Which countries dominate agent-to-agent AI payments on chain?

Chainalysis, geolocating x402 transactions by the same methodology it applies to other on-chain activity, finds the United States leads in agent-to-agent spending on the inputs that build agentic software. "American-funded AI agents are transacting more with x402 services than agents funded from any other country," holding "a 30% share of the agent-to-services transactions that we can geolocate." The US lead concentrates most sharply in model inference: "American wallets account for 88.1% of all geolocated AI model inference transactions on the protocol, 41.8% of payments to agent frameworks and tool catalogs, and 23.2% of payments to data and search APIs." Chainalysis calls these "the three categories that supply the inputs to building agentic software." By contrast, "Chinese-attributed wallets were most prominent in the security space."

Chainalysis frames this as evidence that "AI agents are also paying for AI agents" — one of the largest x402 transaction categories is model inference itself, meaning agents are autonomously purchasing the compute and tooling that other agents (or their own next steps) require, rather than only paying for goods and services on behalf of a human. The data also moves beyond the early memecoin era of x402 activity ("in the early days... the answer was predominately fads; memecoins such as PING attracted huge numbers of new wallets only to fade") toward what Chainalysis describes as more durable spend in "the information sector," including "code review, financial intelligence, and data and search API" services.

This agent-paying-for-inference pattern connects directly to Will AI inference demand outpace supply capacity soon?: if model inference is already one of the largest categories of autonomous machine-to-machine payment on x402, and the US dominates that category at 88.1%, agentic payment rails are a direct channel through which growing token demand converts into paid inference consumption — the same supply-demand mismatch Emberson and Sevilla's estimate describes from the compute side. Are AI agent wallets becoming genuinely operational instead of experimental? establishes, from the same dataset, that this spending is increasingly immediate and non-experimental rather than speculative.

Chainalysis's geolocation methodology attributes wallets to countries by funding source, which the report does not describe in detail, and the figures cover only transactions the firm "can geolocate" — an unspecified share of total x402 volume is excluded. The report does not explain why China concentrates in security services specifically, nor whether the US's inference-spend dominance reflects where AI labs are based, where agents are deployed, or where compute is priced most accessibly on-chain; the finding establishes a geographic concentration in agent-to-agent payment flows without establishing its cause.