Web3

AI Agents Have Made 138M Crypto Payments, Most Under 10 Cents

Chainalysis counts payments over x402 rising from about 1,500 in mid-2025 to more than 138 million by the third quarter.

⏱ 2 min read Web3
Quick Summary
  • More than 87% of the payments were between a hundredth of a cent and 10 cents each
  • US-linked wallets made 88.1% of identifiable payments for AI model access, China-linked ones 0.1%
  • Earlier TRM research suggested much apparent agent activity may be scripts or self-dealing

Autonomous AI agents have made more than 138 million crypto payments through x402, a payment protocol incubated by Coinbase, according to a new Chainalysis report.

A year ago the number was tiny. In mid-2025, Chainalysis counted roughly 1,500 such payments. By the third quarter of 2026, the cumulative total had passed 138 million. The payments run on Base, Coinbase’s layer 2 network.

What agents are buying

x402 lets software pay for online services automatically, without a person approving each purchase. Agents are using it to buy computing power, live web data and coding tools.

The payments are very small. More than 87% were between a hundredth of a cent and 10 cents each. That is the kind of payment card networks struggle to handle, and one reason stablecoins are used for this kind of machine-to-machine spending.

Where the money comes from

The activity is heavily American. US-linked wallets made 88.1% of the geographically identifiable payments for access to AI models, compared with 0.1% for wallets linked to China. Chinese-linked wallets made up most of the spending on security and code-analysis services instead.

Big names moving in

Payment companies are building for it. Visa, Mastercard and Stripe have moved to support payments by AI agents, Google has launched its own agent-payments protocol, and Coinbase now lets business customers accept agent payments.

BlackRock is watching too. Will Su, its head of digital assets research, told Chainalysis that AI agents could create “meaningful demand for frequent, low-value payments around the clock.”

The caveats

The numbers need care. Chainalysis counts only identifiable x402 activity, not the whole market. And earlier research by rival tracing firm TRM suggested that much apparent agent activity may be scripts or self-dealing rather than genuine autonomous agents. A raw count of transactions can overstate how much real commerce is taking place.

⚖️ Our Verdict ⚖️ Watch and Wait

The growth in payment count is real and fast, and big payment firms are building for it. But the payments are tiny, the count may include scripted activity, and the report does not show how much money is actually moving.