Visa and Mastercard Buy Into Stablecoins as AI Agent Payments Split

The infrastructure war over AI agent payments has produced a clear split: card networks control consumer retail, stablecoins dominate machine-to-machine commerce, and both incumbents are now betting on both sides.

According to a report first published by Forbes, Mastercard launched Agent Pay in April 2025 using its existing tokenization technology, bringing Microsoft, IBM's watsonx Orchestrate, and checkout platforms Braintree and Checkout.com into the fold. Visa followed with Visa Intelligent Commerce and AI-Ready Cards, partnering with Anthropic, OpenAI, Perplexity, Mistral, and Samsung. Both approaches keep transactions inside the traditional four-party card model, replacing card details with tokenized credentials that confirm consumer authorization within preset limits.

The stablecoin alternative arrived a month later when Coinbase launched x402, a protocol that revives the dormant HTTP 402 "Payment Required" status code to settle payments in USDC directly over the web. A client requests a resource, receives a payment instruction, sends a signed stablecoin payment in the request header, and receives the resource once the on-chain payment settles. No account, no card, no interchange fees.

Why it matters

The real story is not which rail wins but that the card networks are buying exposure to both. Visa's stablecoin settlement pilot reached a $7 billion annualized run rate by April 2026, up 50 percent from the prior quarter, while expanding to nine blockchains and running more than 130 stablecoin-linked card programs across 50-plus countries. Mastercard agreed to acquire stablecoin platform BVNK for up to $1.8 billion in March 2026. If the incumbents believed agent commerce would settle exclusively on cards, they would not be spending billions on stablecoin infrastructure.

The deployment pattern reveals the divide

Consumer-facing launches chose cards. OpenAI's Instant Checkout in ChatGPT, live from September 2025 and built with Stripe, settles over card rails using a Shared Payment Token scoped to a single merchant and cart. It started with Etsy sellers and expanded to more than a million Shopify merchants. Amazon's Buy for Me agent autofills the user's existing card at checkout.

Stablecoins won the machine layer. Amazon integrated x402 into its Bedrock AgentCore Payments service, where settlement happens in roughly 200 milliseconds on Coinbase's Base network for a fraction of a cent. Coinbase reports x402 processed more than 169 million payments across 590,000 buyers and 100,000 sellers in its first year—transactions representing agents paying for compute, data, and API calls at frequencies and price points cards cannot economically handle.

The identity problem decides consumer adoption

The harder challenge beneath both rails is identity. When software initiates a payment, merchants need to verify it is a legitimate agent acting for a real customer, not a bot exploiting stolen credentials. Visa cited a 4,700 percent surge in AI-driven traffic to US retail sites as the reason it built a Trusted Agent Protocol with Cloudflare to separate sanctioned agents from scrapers.

This is where card networks hold structural advantage. Fifty years of fraud scoring, chargeback rules, and dispute machinery are exactly what an agent economy needs when a bot makes an incorrect purchase. Stablecoin rails settle with finality and no native reversal mechanism. Whoever solves agent identity and dispute resolution—not whoever moves the cheapest payment—may decide the consumer side of this market.

The boundary between consumer retail and machine payments is exactly what the networks are spending to control. The question that decides the outcome is whether agent-driven commerce resembles retail shopping, preserving the card networks' position, or a dense web of tiny machine payments, where stablecoin rails capture volume that never existed before. Visa and Mastercard have answered by buying exposure to both and positioning to charge fees whichever way traffic flows.

Details were first reported by Forbes contributor Zennon Kapron.