Ignore the hype. Follow the gas.
Anthropic just dropped a Q2 revenue bomb: $115 billion. That’s 14x year-over-year. 2.4x quarter-over-quarter. And the kicker? 70% of that—roughly $80 billion—came from a single product: Claude Code. Not a chatbot. Not an API token sale. An autonomous agent that writes, deploys, and debugs code for enterprise clients.
The market is rewarding reliability over raw capability. Enterprise API usage accounts for 80-85% of Anthropic’s revenue. That means corporations are paying for auditable, repeatable, secure execution—not for a clever conversation.
Now ask yourself: where does that trustless execution layer live? In a walled garden? Or on a verifiable, decentralized ledger?
Context: The AI-Crypto Convergence Signal
I’ve been tracking this intersection since 2026, when my fund pivoted heavily into decentralized compute networks like Render and Akash. The thesis was simple: autonomous AI agents need trustless payment rails. No bank account, no legal entity—just a wallet and a smart contract.
Anthropic’s Q2 numbers validate that thesis from the enterprise side. Claude Code is effectively a “proxy economy” where a single agent handles millions of transactions per second. But it’s running on Amazon’s servers. The code is private. The execution is opaque.
Crypto’s answer? Decentralized physical infrastructure networks (DePIN), zero-knowledge proofs for verifiable computation, and machine-to-machine micropayment channels. The infrastructure layer is where the real value accrues—not in the model parameters, but in the pipes that carry the data and the proof that the data is correct.
Core: Deconstructing the Anthropic Liquidity Flows
Let’s break down the $115 billion.
- $80B from Claude Code (agentic code generation)
- $20B from enterprise API calls (non-agentic, but still B2B)
- $15B from consumer subscriptions and other sources
That’s a 7:1 ratio of agent-driven revenue to traditional API revenue. This is not a gradual shift. It’s a pivot. Enterprise spending has moved from “chat as a service” to “agent as a service.”
Now map this to crypto. The same pattern appears in DeFi: from speculative farming to real-world asset tokenization. The macro-liquidity is flowing from speculative narratives to productive infrastructure.
Claude Code’s success is a proxy for the demand for autonomous, auditable execution.
But here’s the catch: Claude Code operates inside a centralized cloud. Every line of code it generates passes through Amazon’s data centers. Every debugging session is logged on a private server. The enterprise pays for the output, but the process is opaque.
Crypto offers a better model: on-chain agent execution. Imagine an AI agent that executes a smart contract, pays for gas, and proves its work via a zk-SNARK—all without human intervention. The enterprise gets the same reliability, but with verifiable trust.
This is not a hypothetical. In Q2 2026, the decentralized compute network Akash processed over 10 million GPU hours for AI inference. Render’s network handled 2 million rendering jobs for architectural firms. The volume is still small compared to AWS, but the growth rate is identical to Anthropic’s early days.
Follow the gas, not the hype.
Contrarian: The Decoupling Thesis is Wrong
Every macro analyst I know is screaming that AI and crypto are decoupling. AI is real economy; crypto is speculative. The numbers suggest otherwise.
Anthropic’s $650 billion ARR is priced at 15x forward revenue. That’s a crypto-level multiple. The market is treating Anthropic the same way it treats a high-growth DeFi protocol: price on narrative, then adjust on execution.
But the contrarian angle is this: The AI boom is actually a hidden bull case for decentralized infrastructure.
Why? Because Claude Code’s success depends on trust. Enterprises trust that the code generated is secure, that the execution environment is not compromised, and that the data is private. In a centralized world, that trust is based on brand reputation and legal contracts. In a decentralized world, that trust is based on mathematics and open-source verification.
The moment an enterprise experiences a security breach from a Claude Code-generated vulnerability—and it will happen—the demand for on-chain verification will skyrocket.
Bets are cheap; exits are expensive.
I’ve seen this movie before. In 2020, DeFi Summer looked like a speculative bubble until the UST crash forced everyone to realize that liquidity is everything. The same pattern is playing out now. The AI agent market is a liquidity pool. The question is: who controls the exit?
Takeaway: Position for the Agent Infrastructure Cycle
Anthropic’s Q2 is a signal, not a destination. The revenue is real, but the infrastructure it runs on is fragile. The next cycle will be defined by protocols that enable verifiable, autonomous agent execution.
I’m already positioned: - Long decentralized compute networks (Render, Akash) - Long zero-knowledge proof verification layers (StarkNet, ZKsync) - Long micro-payment channels for machine-to-machine transactions (Lightning, Coil)
What about the model tokens? I’m short. The model is the commodity; the infrastructure is the scarce resource.
When Claude Code needs to pay for external data, will it call a bank or a smart contract?
Follow the gas. The answer is on-chain.