When AI agents pay each other in crypto, are they buying real services or just shuffling speculative money around?

This explores what AI agents actually spend money on when they pay with crypto over x402, and whether the activity is real. The answer surprises most people: agents are mostly not buying products for humans. They are buying services from other machines. Chainalysis traced x402 transactions to their geographic origin and found the main categories are AI model inference (paying to run a query on someone else's model), agent-framework services (the tooling agents use to coordinate and act), and security services Which countries dominate agent-to-agent AI payments on chain?. Spending also clusters by country. US wallets account for 88.1% of inference payments and 41.8% of agent-framework payments, while Chinese wallets concentrate in security. Chainalysis reads this as a move away from speculative memecoin trading toward spending on agent-to-agent infrastructure.

The spending has also become routine. The median time between funding an agent wallet and its first payment fell from 73 days to 18 minutes, and the share of wallets that get topped up again rose from 71% to 89% Are AI agent wallets becoming genuinely operational instead of experimental?. In practice, people now fund a wallet because the agent has an immediate job to do, and they keep refilling it the way you'd refill a transit card.

Why crypto rather than a credit card? Fintech commentator Simon Taylor offers a useful explanation. He argues agents will use virtual cards first for ordinary purchases, because cards already have mature spending controls and are accepted almost everywhere. Native stablecoin wallets earn their place in multi-agent setups Will agents adopt cards before stablecoins for payments?. A stablecoin is a cryptocurrency pegged to a currency like the dollar. With stablecoins, a 'master' agent can instantly create sub-wallets with custom spending rules and hand them to sub-agents, without waiting for a bank or card issuer to approve each one Can stablecoins let agents create spending authority without approval?. That fits what x402 data shows: many tiny, fast payments between pieces of software, such as one agent paying for an inference call or a security check mid-task.

This makes x402 a different world from consumer agentic shopping, which hit a wall of a different kind. When Amazon blocked Meta's Muse shopping agent, it exposed a missing piece: retailers have no standard way to check who an agent is or whether it has permission to buy What identity standards do shopping agents actually need?. Consumers are cautious too. Only 23% of US adults trust AI to make payments, and that number rises sharply when a familiar brand like Visa is attached Does naming a payments brand increase trust in AI transactions?. Machine-to-machine services on x402 mostly avoid both problems. No retailer has to be persuaded and no nervous shopper has to approve the purchase, which may explain why this is where agent spending is already established.

The corpus has limits here. It describes spending by category and country. It doesn't list specific vendors or typical transaction sizes, and it doesn't show how much x402 spending ends up as goods for human end users. So the honest answer is: compute, tooling, and security for other agents. Agents are mostly paying for their own supplies.