推文

@LiaoMojo · 2026-10-16 02:30

The agent economy is now operational, with the ledger predominantly consisting of minor rental fees. According to a Chainalysis report released this week, AI agents executed over 138 million cryptocurrency payments in the past year. While this suggests the emergence of a machine economy, a closer examination reveals that over 87% of these transactions were valued between a hundredth of a cent and 10 cents. Consequently, the market is open, but the financial records primarily reflect low-value recurring charges. The majority of this activity is facilitated by x402, a payment protocol incubated by Coinbase and settled on the Base blockchain. Agents autonomously purchase data, compute resources, and AI model inference using cryptocurrency, without requiring human approval for each transaction. In mid-2025, the protocol processed approximately 1,500 payments; by the third quarter of 2026, this volume had increased to more than 138 million. Transaction values remain minimal. In July, x402 processed approximately 75 million payments totaling just $24 million over a 30-day period. Traditional card networks cannot economically handle transactions of this magnitude, whereas stablecoins can. This capability has led Visa, Mastercard, and Stripe to support x402, and Coinbase now enables business clients to accept payments from AI agents. It is essential to avoid misinterpreting the 138 million transaction volume. Two velocity metrics provide a more accurate assessment. First, regarding wallet maturation, the average time from initial funding to the first x402 payment was 73 days in late 2025. By the third quarter of 2026, this duration had decreased to 18 minutes. This indicates that agents transitioned from merely possessing wallets to actively utilizing them for spending within less than a year. Second, regarding repeat usage, 89.3% of wallets initially funded in the second quarter of 2026 received additional funding within 30 days, an increase from 70.6% for wallets funded in the third quarter of 2025. The rising refill rates suggest a developing habit rather than a one-off experiment. A counterintuitive geographic distribution is also evident. US-linked wallets accounted for 88.1% of identifiable AI model inference payments, while China’s share was 0.1%. Conversely, Chinese-linked wallets constituted approximately 90% of payments for security and code-analysis services, with a significant portion directed to AgentLisa, an AI smart-contract auditing platform. Eric Jardine, Chainalysis research head, explains that Chinese developers prefer open-weight models such as DeepSeek and Qwen, resulting in lower demand for metered inference. He advises against interpreting this disparity as evidence that open-source AI is diminishing Asia’s on-chain market. These metrics do not assess the quality of agents but rather identify who is paying for specific services. The x402 ledger reflects developer payment habits rather than providing an AI power ranking. Yat Siu of Animoca Brands recently estimated that up to 100 billion AI agents could be transacting online within a few years, with crypto’s next billion users potentially being agents paying in stablecoins. While this figure may appear promotional, the trajectory seems plausible. As every API call and data request can settle independently for fractions of a cent, the definition of a “user” must be revised; every agent represents a new on-chain user. The first page of the machine economy’s ledger is written, primarily consisting of rental fees. However, the infrastructure is established, and momentum is increasing.

曝光 36 · 评论 2 · 点赞 2 · 书签 0