Artificial intelligence agents are beginning to operate as independent digital customers, paying for computing power, data and software services through cryptocurrency networks. More than 138 million payments have been recorded through Coinbase’s x402 protocol on the Base blockchain, according to new research from blockchain analytics firm Chainalysis.
The findings point to an emerging form of digital commerce in which software can purchase services automatically without requiring a person to approve every individual transaction. The United States dominates identified spending on AI model access, while China-linked wallets show a different pattern, concentrated in security and code-analysis services.
US dominates AI model spending
US-linked wallets accounted for 88.1% of geographically identified payments for AI model inference, compared with just 0.1% for China-linked wallets, Chainalysis reported.
The figures do not represent the entire global AI market. They cover identifiable activity through the payment infrastructure analysed, so they should not be interpreted as a complete comparison of AI spending across the two countries.
One possible explanation for the difference is that Chinese developers make greater use of open-weight AI models, including DeepSeek and Qwen, which can reduce reliance on metered, paid inference services. Chainalysis researchers have cautioned against treating this explanation as definitive.
China-linked wallets, meanwhile, accounted for the majority of identified payments for security and code-analysis services, including activity associated with AI-powered smart-contract auditing.
Micropayments fuel the model
The x402 protocol, incubated by Coinbase, allows software to pay automatically for digital services using cryptocurrency. AI agents can use it to buy data, access models or acquire computing resources as needed, without going through a conventional checkout process each time.
More than 87% of the analysed payments were worth between $0.0001 and $0.10. These small transactions are particularly suited to stablecoins, whose blockchain-based settlement can support payments worth fractions of a dollar.
Traditional card networks are generally less economical for such tiny transactions because processing fees can exceed the payment itself. Crypto-based rails could therefore create a new commercial model in which digital services charge for individual queries, data requests or small computing tasks.
The scale of activity has grown rapidly. Payments through x402 increased from roughly 1,500 in mid-2025 to more than 138 million by the third quarter of 2026. In July, the protocol processed around 75 million payments worth approximately $24 million over a 30-day period.
However, a high transaction count does not necessarily mean that AI agents are generating equivalent economic value.
Payments companies move in
Major financial and technology companies are also developing infrastructure for automated payments. Visa, Mastercard and Stripe have backed initiatives related to agent payments, while Coinbase has expanded tools that allow businesses to accept transactions initiated by AI agents.
The opportunity extends beyond cryptocurrency trading. Autonomous systems could eventually pay for cloud computing, research data, software subscriptions, cybersecurity checks and other digital services around the clock.
Stablecoins could be useful because they combine programmable blockchain transfers with a value linked to a conventional currency. But adoption will depend on more than transaction speed: businesses will also need reliable identity controls, spending limits, security safeguards and clear responsibility for errors or fraudulent payments.
Questions remain over real demand
The emerging market faces an important measurement challenge. Some activity may come from automated scripts, testing, repeated transfers or transactions between related wallets rather than genuine purchases by independent AI agents.
Earlier blockchain research has raised concerns that raw payment counts can exaggerate the level of commercial adoption. The Chainalysis figures demonstrate activity on a particular payment protocol, but do not prove that every transaction represents an autonomous agent buying a service from an unrelated provider.
For the crypto industry, the longer-term opportunity is potentially significant. AI agents could become a new category of blockchain users, generating frequent, low-value payments that are difficult to process economically through traditional systems.
Whether that opportunity becomes a durable business will depend on real customer demand, recurring revenue and secure execution—not simply the number of transactions recorded on-chain.
Sources
- CoinDesk — AI Agents Make 138 Million Crypto Payments as U.S. Dominates Spending on AI Models
- Yahoo Finance — AI Agents Made 138 Million Crypto Payments in a Year. Most Were Tiny
- Decrypt — AI Agents Made 138 Million Crypto Payments in One Year
- Coinbase Developer Platform — x402 Payment Protocol
