AI Agents Have Already Paid 138 Million Bills Themselves: US Wallets 88%, China Just 0.1%
You're still debating whether AI can transfer money for you. AI has already transferred 138 million payments on its own — and it's billing by the cent.
Three key facts
AI agents have made more than 138 million transactions through Coinbase's x402 protocol, using cryptocurrency to buy data, computing power, and access to AI models. The figures come from Chainalysis's report "Inside North America's $3 trillion crypto economy: Agentic commerce accelerates." The transactions occur on Coinbase's Base chain; x402 is a protocol that lets AI agents pay automatically with stablecoins, and payment giants including Visa, Mastercard, and Stripe have already backed it.
The geographic split is highly concentrated: US-linked wallets accounted for 88.1% of identified payments for AI model inference, versus just 0.1% for Chinese-linked wallets. Chainalysis head of research Eric Jardine offers a very practical explanation: Chinese developers are more inclined to use open-weight models such as DeepSeek and Qwen, so their need to buy metered inference services is inherently lower. In other words, they aren't failing to spend — they're spending on buying the model (weights) rather than renting one.
Individual amounts are startlingly small: more than 87% of analyzed x402 payments were worth 10 cents or less. The report says this underscores stablecoins' advantage for micropayments — traditional card fees simply can't cover a transaction worth a few cents. But the report also leaves a question mark: how much of this is "genuine commercial demand" versus testing and technical demos remains to be seen.
Wang Dou's Take
138 million sounds impressive, but stack that next to 87% being under 10 cents and the picture changes: it looks more like a bunch of machines passing pennies among themselves than a booming market. On average, if the vast majority are a few cents, the total volume of 138 million payments may not exceed one mid-size e-commerce company's monthly sales. The number is big because machines don't mind the hassle — not because the business is big.
What's really worth pondering is that 88% vs 0.1%. Many people's first reaction is "China is falling behind again," but Jardine's explanation is the opposite: not behind — a different path. If you have your own open-source model, you hardly need to rent someone else's inference on-chain, one payment at a time. That's a genuinely clever observation: when your model is free and open, you don't even need to participate much in the "AI agents pay automatically" business. The flourishing of paid infrastructure is, to some degree, a product of models not being free.
That final question mark is the most valuable part — "genuine commercial demand" is in doubt. Whenever a new concept takes off, idle numbers rush in first. The agent economy may be the future, but today this report proves "machines' ability to move money," not "the value of the money they move." Anyone who conflates the two will end up holding someone else's test invoice at the top.
Source: CoinDesk · Chainalysis