BroadChain reports that at 16:06 on April 24, the market has been chasing "AI tokens" for three years, but the tokens actually used by AI may have never entered most people's sight. Tokens like FET, TAO, and RENDER, which carry "AI" in their names, are not used by the AI agents running in practice to pay for GPU computing power, call APIs, or pay model royalties. AI agents can initiate hundreds of payments per second, making the 2.9% + $0.30 fee of credit cards unsustainable, the three-day cycle of ACH wire transfers too slow, and any payment process requiring human confirmation outdated.
Hundreds of millions of dollars still flow between machines every day, but what flows is not FET, TAO, or RENDER, but USDC. In 2025, Coinbase partnered with Circle and Google to revive the long-dormant HTTP 402 status code from 30 years ago—the x402 protocol. When an AI agent accesses a paid API, the server directly returns a 402 status code containing payment metadata, the agent wallet automatically signs a USDC transfer and resubmits the request, completing the entire process within two seconds without the need for accounts, API keys, or human confirmation.
Circle's 2026 data shows that over 98% of agent-driven payments chose USDC, with an average transaction amount of just $0.31—a scale that can only be machine-to-machine. To this end, Circle built the Arc chain, which reduces USDC transfer costs to the level of $0.00001 through off-chain aggregation plus on-chain batch settlement. In contrast, the fixed fees of credit cards alone exceed the total value of machine-level micropayments by an order of magnitude.
Tokens like FET, TAO, and RENDER, dubbed "AI tokens," hardly appear in AI agent wallets; they sit in the smart contract accounts of speculators, fluctuating 5-10% intraday. If an AI agent paid for computing power rent with TAO, it could rent 1,000 hours today but might only afford 900 hours tomorrow, making budget planning impossible and financial plans unfeasible. Ironically, tokens touted as "AI" have zero circulation in the real AI economy—they are traded by humans, not used by machines.