Via nbclosangeles.com
The protocol built on HTTP's forgotten error code is quietly turning AI agents into paying customers, with 75 million transactions processed in the last month alone
AI agents have initiated 14 million transfers through the x402 protocol over the past 30 days, a signal that machine-to-machine payments are moving from theoretical curiosity to measurable infrastructure. The protocol, which piggybacks on the long-dormant HTTP 402 “Payment Required” status code, has turned that dusty web standard into a live payment rail, and Coinbase’s Layer-2 chain Base is where most of the action is happening.
The broader x402 ecosystem is even larger than that 14 million figure suggests. Across all participants, the protocol has processed roughly 75 million transactions and $24 million in volume over the last 30 days, with an average payment hovering around $0.32. These are not humans buying coffee. These are AI agents paying for API calls, compute resources, data feeds, and content, automatically, without anyone clicking “confirm.”
How a forgotten error code became a payment protocol #
Every web developer has encountered the HTTP 402 status code. It was baked into the original HTTP specification back in the 1990s as a placeholder for future digital payment systems that never materialized. Coinbase picked up that unfinished business and built x402, an open standard that lets any client, human or AI, attach a stablecoin payment directly to an HTTP request.
The mechanics are straightforward. When an AI agent hits a server that requires payment, the server responds with a 402 status code and the payment terms. The agent sends a USDC transaction on a supported chain, the server verifies it, and the content or service is delivered. No credit card, no subscription, no billing department. The whole exchange happens within the standard HTTP flow.
Transaction fees typically land below one cent. Credit card networks charge a minimum of roughly 20-30 cents per transaction, making sub-dollar payments economically nonsensical on traditional rails. When your average transaction is $0.32, Visa’s interchange fee would eat most of the payment. On Base, the math actually works.
Base emerges as the default settlement layer #
Base has emerged as the dominant chain for x402 deployments, accounting for the majority of the protocol’s transaction volume. USDC serves as the primary settlement asset across the protocol. For AI agents that need to pay for services programmatically, stablecoins eliminate the volatility problem that would make budgeting impossible with assets like ETH or Bitcoin.
The cumulative transaction count tells the scaling story. By mid-2026, between 100 and 119 million total x402 transactions had been processed, with the growth curve steepening considerably compared to earlier activity levels in 2025 and early 2026.
The foundation and its heavyweight backers #
Coinbase handed governance of the protocol to the x402 Foundation, which launched operationally in July 2026 under the Linux Foundation’s umbrella. The foundation counts 40 members, including Visa, Mastercard, and Google.
Visa and Mastercard built empires on facilitating human-initiated transactions with minimum viable amounts well above a dollar. They’re joining because AI agent commerce could become a transaction category large enough to matter, and the economics make it impossible to serve on their existing rails. Google’s participation points to the demand side of the equation: as a provider of AI models and cloud infrastructure, Google has a direct interest in how AI agents pay for services.
What 75 million transactions actually mean #
At 75 million transactions and $24 million in volume over 30 days, x402 is processing roughly 2.5 million transactions daily. These are transactions that would not exist on traditional payment rails. No human is going to pull out a credit card to pay $0.32 for an API call that an AI agent makes autonomously.
When transaction costs sit below one cent on Base, even the smallest micropayments remain economically rational. This opens up pricing models that were previously impossible: pay-per-query for data, pay-per-inference for AI models, pay-per-byte for storage.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our