AI Agents Pay for Services in Crypto via x402 Protocol

AI agents have made over 138 million payments through the x402 protocol on Base, according to a Chainalysis study published on October 14. However, this figure measures the number of transactions, not 138 million dollars: over 87% of the analyzed operations are worth less than ten cents.

In Brief

• Over 138 million payments have been made via the x402 protocol on Base, according to Chainalysis.
• More than 87% of transactions involve amounts less than ten cents, illustrating the rise of micropayments between AI agents.
• The United States dominates payments related to AI models, while China stands out in security services and code analysis.
• Stablecoins facilitate automated transactions, but the high number of operations is not enough to demonstrate massive adoption of AI agents.

Tiny Payments, But Rapid Growth

The x402 protocol allows autonomous software to pay directly for an online service without requiring human validation for each expense. Initially developed by Coinbase, it relies on stablecoins to settle amounts too low for traditional credit card networks.

The key figures from the study highlight the growth of this still experimental economy:

• Over 138 million x402 transactions have been recorded on Base since the protocol's launch;
• More than 87% of payments were between 0.01 cents and 10 cents;
• The most common range was between 1 and 10 cents;
• The median time between funding a wallet and its first payment dropped from 73 days to 18 minutes;
• The share of wallets refilled within 30 days increased from 70.6% to 89.3%.

These operations are primarily used to purchase data, query a programming interface, access an AI model, or have code reviewed. An agent tasked with a mission can thus independently acquire the digital resources necessary for its execution.

Stablecoins present an economic advantage here. A transaction of a few cents would generally be absorbed by the fees of a card payment. On a low-cost blockchain like Base, it can remain viable.

Circle summarizes this distribution of uses with a formula:

Cards for payments triggered by consumers, stablecoins for machine-to-machine settlements too small for the card economy.

The United States Dominates AI Model Purchases

Wallets linked to the United States account for 30% of geolocatable payments made between agents and services on x402. Their dominance is even clearer in a specific category: they concentrate 88.1% of payments intended for using AI models.

The share of wallets linked to China only reaches 0.1% in this category. Chainalysis offers a possible explanation: Chinese developers may rely more on domestic open models, such as DeepSeek or Qwen, while major American models are often charged per use.

The situation, however, is reversed for certain specialized services. Chinese wallets account for nine out of ten payments dedicated to security and code analysis. The bulk of this activity has been directed towards AgentLISA, a tool that uses a language model to audit Solidity smart contracts.

On-chain data therefore does not describe the entirety of the global market. It only covers the x402 transactions that Chainalysis has been able to identify and geolocate. The significant American share does not mean that the United States accounts for 88% of all payments between AI agents.

The Number of Transactions Is Not Enough to Prove Adoption

The growth of payments and the more frequent reloading of wallets suggest that some projects are moving from simple tests to recurring uses. Hoolie Tejwani, head of Coinbase Ventures, thus estimates that "AI agents are becoming economic actors, with each agent effectively constituting a new user of the blockchain."

This conclusion must, however, remain cautious. A transaction recorded as coming from an agent does not necessarily prove that an AI system has independently made a genuine economic decision. Classic scripts, automated tests, or operations between wallets controlled by the same actor can also inflate the statistics.

The cumulative amount transferred would also be a more telling indicator than just the number of operations. With a majority of payments under ten cents, 138 million transactions do not necessarily represent a considerable financial volume.

The most significant change lies rather in the ability to monetize each request. Data, a search, or a few seconds of computation can be sold individually without subscription and without heavy banking infrastructure. Will Su, head of digital asset research at BlackRock, estimates that agents could thus create significant demand for frequent, low-value payments executed continuously.

The next step will be to determine whether these flows correspond to useful and sustainable services. For now, x402 mainly demonstrates that a blockchain can provide autonomous software with a payment method suitable for tiny expenditures.