AI agents drive x402 wallet refills to 89%

Chainalysis found that 89.3% of active x402 wallets first funded in the second quarter of 2026 received another deposit within 30 days. The figure covers wallets that made payments on at least two separate days.

The rate was 70.6% for wallets first funded in the third quarter of 2025. Chainalysis attributed the difference to developers moving from one-time tests to recurring budgets for software that can spend money independently.

In the third quarter of 2026, about two-thirds of newly funded x402 wallets made a payment within an hour of receiving funds. The share was 4% a year earlier. The median time between initial funding and the first payment fell to 18 minutes from 73 days in late 2025.

Coinbase developed x402 in 2025. The payment standard runs on the company’s Base blockchain and lets applications pay for online resources through web requests. An AI agent can request data, computing capacity or another digital service, receive payment instructions and settle the charge in cryptocurrency before accessing the resource.

By July 2026, x402 had processed more than 138 million transactions. More than 87% of payments recorded during the 16 months through September were between $0.0001 and $0.10. Payments between 1 cent and 10 cents were the most common.

The transactions supported services including AI model inference, code review, financial intelligence and data retrieval. AI model inference allows one application to pay another AI system for a generated response. Developers can buy computing resources when needed instead of using a fixed subscription or enterprise contract.

The United States accounted for about 30% of geographically identified transactions between AI agents and service providers, according to Chainalysis. U.S.-linked wallets generated 88.1% of AI model inference payments, 41.8% of payments to agent frameworks and development tools, and 23.2% of transactions involving data and search services.

Will Su, BlackRock’s head of digital assets research, told Chainalysis that the investment firm is monitoring the sector because autonomous software could create continuous demand for small payments during the day.

Chinese-linked wallets showed a different pattern. They accounted for about 90% of geographically attributed payments for security and code-analysis services. Much of that activity went to AgentLISA, a platform that uses artificial intelligence to audit blockchain smart contracts. Chinese wallets represented 0.1% of identified AI model inference transactions.

Chainalysis linked the difference in part to the use of open-weight models such as DeepSeek and Qwen. These models can operate without a separate payment to an external provider for each request.

A July academic study of 136.7 million Base settlements worth $44.1 million found that 21.2% were fictitious, including self-payments and closed loops between wallets. Another 63.78% occurred within wallet clusters connected through funding relationships, which can include related operators and legitimate transactions.

The study identified about $187,861 in payments reaching recognizable services without detected links between buyers and sellers. Its broader estimate put potentially genuine economic activity at as much as $20.26 million. Of 25,163 listed Base resources, 811 had distinct payment recipients, while 249 recipients earned at least $10 during the period examined.