In the span of a single year, autonomous AI agents have made more than 138 million crypto payments—the vast majority worth less than a dime—through Coinbase’s x402 protocol on Base. According to a new Chainalysis report, the number of agentic transactions exploded from around 1,500 in mid-2025 to over 138 million by the third quarter of 2026. What started as an experiment in machine-to-machine micropayments has quietly become a backbone for autonomous software buying data, compute power, and access to AI models.
For developers, this shift matters because it rewrites the economic relationship between software and services. Traditionally, payment flows are initiated by humans—clicking a button, approving a card charge. Here, agents are making split-second decisions about what to purchase and paying without a human in the loop. The x402 protocol enables transactions as small as a fraction of a cent, which unlocks entirely new categories of utility: paying per API call, per inference, per megabyte of storage. The infrastructure exists for a machine economy where services are consumed by software not just through subscriptions but through dynamic, real-time micro-transactions.
Yet the data also reveals a lopsided geography. U.S.-linked wallets accounted for 88.1% of identified payments for AI model inference, while China’s share was just 0.1%. That concentration mirrors the dominance of American cloud and AI model providers, but it also raises questions about the resilience and inclusivity of this emerging ecosystem. If agentic commerce becomes central to how AI systems operate, a world in which a handful of jurisdictions control the economic rails could create new vulnerabilities—both in terms of access and governance.
The promise of agents handling their own costs is a step toward more autonomous and efficient software, but the current pattern suggests we are building an infrastructure whose balance of power is already heavily skewed. Developers working in this space should consider what it would mean to design for a more distributed agent economy, one where IP, compute, and model access are not channelled through a single geopolitical hub. The technology is still young, and the architecture we choose now will shape whether autonomous commerce remains open or becomes a new kind of bottleneck.