Welcome to #91 of the AI edge.
BTC ripped past $75K early in the week after Trump declared Iran "obliterated." The relief rally lasted two days. Escalation resumed, PPI came in hot, the Fed held steady, and BTC bled back to $69K.
Kalshi is now pricing a 75% chance it falls below $60K. That's not noise.
While BTC chopped, TAO took over the CT timeline. Jumped ~35% on the week, touching $300, off the back of Grayscale's SEC reporting status, Templar's record-breaking decentralized training run, and a broader rotation into AI tokens. Every other post is about which subnet breaks out next.
The other headline that caught attention was Jensen at GTC. A $1 trillion chip revenue forecast through 2027, the full Vera Rubin reveal, and the launch of NemoClaw, an open-source agent platform for enterprises.
Hardware dominance wasn't enough…is Nvidia now reaching into the software stack too?
We’ve got a great edition this week. Let’s dive in.
The Big Story: Stripe Ships the Agent Stack
Stripe shipped its own blockchain this week. Tempo, built with Paradigm, went live on mainnet alongside something potentially bigger: the Machine Payments Protocol.
MPP is an open standard that lets AI agents pay for services on their own. An agent requests a resource, the service responds with a price, the agent authorizes payment from its wallet, settlement happens instantly, and the resource is delivered.
No human in the loop. Stablecoins, cards, and even Bitcoin Lightning are all supported out of the box, with Visa, Lightspark, and Stripe itself extending the protocol across rails.
The Wedge
This isn't a crypto team trying to get Stripe's attention. This is Stripe building the rails itself. Over 100 services are already listed in Tempo's payments directory at launch, including Alchemy, Dune, Anthropic, and OpenAI. For any business already running on Stripe, accepting agent payments is a few lines of code. That integration depth is a moat no crypto-native competitor can match overnight.
The "sessions" primitive is the design choice worth watching. Instead of paying per request (which creates massive on-chain spam at agent scale), an agent authenticates once, sets a spending limit, and streams payments within that window. Think OAuth for money. That's what makes microtransactions actually viable for autonomous workflows.
The Fine Print
The obvious question around any Stripe-owned chain is how "open" it stays long term. MPP is open-source and designed to be rail-agnostic, but Tempo is the default settlement layer, and Stripe controls the most frictionless integration path. Decentralization is a promise only.
Coinbase got there first with x402, its own standard for agent payments on Base. It's processed hundreds of millions of transactions since launch, which shows real traction. But daily volume has cooled off.
MPP is coming from a very different place. Stripe moves trillions of dollars a year and already works with millions of merchants. I think that is the real split: Coinbase built the crypto-native system first, while Stripe built the version that can slide into the economy that already exists. The winner will probably be the one merchants decide to back.
GEODNET hit $8.3M ARR with 21,727 active stations across 158 countries, reinforcing its RTK network as a global “precision layer” for drones/robots and physical AI.
peaq released its “Robot Money” Purple Paper, laying out how machines could transact and run businesses across blockchains as a true machine-to-machine economy.
Score (Bittensor SN44) received new funding from Stillcore Capital, adding fresh capital to accelerate the subnet’s roadmap and ecosystem expansion.
NexusLabs launched mainnet with zkVM 3.0, following a massive testnet (2.6M participants + 96 partners) and pushing zk compute closer to production use.
RESI Labs (Bittensor SN46) shipped its US property appraisal model on Chutes, running encrypted inference via SN64 and boosting distribution + miner rewards through “subnet-on-subnet” demand.
Sky (prev. MakerDAO) proposed ~70M USDS in Genesis Capital allocations to seed more executor agents, scaling its agent network of capital allocators ahead of the March 26 executive vote.
Chutes (Bittensor SN64) became OpenRouter’s top inference provider, claiming ~85% cost savings vs AWS for comparable inference workloads.
Sahara AI introduced Proof of Guardrails, aiming to cryptographically verify agent safety/compliance instead of relying on unverifiable “trust me” guardrail claims.
Read our deep dive on Gradient and how it can make reinforcement learning much cheaper and more accessible to all.
AI Infra is booming
So far in 2026, AI infra/semis are the top-performing theme (+18%), leading a market where almost everything else is red. There’s no better time to be paying attention to the AI stack.
Every protocol and startup I write about in this newsletter depends on GPUs, memory, networking, power. This is where the pressure is building, and where a lot of the money is getting made. Micron is up 30% this year because HBM demand broke past supply. Moves like that leave clues early. They’re just scattered across the stack and hard to track.
That’s why I built Tessara - a new product by Chain of Thought.
Tessara helps you track bottlenecks, pricing power, and where value is moving across the AI stack before the market catches up.
I’m opening a small private beta now. Reply “beta” and I’ll put you on the list.
🔥 Our Weekly Top Tweets
#1 Simulating the Market’s Mind
A Chinese college student built MiroFish, an AI system that tries to predict markets by simulating them, spawning up to 1M agent traders to “play out” future events. It earned ~60K GitHub stars and $4M in backing in under six months.
#2 Always-On Trading Agents
Gigabrain introduced SuperAgents, always-on AI trading agents (not OpenClaw wrappers) running on a custom cloud runtime with persistent memory, wallets, and real-time market intelligence. At $5/month. Early days, but if it works, it could be huge.
Cheers,
Teng Yan & 0xAce
P.S. If you’re a startup doing serious work in AI or robotics and want to work closely with us, reply to this email and let’s chat.
