14.33, $65.2M): Recovering. Intel TDX whitepaper co-authorship and $10.4M ARR give it the best subsidy-to-revenue ratio in the ecosystem (1.7:1)

• SN44 / Score (+21.0% 7d, $9.44, $42.2M): Week’s top performer. PwC France partnership for AI-driven sports analytics. Disrupting the $600B soccer industry with 70% prediction accuracy using DKING AI agents

• SN75 / Hippius (+9.8% 7d, $6.67, $26.8M): Decentralized data storage gaining traction; created by the same person (Mogmachine) that created TAOStats

• SN3 / Teutonic (-17.1% 7d, $6.08, $26.4M): Still bleeding from the Covenant exit but the trillion-parameter training target in May is the catalyst to watch

B) Robotics Market Cap Analysis

• GEOD (-10.4% 7d, $0.1237, $53M): Down despite the Hyfix raise (below). DePIN sector rotation and macro risk-off overriding the fundamental catalyst

• PEAQ (+22.6% 7d, $0.01647, $34M): High-beta DePIN play amplifying renewed Machine Economy interest post-Hyfix. Hashkey Capital flagged PEAQ as a foundational L1 for Web3 robotics. Dubai VARA MoU adds institutional credibility. Note: 84.84M unlock on May 12 (~1.9% of supply)

• AUKI (+12.3% 7d, $0.006802, $28M): Executing. 1,000+ retail store pilots in Northern Europe, expanding into a $15B-revenue retailer (~2,000 locations). GAIB partnership for robot financing. Live teleoperation for planogram verification. 7-digit ARR

• NATIX (-21.3% 7d, $0.0000727, $2.9M): Vesting bleed. ~45% of supply entering circulation (team + early backer unlocks since Feb 2026) against $386K daily volume. DePIN mapping vertical lost narrative momentum to AI agent infrastructure

Hyfix Raises $15M: GEODNET’s Hardware Play Gets Serious

Hyfix Spatial Intelligence raised a $15M seed (Craft Ventures, Multicoin, Catapult) to build a US-made autonomous systems chip. CEO Mike Horton is also the creator of GEODNET

The chip combines flight control, GNSS positioning, encrypted comms, and onboard AI inference into a single SoC. Integrates with GEODNET’s 21,000+ reference stations for centimeter-level positioning when GPS is jammed or spoofed. Supports ROS 2, ArduPilot, PX4

First time a DePIN network has a direct hardware integration path into a venture-backed chip company targeting US defense and commercial drone markets

Craft Ventures’ Fluhr sees the architecture extending into humanoid robotics within two years. Validates GEODNET as enterprise-grade infrastructure beyond DePIN token rewards

2. AI Token 7-Day Analysis

AI Tokens Capital Flows ($21.05B mcap, $1.99B 24h vol)

SURF: 7D relative performance: all six large caps within -6% to +6%. Sector digesting Covenant + Kelp and finding equilibrium

For specific token analysis, see commentary in 1A above.

x402: The Collateral Gap

Kelp reframes the x402 conversation: the entire agentic payments stack assumes on-chain state is sound. Kelp proves it is not

Agents need tooling that assesses collateral quality and protocol health before committing capital

Khala Research x402 Report

Pieverse: x402 Adjacent or Upbit Pump?

Pieverse (+136.1% 7d, $483M 24h vol, $233M mcap) is the anomaly on both CoinGecko and SURF Studio this week

What it is: “The Agentic Neobank.” x402-based rails for gasless, auditable agent-to-agent payments via Universal Commerce Protocol (UCP)

Compliance-first positioning (timestamped invoicing, jurisdiction-aware payments) differentiates it from Bankr, Virtuals ACP, Daydreams

What happened: Upbit (~80% of Korean exchange volume) listed PIEVERSE on April 20; Surged from $0.48 to ATH $1.65, crashed 49% to $0.83 same day

Caution: 3x volume-to-mcap ratio. Team tokens (20%) on 1-year cliff with 3-year vest, meaning unlocks are coming. Textbook exchange listing pump pattern

3. AI Agents Developments: Innovations and Market Developments

Here is what happened on crypto AI & robotics related developments this week:

That’s a wrap for issue 173 of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content of this newsletter is for informational purposes only. Nothing in this newsletter constitutes financial advice or a recommendation to buy or sell any asset. Always do your own research before making any investment decisions.

I hold positions in many of the assets discussed in this newsletter. For a full list of disclosures, please refer to the Khala Research website.

Follow me on X | Khala Research

This Crypto AI & Robotics newsletter consists of three parts:

1. Snippet Partner (Nunet)

2. Crypto AI & Robotics Market Overview

3. Emerging Developments

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’

NuNet is a peer-to-peer protocol for discovering, orchestrating, and settling compute across distributed infrastructure

Think of it as the “Limewire for Compute”

Where most projects in the decentralized compute space are GPU marketplaces or cloud wrappers, NuNet built a complete orchestration and coordination layer from first principles over four and a half years of R&D

On March 2, 2026, NuNet launched Network Live: a production decentralized compute network where software finds compute autonomously across any device from edge to cloud, every machine carries a cryptographic identity verified at every interaction, and orchestration and value exchange are unified in a single system

This is live infrastructure with real AI workloads running on it today:

What has been shipped with Network Live:

1. Multi-chain settlement: NTX transactions flow natively on both Ethereum and Cardano. Compute providers select their preferred chain when entering contracts. No bridges or wrapped tokens

2. Open provider access: The NuNet Appliance runs on a PC, dedicated server, Raspberry Pi, cloud instance, or data centre hardware. If a machine has compute to contribute, it can be onboarded

3. Real AI deployments already running: Private AI agents via OpenClaw with Qwen and Ollama (fully private, no cloud APIs), one-click LLM deployment with Hermes Agent on consumer GPUs (RTX 3060 running Google’s Gemma 4 26B), workflow automation via n8n, and ASI:Chain validator node deployment

NuNet has active partnerships across six verticals:

1. smart infrastructure (paid innovation contract with AL’MA Action Logement covering 1M+ homes),

2. edge AI and DePIN (Auki Labs/Posemesh PoC delivered),

3. robotics (Intercognitive Foundation founding member),

4. private agentic AI,

5. Web3 compute (10 Cardano SPO partnerships), and

6. sovereign compute.

NuNet is a member of the Linux Foundation. NTX is the utility token facilitating peer-to-peer compute settlement on the live network.

Learn more: NuNet Docs | Live Dashboard | Download the Appliance

This newsletter goes out weekly to 7.2k+ subscribers.

Please don’t hesitate to message me directly for sponsorship or partnership enquiries.

1. Crypto AI & Robotics Market Overview

The Covenant AI Exit: What Happened, What It Means

This is the section I would normally save for Emerging Developments, but the scale of this event warrants leading with it

On April 10, Covenant AI founder Sam Dare published a public statement announcing the immediate withdrawal of all three Covenant subnets from Bittensor: Templar (SN3), Basilica (SN39), and Grail (SN81).

The same team that trained Covenant-72B, the 72-billion-parameter model across 70+ contributors on commodity hardware that triggered TAO’s 90% March rally, Jensen Huang’s public endorsement, and the Grayscale ETF filing momentum.

Dare accused Bittensor co-founder Jacob Steeves (Const) of maintaining centralized control over what is marketed as a decentralized network

The specific allegations include:

• suspension of emissions to Covenant’s subnets,

• stripping moderation access over Covenant’s own community channels,

• unilateral deprecation of subnet infrastructure, and;

• strategically timed token sales used as economic pressure

But Const has responded to all allegations and highlighted that this will strengthen the network in the long term:

Dare simultaneously sold 37,000 TAO (approximately $11M) across the three subnets, intensifying the selloff.

The Tao Papers

On the same day, a whistleblower site called Tao Papers went live, publishing internal documents, chat transcripts, and on-chain forensics from multiple sources inside the Bittensor ecosystem.

The most damaging claim: of 41 Bittensor network upgrades between 2023 and 2026, 38 were proposed, first-signed, and deployed from infrastructure controlled by Steeves, with the other two multisig signers co-signing within minutes and without public discussion

This directly challenges the “triumvirate” governance narrative Bittensor has promoted to institutional allocators

Steeves’ Response

• Steeves responded publicly, denying the ability to suspend subnet emissions and stating that his token sales amounted to less than 1% of his total investment in Covenant-related subnets

• On Discord moderation, he claimed Dare himself deprecated his own channels and that moderation access was temporarily restricted because Dare was removing genuine criticism, then restored

• On infrastructure deprecation, Steeves stated he was unsure what the allegation referred to

My Take: The Bittensor ecosystem lost a team with misaligned interests.

Covenant trained a 72B parameter model using Bittensor emissions and network compute, built the narrative that repriced TAO from $180 to $380, and then walked away with 8 figures in capital

The governance allegations are a sideshow to the structural problem: there is limited in the protocol today that prevents this from happening again

The biggest friction point to institutional adoption of the Bittensor subnet ecosystem is the misalignment between subnet founders and network stakeholders

Very few institutions will deploy serious capital into subnet alpha tokens until a tangible solution is in place

A founding team can farm emissions, build on network resources, extract value, and leave. The 37,000 TAO dump ($11M) is the cost the ecosystem paid for learning this lesson in production

TAO crashed 27% on the news, dropping from above $340 to $253. Over $9M in long positions were liquidated. There has been no meaningful recovery. As of writing, TAO sits at $253, still pinned near the lows

The Potential Solutions

I published a thread with 10 suggestions for how the protocol could address this:

TAO Institute Goes Live This Week

This brings me to TAO Institute. We are launching the Subnet Risk Index (SRI) this week; the first institutional-grade risk evaluation framework for Bittensor subnets

Pitfalls like the Covenant debacle illustrate the need for tooling that helps investors form decisions around associated subnet risks.

The SRI provides that lens. More details on launch day.

A) DeAI Market Cap Analysis

The overall DeAI market cap sits at $21.6B (+3.4% 24h):

Bittensor is trending, though this time for the wrong reasons.

The broader AI category is showing resilience with a 3.4% 24h uptick, but the 7d picture tells a more nuanced story.

• TAO (-18.6% 7d, $252.77, $2.43B mcap): The Covenant AI exit and simultaneous 37,000 TAO dump extracted roughly $10M from the ecosystem and cratered the price from $340+ to $253. No meaningful bounce. Over $9M in long liquidations. Covered in depth above.

• The Grayscale ETF filing (GTAO on NYSE Arca) and 43% allocation within Grayscale’s DeAI Fund remain live catalysts, but governance risk now directly competes with the institutional narrative

• NEAR (+16.4% 7d, $1.44, $1.86B mcap): The week’s best performer among large caps. Strong momentum with $238M in 24h volume suggesting real capital rotation rather than a low-liquidity drift; could this be AI capital rotating from TAO?

• RENDER (+1.2% 7d, $1.90, $984.71M mcap): Holding steady. RenderCon 2026 momentum continues but the token is treading water relative to the broader sector’s bounce

• VIRTUAL (+8.2% 7d, $0.6802, $446.10M mcap): Bouncing back after last week’s dip. Eastworld robotics positioning and ACP facilitator activity providing a bid. The x402 usage-based pricing shift from Coinbase could re-ignite interest given Virtuals ACP is one of the largest x402 facilitators

Bittensor Subnet Ecosystem

The Bittensor Subnets market cap has decreased to $1.03B (-1.8% 24h):

The subnet ecosystem has taken a significant hit. Total market cap down from $1.36B last week to $1.03B, a roughly 24% decline.

TLDR: The Covenant exit triggered a sector-wide repricing across the subnet ecosystem. The question now is whether this stabilizes as an isolated governance event or whether other subnet operators begin to echo Dare’s concerns.

The contagion effect has pulled essentially every subnet into the red on the week:

• SN64 / Chutes remains the dominant subnet at $108.88M mcap ($22.70, -16.1% 7d). Down from $129M last week as the broader TAO selloff dragged it lower. OpenRouter default routing and approaching $10M ARR remain strong fundamentals + hints at training a trillion parameter model next week:

• SN4 / Targon at $59.96M ($13.34, -31.0% 7d). The hardest hit major subnet, giving back the Intel partnership pump from two weeks ago

B) Robotics Market Cap Analysis

The Robotics market cap has increased to $674M (+4.4% 24h):

A recovery week for robotics after last week’s broad selloff. The category is up 4.4% and showing green across most assets, with the notable rally from Fabric Protocol

• VIRTUAL (+8.2% 7d, $0.68, $446.46M mcap): Leading the robotics category recovery. Eastworld robotics positioning and ACP facilitator activity providing sustained interest

• GEOD / Geodnet (+1.6% 7d, $0.1382, $59.20M mcap): Slight green on the week. Continues executing with 20,000+ stations across 150+ countries

• ROBO / Fabric Protocol (+11.1% 7d, $0.01932, $43.04M mcap): Bouncing back after last week’s brutal -26.6% decline. Still down from recent highs but showing signs of a relief rally

• IOTX / IoTeX (-2.2% 7d, $0.004476, $42.27M mcap): Slight red on the week

• PEAQ (+3.5% 7d, $0.01343, $27.40M mcap): Recovering from last week’s -9.2% decline

• AUKI (+1.9% 7d, $0.006074, $24.56M mcap): Also recovering after last week’s -19.8% drawdown

2. AI Token 7-Day Analysis

The 7D relative performance chart tells the story of the week in a single image. From April 7 through April 9, the top six AI large caps (LINK, TAO, NEAR, ICP, RENDER, FET) were clustered between 0% and +10%, moving in broad correlation

Then April 10 hits and TAO falls off a cliff, dropping to -20% while every other asset holds its range or continues higher. NEAR pushes to +10%, LINK and ICP remain flat-to-positive, RENDER and FET hold steady.

The divergence is entirely Covenant-driven; TAO decoupled from its peer group on a single catalyst: Sam Dare’s public exit statement, the 37,000 TAO dump, and the Tao Papers whistleblower site launching simultaneously. The rest of the AI sector barely registered the event

That isolation is both reassuring for the broader category (no contagion) and damning for TAO specifically (the market is treating this as a Bittensor-specific governance problem, not a DeAI narrative risk)

x402: The “Upto” Scheme and Why It Matters

The most significant x402 development this week: Coinbase Developer Platform shipped the Upto scheme, a new billing primitive that allows variable-cost settlement within a single HTTP request cycle.

Until now, x402 only supported exact, fixed-price payments

A client knew the price before making a request, paid it, and received the response. This works well for deterministic API calls with predictable costs, but it blocked an entire category of services where the cost depends on usage:

• LLM inference (token count varies per prompt)

• compute tasks (processing time varies by complexity), and;

• data queries (result size unknown upfront)

Upto solves this by letting the client authorize a maximum spend per request

The server processes the work, determines the actual cost, and settles only what was consumed. Unused funds remain in the client’s wallet

The diagram in the screenshot illustrates this cleanly: client authorizes max, server settles exact use ($0.42 in this example), funds remain in wallet

Competition is Ramping Up (MPP vs x402)

This is a direct response to the scaling critique that Stripe’s Machine Payments Protocol (MPP) posed to x402

MPP, co-authored by Stripe and Tempo and launched on March 18, introduced “sessions” as its core primitive

An agent pre-authorizes a spending limit once, then streams micropayments against that session continuously using lightweight off-chain vouchers, settling in bulk rather than per-request on-chain

For high-frequency scenarios (an agent querying a data feed thousands of times per hour), MPP’s session model avoids the gas overhead of per-request on-chain settlement that was a real constraint in x402’s original fixed-price design.

The Comparison:

Upto does not replicate MPP sessions. It remains a per-request model. But it closes the gap meaningfully for the most common agentic workload pattern today: one prompt, one response, variable cost

For LLM inference, where each call generates a different number of tokens and the cost is only known after completion, Upto removes the need to either overpay via a padded fixed price or open a full MPP session for a single request.

Our definitive institutional breakdown of x402 remains the most comprehensive resource available: Khala Research x402 Report

3. AI Agents Developments: Innovations and Market Developments

Here is what happened on crypto AI & robotics related developments this week:

That’s a wrap for issue 172 of Sammy’s Snippets. I hope you enjoyed it.

Please leave me any questions or thoughts here - I will respond to everyone!

If you found this interesting, please consider subscribing to this Substack and following me on X for more related insights.

If you are interested in more formal reporting on Crypto AI and Robotics then Khala is my research product.

Disclaimer: The content covered in this newsletter is not to be considered investment or financial advice. It is for informational and educational purposes only.

Disclosure: I hold some of the assets and have partnerships with some of the projects mentioned in this newsletter.

This Crypto AI & Robotics newsletter consists of three parts:

1. Snippet Partner (CodecFlow)

2. Crypto AI & Robotics Market Overview

3. Emerging Developments

If you have any questions feel free to reach out to me on X or message my business X account ‘Khala Research’