AI Agents Made 138 Million Crypto Payments in a Year. Most Were Tiny

What it claims

Chainalysis, in the North America edition of its annual Geography of Crypto report (framing a $3 trillion regional crypto economy), counts more than 138 million cumulative AI-agent payments through Coinbase's x402 protocol on Base by Q3 2026. That is up from roughly 1,500 in mid-2025. The page names what agents are buying: computing power, live web data and coding tools, paid for out of budgets that developers assign to the agents. More than 87% of payments fell between $0.0001 and $0.10.

Evidence

• Geography by service type differs sharply. US-linked wallets made 88.1% of geographically identifiable payments for AI model inference, and China-linked wallets made 0.1%. In security and code-analysis services the pattern reverses: Chinese-linked wallets account for most of the spending. The summary gives only the US inference figure. This split matters if you are deciding which x402-paid services have non-US demand.
• Adjacent infrastructure the page lists:
• Coinbase has a tool that lets agents trade and transact.
• Coinbase has begun letting business customers accept agent payments. This is the merchant side, which matters if you sell an API.
• Visa, Mastercard and Stripe have moved to support agent payments.
• Google has a dedicated agent-payments protocol.

The article gives no technical details for any of these.

• Demand-side signal: Will Su, BlackRock's head of digital assets research, told Chainalysis that BlackRock is watching for "meaningful demand for frequent, low-value payments around the clock."

Limits

• Coverage is partial. Chainalysis itself cautions that the figures capture only identifiable x402 activity, not the global market. The US/China gap should not be read as market share.
• Scope is x402 on Base only. Other chains and other agent-payment protocols, including Google's, are not counted.
• Counts are not revenue. The article reports transaction counts and size bands, but gives no total dollar volume and no per-service revenue.
• Many payments may not come from real agents. Earlier research by TRM, a rival tracing firm, suggests much of the apparent agent activity may be scripts or self-dealing. Raw counts may therefore overstate genuine agentic commerce.
• This is secondhand reporting. The page is a news article about the report, with no methodology. Before relying on any figure, check the Chainalysis report directly, in particular how it attributes wallets to geographies and how it classifies a payer as an agent.