Why Stablecoins Are Powering the First Wave of AI Agent Payments

The first wave of autonomous AI commerce is emerging—and it looks very different from traditional online payments.

A report from crypto market maker Keyrock analyzing activity between May 2025 and April 2026 found that AI agents executed around 176 million blockchain transactions totaling more than $73 million. The data suggests that stablecoins are rapidly becoming the default payment rail for machine‑to‑machine transactions, particularly for the tiny payments typical of automated services.

The reason is simple: AI agents pay differently than humans. Instead of occasional purchases, they often perform thousands of automated microtransactions for APIs, data, compute, or model inference—payments that conventional card networks were never designed to handle.

Why Credit Card Networks Don’t Work for AI Transactions

The economics of card payments break down at the scale AI agents operate.

According to summaries of the Keyrock analysis, the median AI‑agent transaction size falls between roughly $0.01 and $0.10. Even more striking, about 76% of payments occur below the roughly $0.30 fixed fee floor common in card processing.

That means the processing fee alone can exceed the value of the transaction. For example, an agent paying $0.03 for a weather API request would incur a fixed card fee that is an order of magnitude larger than the purchase itself.

Traditional payment infrastructure also introduces other mismatches for autonomous software:

• Card networks require human‑managed accounts and credentials.
• Payments typically settle with delays through multiple intermediaries.
• The system is optimized for consumer purchases, not continuous machine‑to‑machine activity.

For automated services that may execute thousands of small transactions per hour, these constraints make card rails economically impractical.

Why Stablecoins Fit Machine Payments Better

Stablecoins solve several structural problems for AI agents.

On blockchain networks—especially Layer‑2 systems—payments can settle in seconds with fees measured in fractions of a cent, making them viable even for extremely small transactions.

Instead of storing card credentials or opening merchant accounts, an AI agent can simply sign a transaction that transfers digital dollars such as USDC directly to the service provider. The result is an internet‑native payment model where software can pay software instantly.

This architecture also enables new pricing models: pay‑per‑API call, pay‑per‑inference request, or real‑time consumption billing for data and compute services.

How USDC Became the Dominant Settlement Asset

One of the most striking findings in the Keyrock analysis is how quickly the market converged on a single stablecoin.

Across AI‑agent transactions analyzed in the report period, about 98.6% of payments were settled in USDC.

Several factors likely contributed to this concentration:

• USDC is widely integrated across developer platforms and payment tooling.
• The token is dollar‑pegged, making pricing stable for services.
• It is commonly used across exchanges, wallets, and blockchain ecosystems.

Because machine payments require interoperability across platforms and services, developers often standardize on the settlement asset that is easiest to integrate across the ecosystem.

The New Protocol Stack for AI Agent Payments

Alongside stablecoins, an entirely new infrastructure layer is forming to support autonomous commerce.

Coinbase’s x402 Protocol

Coinbase introduced x402, an open payment protocol that revives the long‑unused HTTP 402 “Payment Required” status code to enable automated payments over the web.

With x402, an API or website can request payment directly within an HTTP response. The client—human or AI agent—can then automatically submit a stablecoin payment (such as USDC) and continue the request flow without creating accounts or managing subscriptions.

This model allows services to monetize APIs or digital resources with per‑request micropayments.

Stripe and Machine‑to‑Machine Payments

Stripe is also developing infrastructure for automated payments, including systems designed for machine‑to‑machine transactions and agent‑driven commerce. Some reporting links these efforts to the Machine Payments Protocol (MPP) developed with the Tempo blockchain.

These systems aim to let developers charge AI agents programmatically for API calls, data access, or automated services.

Visa’s Agentic Commerce Infrastructure

Traditional payment networks are not standing still. Visa has introduced Intelligent Commerce Connect, part of its broader Intelligent Commerce platform designed to support transactions initiated by AI agents.

The system provides a single integration that allows merchants to accept agent‑initiated purchases while incorporating authentication, tokenization, and payment controls designed for automated commerce.

Big Tech Is Building the Infrastructure Layer

Beyond payment networks and crypto platforms, major cloud providers are also entering the space.

Companies such as AWS and Google are building infrastructure that allows AI agents to manage credentials, spending authorization, and transaction workflows within agent frameworks and cloud environments. Reporting around the Keyrock findings places these firms in the orchestration layer of the emerging “agentic commerce” stack.

In practice, this means future AI systems may be able to:

• Discover paid APIs or datasets
• Authorize spending within a defined budget
• Execute purchases automatically
• Log transactions for governance and auditing

Why Financial Firms Are Investing Now

Although the total payment volume observed in the Keyrock report—about $73 million over a year—is still tiny compared with global payment networks, the trend matters because of the potential scale.

If AI agents begin performing tasks autonomously across the internet—purchasing data, compute, media generation, or cloud services—the number of transactions could grow dramatically. Each agent might perform thousands or millions of purchases per year.

In that scenario, the most valuable infrastructure layer may not be the AI models themselves, but the transaction rails that enable autonomous economic activity.

The companies competing to build that layer—payment networks, crypto platforms, and cloud providers—are effectively racing to define the financial operating system of machine‑to‑machine commerce.

The Bigger Shift: Software Buying From Software

The Keyrock report highlights a fundamental shift in how commerce may work online.

Instead of humans initiating most transactions, autonomous software agents could continuously purchase services from other software systems—paying for data feeds, compute time, model inference, storage, or digital content.

That new economic model favors payment systems that are:

• Programmable
• Instant
• extremely low‑cost
• interoperable across platforms

Stablecoins and internet‑native payment protocols are currently the closest match to those requirements. If agent‑driven commerce continues to grow, the infrastructure being built today could become the foundation for a much larger machine economy.