The payment market for artificial intelligence is rapidly evolving, and as recent research shows, traditional cards and stablecoins will occupy fundamentally different niches in this ecosystem. A joint analysis by Visa experts and the blockchain analytics company Artemis clearly divided AI agent transactions into two types.

The first type is purchases made on behalf of a human: booking tickets, managing subscriptions, ordering goods. Such operations are essentially identical to regular online commerce and work comfortably through the existing card infrastructure. The second type is frequent micropayments between programs, where the transaction amount rarely exceeds one dollar. This involves paying for individual API requests, accessing computing power, purchasing data, or using software tools.

This is precisely where the fundamental advantage of blockchains becomes apparent. Fixed fees make the card infrastructure completely unprofitable for such operations. At the same time, in the Base and Solana networks, transaction fees amount to fractions of a cent. The report examines two open payment protocols: x402, which is most actively used on Base, Solana, and Polygon, and the Machine Payment Protocol (MPP), which supports settlements both through blockchains and traditional payment systems.

Hybrid Future and Scale

The authors see the most likely model as a hybrid approach: cards will remain for authorization and purchases on behalf of humans, while stablecoins will handle settlements between programs. Both methods can be used at different stages of a single task. According to Artemis data as of April 21, after filtering out fictitious and test operations, approximately $15 million passed through x402 across 109.6 million transactions. For MPP, the figures are more modest — roughly $25,000 and 115,000 transfers, respectively.

Interestingly, the raw data is significantly higher: $135.7 million and 178.3 million operations via x402, as well as over $38,000 and about 184,600 transactions via MPP. The difference is explained by the exclusion of operations that Artemis classified as fictitious activity or testing. This resembles the situation in March, when Artemis analysts excluded fictitious operations from x402 statistics, after which the real volume of AI agent transactions turned out to be 15 times lower than Bloomberg's initial estimates.

The Problem of Trust and Legal Liability

The authors identify trust as the main unresolved issue. It is currently completely unclear who should be responsible for an erroneous or unauthorized purchase: the owner of the AI agent, the platform operator, the model developer, or the seller. Existing payment reversal rules are designed exclusively for human transactions, and there is simply no established mechanism for disputing thousands of automatic transactions per hour. An additional risk is posed by malicious instructions capable of redirecting a payment or forcing an AI agent to make an unplanned purchase.

My comment: While lawyers and regulators sort out the issue of liability, stablecoins are already beginning to capture the niche of inter-program settlements. The June launch of the AgentCard service from Alchemy and Visa is just the first sign. I believe that the hybrid model will become the de facto standard, but only after the industry solves the chargeback problem for AI agents. Without this, mass adoption will not happen.