138 million payments, most under 10 cents: what x402 data says about AI agent commerce
New Chainalysis figures show software agents paying for inference and data in tiny stablecoin amounts. We read the numbers, the patterns and the caveats.
AI & CRYPTOSTABLECOINS & CBDCANALYSIS
Hana Kimura
10/28/20263 min read
For two years, agentic commerce has been mostly a slide in investor decks: software that buys what it needs without a human clicking “pay”. The first large dataset on how that actually works now exists, and it describes a market that is real, fast-growing and still very small in dollar terms. According to the North America chapter of Chainalysis' 2026 Geography of Cryptocurrency Report, published on October 14, AI agents have made more than 138 million payments through x402, the payment standard incubated by Coinbase, on the Base network.
x402 revives a long-unused part of the web's plumbing. The HTTP status code 402, “Payment Required”, was reserved decades ago and never widely adopted. The protocol lets a server answer a request with a price, and lets the client, often an AI agent, pay it in stablecoins and retry within the same exchange. No account sign-up, card form or invoice is needed, which makes it practical to charge for a single API call.
Tiny tickets, very high frequency
The most telling figure in the report is payment size. More than 87% of the analysed transactions fell between one-hundredth of a cent and 10 cents, and the most common band was one to ten cents. Chainalysis did not disclose a total dollar value. For comparison, CoinDesk reported in July that x402 had processed about 75 million payments worth roughly $24 million over a 30-day window, which implies an average well under a dollar.
Those amounts explain why the payments run on stablecoins rather than cards. Card networks carry fixed fees that make a one-cent charge uneconomic, while a stablecoin transfer on a Layer 2 costs a fraction of that. Visa, Mastercard and Stripe have nonetheless backed x402, a sign that incumbent payment firms would rather shape this channel than watch it grow outside their networks.
What agents are buying
Chainalysis ranks AI model inference as the largest spending category, followed by information services such as code review, financial intelligence, data and search APIs, and then tools and frameworks for building agents. Early volume was partly driven by memecoin activity, which has since faded. That shift matters because it suggests the remaining traffic is closer to genuine usage: agents paying per request for a model answer, a dataset or an audit.
The geography is uneven. Wallets linked to the United States accounted for 88.1% of geolocated inference payments, while Chinese-linked wallets made up just 0.1%. Yet Chinese-linked wallets accounted for about 90% of payments for security and code-analysis services, most of it going to one smart contract auditing provider. Chainalysis offers a plausible explanation for the inference gap: leading Chinese open-weight models can be run for free, while the top US models are metered per call.
Automation is taking over the wallet
The behavioural data may be the most important part of the report. The median time between a wallet's first funding and its first x402 payment fell from 73 days in late 2025 to 18 minutes in the third quarter of 2026. The share of wallets that spend within an hour of being funded rose from 4% to 67% over the same period. Retention also improved: 89.3% of wallets first funded in the second quarter of 2026 received more money within 30 days, compared with 70.6% for wallets funded in the third quarter of 2025.
Read together, these numbers describe wallets created by software for software. A developer or a platform spins up an address, tops it up and lets an agent spend from it almost immediately. That is a very different user profile from a retail investor who buys crypto and holds it for months.
The caveats
Transaction counts flatter young networks. A single agent in a loop can generate thousands of payments, and earlier research raised questions about artificial activity on x402. Chainalysis itself excluded wallets that paid on only one day, and its figures cover identifiable public-chain transactions rather than the whole market for AI payments. The dollar volumes involved remain tiny next to the $482 billion in stablecoin payments recorded through September.
Still, the direction is clear. Coinbase now lets AI agents operate inside customer accounts and lets businesses accept agent payments, and more than 10,000 merchants had integrated x402 by April, according to The Block. The next data points to watch are total settled value, the share of payments going to services outside crypto, and whether non-US model providers adopt per-call pricing. If those move, the micro-payment economy will stop being a niche experiment and start showing up in revenue lines.