Chainalysis counts 138 million x402 agent payments, most of them worth 10 cents or less

The 138 million figure is a count, and the dollar volume behind it is modest. In July, x402 processed roughly 75 million payments worth $24 million over 30 days ^[14], an average of 32 cents ^[16]. Set that against the 87% of analyzed payments worth 10 cents or less ^[2] and the average implies a minority of much larger payments, assuming July's mix matched the report's (the samples differ, so treat the result as a rough floor). The cheap 87% adds at most 8.7 cents to the average ^[17]. The other 13% or fewer would then average at least $1.79 ^[18] and carry roughly 73% of the money ^[19].

The case for reading the penny payments as the product is strong. At those sizes the fee on a card payment can cost more than the service itself, according to Fortune ^[9], and Eric Jardine, Chainalysis' head of research, told Fortune: "You have this perfect combination of speed plus low cost to settle, and that's the sort of thing that you would need if you were doing true AI-to-AI commerce." ^[11] The wallet-funding figures point the same way, with two qualifications.

The repeat-funding comparison excludes wallets that made payments on only one day, according to CoinDesk's account of the report ^[12], so it describes wallets that kept going. CoinDesk also noted that earlier research raised questions about artificial activity on the network ^[21].

Jardine expects the run to continue. "I would hazard that the trend is likely to be exponential," he said ^[24]. Cumulative payments on Base started from about 1,500 in mid-2025 ^[8], so the rise to 138 million is a factor of roughly 92,000 ^[15].

The geographic split needs the same care. Jardine's explanation for the thin Chinese-linked share of inference spending is hedged in his own words: "One possible explanation is that Chinese developers are more inclined to use open-weight models such as DeepSeek and Qwen, reducing their need to purchase metered inference services," he told CoinDesk ^[23]. The figures also measure identifiable x402 transactions, not the entire market for AI payments ^[10].

Payments above a dime could grow faster than the count, pushing the average ticket up from July's 32 cents, in which case the pennies are the rails and the tail is the business. Or the average could hold near 32 cents while the count climbs, leaving x402 a large number of very small sales. Or more of the count could be artificial than the funding data suggests, and demand would be smaller than any of these figures. We think the middle case describes today's evidence, or rather, the middle case with the first as its test: 138 million payments show that agents can pay without a person approving each purchase, and whether buyers would otherwise have paid for the same work is untested. We would be wrong if payments above a dime are already growing faster than the count. The coverage does not include a dollar split by payment size.