AI agents are beginning to spend cryptocurrency at scale, although the value of most individual transactions is extremely small. More than 138 million payments were made through Coinbase’s x402 protocol on Base by the third quarter of 2026, according to Chainalysis research reported by Decrypt.
The activity offers an early view of “agentic commerce”: software that can use an assigned budget to purchase digital services without a person approving every transaction. The payments have primarily involved services such as computing power, live web data, AI model inference and coding tools.
A rapid rise in automated payments
Chainalysis included the figures in its annual Geography of Crypto report, which examines North America’s $3 trillion cryptocurrency economy. The report found that cumulative payments through x402 on Base increased from approximately 1,500 in mid-2025 to more than 138 million by the third quarter of 2026.
x402 is a protocol incubated by Coinbase that enables software to pay automatically for online services. Base, Coinbase’s layer-2 network, is the blockchain network on which the reported activity took place. In practical terms, an AI agent can receive spending authority and use cryptocurrency to obtain a service when it needs one, rather than waiting for a human operator to complete a separate payment.
The scale of the transaction count is striking, but the size of the payments is just as important. More than 87% of the recorded transactions were between one-hundredth of a cent and 10 cents. Such payments are often described as micro-transactions. Stablecoins, which are crypto tokens designed to maintain a stable value relative to a reference asset such as a fiat currency, can process these very small transfers. Traditional card payment systems are generally not designed for payments at this scale.
The model could allow software to pay repeatedly for individual requests or small amounts of access. However, the transaction total alone does not establish how much economic value was exchanged or how many distinct autonomous agents were involved.
The United States leads identifiable activity
The United States accounted for 88.1% of geographically identifiable payments for AI model inference in the Chainalysis data. China represented 0.1% of that activity, although wallets linked to China made up most of the spending on security and code-analysis services.
Chainalysis cautioned that these figures should not be treated as a complete map of the global market. They cover only geographically identifiable activity recorded through x402, rather than every payment made by AI agents worldwide or through other systems.
That limitation matters because the sector is still developing and because wallet location does not necessarily provide a full picture of the people, companies or software behind a transaction. The data nevertheless indicates that automated, low-value crypto payments have become a visible use case on the infrastructure covered by the report.
Payment infrastructure is expanding
The growth in x402 activity comes as major financial and technology companies work on ways for AI systems to make payments. Visa, Mastercard and Stripe have each moved to support payments made by AI agents. Coinbase has introduced a tool that allows agents to trade and transact, and it has started enabling business customers to accept payments from agents. Google has also announced a protocol dedicated to agent payments.
The emerging systems are intended to address a practical problem: an autonomous program may need to buy a service frequently, quickly and in small increments. A payment method built for software could, in principle, allow an agent to obtain data, computing capacity or another digital resource as soon as it is required.
BlackRock has identified autonomous agents as a possible future source of crypto demand. Will Su, the firm’s head of digital assets research, told Chainalysis that BlackRock is watching the area, with AI agents potentially generating demand for frequent, low-value payments around the clock. That observation describes a possible use of the infrastructure, rather than a forecast of how large the market will become.
Questions remain over what counts as an agent
The transaction numbers come with an important qualification. Earlier research from crypto tracing firm TRM suggested that a significant share of apparent agent activity on these payment rails could instead be produced by scripts or self-dealing transactions. In other words, not every automated payment necessarily reflects an independent AI system making a useful purchase on behalf of a user.
That distinction affects how the market should be measured. A high number of transactions may reflect repeated tests, automated internal transfers or other programmatic activity rather than broad adoption by genuine autonomous agents. Determining the proportion of payments generated by capable agents remains a challenge for analysts.
Even with that uncertainty, the rise from roughly 1,500 payments to more than 138 million shows how quickly the supporting infrastructure has developed. The central idea behind agentic commerce is that future crypto users may include software systems, not only people. Whether those systems become a major source of demand will depend on how widely they are deployed and how much of their activity represents real purchases.
Current market picture
Moneyiar market data at the stated times shows mixed performance among the listed assets:
- Decimal (DEL) was $0.4516439383, down 2.62% over 24 hours and up 17.55% over seven days, as of October 15, 2026, at 21:28 +03:30.
- BAS (BAS) was $0.0203143918, down 0.99% over 24 hours and 6.87% over seven days, as of October 15, 2026, at 21:29 +03:30.
- Walrus (WAL) was $0.0516680316, down 18.71% over 24 hours and up 55.36% over seven days, as of October 15, 2026, at 21:28 +03:30.
- ECOMI (OMI) was $0.0002653436, down 2.18% over 24 hours and 1.99% over seven days, as of October 15, 2026, at 21:28 +03:30.
These market prices are separate from Chainalysis’s x402 payment-count data and do not measure activity by AI agents.
