Crypto rails are fast becoming the payment backbone for autonomous AI agents, according to a new report from Keyrock — and big tech, payments and crypto firms are racing to build the plumbing. What Keyrock found - Between May 2025 and April 2026, AI agents settled roughly $73 million across about 176 million on-chain transactions. That dollar figure is tiny next to TradFi — Visa alone processes about $14.5 trillion a year — but the rapid formation of an infrastructure stack is the real story. - Major players are already shipping competing systems: Coinbase’s x402 protocol, Stripe’s Tempo with its Machine Payments Protocol (MPP), Google’s AP2 for delegated spending, and Visa’s tokenized credentials for AI-driven commerce. Why crypto rails matter for AI agents Agentic payments describe software that autonomously purchases digital services — think an AI trading bot buying continuous market data, cloud compute or API-based analysis in tiny increments without human sign-off. Those microtransactions are often just a few cents or less, making card rails impractical: Keyrock notes 76% of agent payments fall below the typical $0.30 fixed-fee floor for cards, and most payments are between $0.01 and $0.10. By contrast, settling in stablecoins on layer-2s like Base and Tempo can cost fractions of a cent. Stablecoins and concentration risk Stablecoins have emerged as the settlement medium of choice: around 98.6% of machine payments in Keyrock’s sample used USDC, the Circle-issued stablecoin. That dominance cements USDC’s role in crypto payments but also concentrates risk around a single issuer. How big could this get? Analyst forecasts are eye-popping: Gartner projects AI agents could intermediate $15 trillion in purchases by 2028; McKinsey estimates retail agentic commerce could reach $3–5 trillion by 2030. Keyrock argues those growth rates, if realized, would outpace even the breakout years of stablecoins — and the speed of current infrastructure rollout suggests the market is moving beyond pure experimentation. Regulatory question marks Regulation may become a constraint. Several major regimes — Europe’s MiCA, the U.S. GENIUS Act and the EU AI Act — are expected to take effect around mid‑2026, but Keyrock says none directly tackles the legal gray areas of autonomous machine-to-machine transactions, such as liability and agent identity. Bottom line The dollar totals today are modest, but infrastructure and protocol competition are heating up. Expect continued experimentation and rapid product launches from crypto, payments and tech firms — alongside scrutiny over concentration in USDC and future regulatory responses as machine-driven commerce scales. Read more AI-generated news on: undefined/news