Transactions Between AI Agents Create a New Segment in Digital Payments

a joint study by analysts at Keyrock, Coinbase, Tempo, and Virtuals, the market for automated payments between AI agents has evolved over the past year from isolated pilot experiments into a fully emerging segment of the digital payments industry.

The report’s authors argue that the key trend is not transaction volume itself, but the speed at which the ecosystem is taking shape. While the first transaction between AI agents on a public blockchain in May 2025 was largely viewed as a proof of concept, by mid-2026 the market had already developed four competing payment architectures:

• The Linux Foundation is developing the x402 payment protocol originally created by Coinbase for stablecoin payments between AI agents;
• Stripe and Tempo are building the universal MPP protocol;
• Google is focusing on delegated authorization infrastructure through AP2;
• Visa is adapting its existing payments infrastructure for AI agents through tokenized credentials.

The report notes that these solutions are not directly competing with each other, but are instead forming complementary layers of a broader ecosystem.

Micropayments have become the dominant use case within the new segment. According to the study, 76% of all transactions between AI agents are below $0.30 in value — a level at which traditional payment network fees make such operations economically inefficient. For comparison, Stripe charges 2.9% plus $0.30 per transaction, while a USDC transfer on the Base network costs roughly $0.0001. As a result, nearly all payment infrastructure for autonomous AI agents is currently being built around stablecoins.

The study shows that 98.6% of all agent payments are settled in USDC. Researchers describe this both as proof of stablecoins’ dominance in the sector and as a potential systemic risk, since the market is becoming heavily dependent on a single infrastructure provider.

At the same time, consolidation within the sector is accelerating. Over the past year, companies have invested more than $8 billion in acquisitions and infrastructure deals related to agent payments.

The report also highlights that a significant share of activity on some blockchain networks is already generated not by humans, but by automated AI agents. On Base and Optimism, automated systems account for more than half of all network activity, while on Gnosis Chain over 75% of transactions come from automated users operating through Safe wallet infrastructure. At the same time, the market is gradually shifting away from arbitrage and MEV toward practical commercial use cases, including payments for API access, analytics, computing resources, and other digital services.

Researchers identify regulation as the main constraint on further market growth. None of the existing legal frameworks currently provide a comprehensive regulatory basis for transactions between autonomous AI agents. According to the report, issues surrounding agent identity, liability allocation, and compliance will ultimately determine the pace at which the agent economy can scale.

Earlier forecasts suggested that by 2030 AI agents could process up to $5 trillion in transactions, with stablecoins and blockchain networks serving as the core infrastructure layer for these payments. Ethereum, in particular, could emerge as a foundational network for the agent economy.