Chainalysis says the US dominates AI agent payments — but how much of all that payment activity did it actually manage to trace?

This explores how much of all x402 activity Chainalysis can actually trace to a country, because that coverage decides how far you can trust its country rankings. The short answer is that the corpus doesn't know, and Chainalysis doesn't say. Every country figure in the report is a share of the transactions the firm "can geolocate." That includes US wallets making up 88.1% of AI model inference payments, 41.8% of payments to agent frameworks and tool catalogs, and about 30% of all agent-to-services transactions. The report never says what fraction of total x402 volume that geolocatable slice is, so some unknown amount of activity is left out of every percentage Which countries dominate agent-to-agent AI payments on chain?.

The gap matters because the overall volume is large and well counted. Chainalysis reports more than 138 million x402 transactions on Base, the blockchain where most of this activity happens. More than 87% of them are micropayments between a hundredth of a cent and a dime Are AI agent wallets becoming genuinely operational instead of experimental?. So we know a lot about how much is moving and very little about how much of it can be placed on a map. "88.1% of geolocated inference payments" could be describing almost all inference traffic or a small, unusual corner of it, and the report gives no way to tell which.

It also helps to know how the attribution works. Chainalysis assigns a wallet to a country based on where its funding came from, not where the agent runs or who instructs it. The report doesn't describe that method in detail Which countries dominate agent-to-agent AI payments on chain?. In practice, a country can only be assigned when the money trail leads back to something tied to a place, such as a regulated exchange. Wallets funded entirely from other on-chain sources are much harder to place. If many agent wallets are topped up automatically by other software, the part that can't be attributed might be exactly the most autonomous, machine-to-machine traffic. That last point is an inference, not a finding in either note. Either way, the attributed sample may be skewed toward agents with an obvious human or company funding them.

The companion finding on wallet behavior has the same blind spot one level up. Chainalysis shows wallets going from funding to first spend in a median of 18 minutes, down from 73 days, and repeat-funding rates rising to about 89%. That's strong evidence that a slice of activity has become routine. But the report doesn't say who or what is telling these wallets to spend: people delegating to agents, fully autonomous agents, or bots gaming the protocol. It also doesn't say how big this slice is compared with all AI-agent economic activity Are AI agent wallets becoming genuinely operational instead of experimental?. It's a useful pattern to notice. Agent-payment analytics are currently very precise about how many transactions happen and how fast, and quiet about who is behind them and how complete the picture is. Before acting on any country share, first ask what it is a share of.