AI Agents Logged 138 Million Crypto Payments on Coinbase's x402, Chainalysis Reports

138 Million Payments Through x402

Blockchain analytics firm Chainalysis published a report on October 14 sizing up how autonomous AI agents spend on-chain, and the topline figure is large: cumulative transactions through x402, the Coinbase-incubated payment protocol, had cleared 138 million on Base, the layer-2 network that periodically settles to an underlying layer 1 blockchain protocol. x402 is built for machines rather than people: it lets software pay automatically for digital services such as live data, computing power and access to AI models, settling each charge in cryptocurrency with no checkout step. The report sits inside Chainalysis's annual Geography of Crypto series, which this year focuses on North America's roughly $3 trillion crypto economy, and it frames agentic commerce, the emerging pattern of software agents transacting on their own, as one of the sector's fastest-emerging use cases. The payment sizes are as telling as the totals: more than 87% of x402 payments recorded across the 16 months to September 2026 fell between one hundredth of a cent and 10 cents, micro-charges traditional card rails struggle to process profitably. Stablecoins can settle sums this small; card networks, with fixed per-transaction fees, generally cannot. Geography splits just as sharply. Wallets tied to the United States accounted for 88.1% of geographically identifiable payments for AI model inference, while China-linked wallets made up 0.1%. For security and code-analysis services the ranking flips: Chinese-linked wallets held roughly 90% of identifiable spending there, most of it flowing to AgentLISA, an AI-powered smart contract auditing service. Chainalysis ties the gap to developer habits, noting builders in China lean on open-weight models such as DeepSeek and Qwen, which lowers demand for pay-per-use inference. The firm also cautioned that the figures capture identifiable x402 activity only, not the entire global agent-payment market.

Visa, Google and BlackRock Circle the Rails

The curve behind that count is steep. Roughly 1,500 payments had cleared through x402 by mid-2025; by the third quarter of 2026 the running total had passed 138 million, as developers increasingly hand each agent a budget and an AI crypto wallet and let it buy computing power, live web data and coding tools in the middle of a task. The rails beneath these flows matured in parallel. Visa, Mastercard and Stripe have each moved to support payments initiated by AI agents, and Google has unveiled a dedicated agent-payments protocol. Coinbase, which incubated x402, has shipped tooling that lets agents trade and transact, and has begun letting its business customers accept agent payments. Asset managers are paying attention as well. Will Su, head of digital assets research at BlackRock, told Chainalysis the firm follows the space closely, “with AI agents potentially creating meaningful demand for frequent, low-value payments around the clock.” The counts carry a caveat. Earlier research from TRM, a rival crypto tracing firm, suggested much of the apparent agent activity on these rails may be scripts or self-dealing rather than genuine autonomous agents, meaning raw transaction totals can overstate how much real agentic commerce is taking place. The direction of travel still points one way, and the thesis gaining ground across the industry is that crypto's next wave of users may not be people at all.

Where the Count Overstates Demand

Our reading: the 138 million headline needs an asterisk before it is treated as commercial traction. The report itself notes that early x402 volumes drew partly on memecoin-related activity, so transaction growth does not map one-to-one onto real demand for agent services. The more durable datum is the payment mix, with more than 87% of charges under 10 cents, because it shows agents buying services in increments card networks price out. TRM's script warning is the sharpest counterweight in the dataset, and dismissing it as FUD would ignore a measurement problem nobody has solved: separating genuine autonomous agents from scripted self-dealing at scale. Watch whether the sub-10-cent share holds as agent budgets grow; that mix, not the raw count, will show whether agentic commerce is an economy or an echo.