Visa, Artemis: Card rails can’t handle AI agent micropayments

Visa and Artemis released a joint report Wednesday saying current card payment systems cannot handle the high-frequency, near-zero-fee micropayments required by autonomous AI agents. The firms said that shortfall is slowing the growth of agent-driven commerce.

The report states that card rails were built for human shoppers and relatively infrequent transactions. AI agents now require much lower fees and faster settlement so tiny, rapid payments become commercially viable.

The report says AI agents crossed a capability threshold in mid-2025 that allows them to discover unfamiliar APIs, compare prices and make payments without human intervention.

The document points to early signs of machine-native payment systems. Coinbase’s x402 protocol recorded about $15 million in adjusted volume across more than 109 million adjusted transactions since its May 2025 launch. In October 2025 the protocol’s monthly transaction count rose from roughly 40,000 to 3.8 million, and it processed about 38 million transactions that month.

The report argues a single machine-payments framework could support both stablecoin and card-based flows. “The trajectory points toward convergence rather than competition: cards for proxy purchases inside existing merchant networks, stablecoins for machine-native micropayments, and hybrid flows where both are used within the same workflow,” the report says.

The paper highlights Tempo’s Machine Payment Protocol, which links onchain crypto payments and fiat through shared payment tokens, and Visa’s Card Specification SDK, which is intended to extend machine payment standards into card-based commerce.

Payments firms have begun introducing tools for AI actors. In March, Visa’s crypto division rolled out an offering allowing same-day payments by AI agents, and Tempo launched its Machine Payments Protocol to simplify sending and receiving money for automated systems.

The report cites market estimates including an analysis from Australian exchange Swyftx that projects AI-enabled microbusinesses could add about $262 billion in stablecoin volume by 2033 if roughly one-third of eligible activity adopts machine-native payments.

The report concludes current infrastructure gaps-transaction fees that are too high for tiny payments and settlement times that are too slow-are a barrier to wider adoption of agent-driven commerce. It recommends building payment rails that support near-zero fees, faster settlement and protocols that let cards and stablecoins interoperate within the same workflows.