Chainalysis counts 138 million payments through AI-focused crypto protocol

Software wallets are spending sooner and getting replenished more often, according to new Chainalysis research covered by Fortune. Cumulative payments through x402, a protocol that lets AI agents pay for online services with crypto, exceeded 138 million on Base by Q3 2026. The activity suggests sustained use, but transaction counts alone cannot establish widespread autonomous commerce.

Tiny purchases, not a retail shopping boom

The rise is steep: cumulative x402 payments on Base, the Coinbase-incubated blockchain, stood at about 1,500 in mid-2025. Chainalysis examined micropayments over the 16 months ending in September 2026. Its choice of x402 was deliberate: the protocol was designed for agent-to-agent commerce, making it a focused window into that activity rather than a survey of every way AI systems make purchases.

The spending model is closer to an allowance for software than a shopping cart. A developer deposits stablecoins—typically Circle’s dollar-pegged USDC—into a bot’s wallet. The bot can then purchase resources as it works without seeking human approval for each payment. Those resources include computing power, fresh web information, help from a more capable AI model and automated checks for security flaws in code.

More than 87% of payments were for less than 10 cents. At that size, a credit-card processing fee can exceed the price of the service itself. Eric Jardine, Chainalysis’s head of research, told Fortune that fast settlement and low costs make blockchains well suited to these small, automated purchases. That is the proposed economic advantage: paying separately for a narrow digital input when a task needs it.

Repeat funding offers a stronger signal—with a filter

The wallet behavior adds something the headline payment total cannot: evidence about how quickly funds enter use and whether developers supply more. Roughly two-thirds of new wallets spent within an hour of receiving funds. Fortune described the combination of faster spending and repeat deposits as suggesting that developers are financing ongoing tasks, rather than only one-off experiments.

But the repeat-funding result applies to a selected group. As PYMNTS explained, Chainalysis excluded wallets that had not sent payments on at least two separate days. Among those qualifying wallets, 89.3% of the group first funded in Q2 2026 received a second deposit within 30 days. That compares with 70.6% for wallets first funded in Q3 2025.

The distinction changes the reading of that near-90% figure. It describes replenishment among wallets already active across multiple days, not the likelihood that any newly funded wallet will become a repeat user. It supports a narrower finding about continued funding among qualifying users, rather than a retention rate for everyone who tried the system.

Payment activity is not proof of autonomous demand

PYMNTS raised a more fundamental qualification. Citing Decrypt’s account of earlier TRM research, it noted that scripts and self-dealing may account for much apparent agent activity. That is a caution about interpreting these transactions, not a demonstrated breakdown of the 138 million payments. A high count does not, by itself, show how many independent AI purchasing decisions occurred or how much useful commerce they generated.

Even the geographic results have a defined boundary. American-linked wallets accounted for 88.1% of geographically identifiable payments for AI model responses, compared with 0.1% for China. Those shares describe identifiable x402 activity—not the global market for AI services. They cannot be read as a country-by-country map of all AI spending.

Jardine expects growth to continue as more platforms accept agent payments, arguing that network effects will attract further activity. PYMNTS set a different test for durability: whether purchases deliver useful services, customers return and spending stays within authorized budgets. The contrast is between a forecast driven by expanding participation and a standard grounded in economic usefulness and spending control.