138 Million Agent Payments, Almost No Money: What the x402 Numbers Measure

Where the 138 million comes from

The figure is from Chainalysis. Its North America crypto adoption report, published October 14, 2026, says more than 138 million x402 transactions had occurred on Base. The report's general reporting period is the 12 months to July 1, 2026, but its x402 review window is the 16 months ending in September 2026, and it does not attach a precise as-of date to the count. Coverage describing the total as reached by the third quarter of 2026 is consistent with the x402 window.

Two things in that report explain the headline. First, size: over the same 16-month window, Chainalysis found that over 87% of x402 payments fall between a hundredth of a cent and a dime, with one to ten cents the most common size. Second, absence: the report gives no dollar total for x402 payments. The count is the number of settlement events, and nothing more.

An earlier Chainalysis post, from June 3, 2026, reported well over 100 million cumulative x402 transactions on Base through the first quarter of 2026.

Why count and value point in different directions

A transaction count weights every payment equally. A one-cent call to a data endpoint and a large purchase each add one. Value does the opposite and is dominated by the few large payments.

Chainalysis's June post shows how far apart the two populations sit. Transactions of $1 or more made up 95% of x402 volume in early 2026, up from 49% in early 2025, while transactions between 10 cents and $1 fell from 46% to 4% of volume. So the count is made mostly of tiny payments, and the money is made mostly of a minority of larger ones.

That is not a flaw in the data. Pay-per-call is supposed to generate many small payments. It does mean a count can grow by orders of magnitude while value barely moves, and a headline built on the count says little about how much money changed hands.

What inflates the count

Speculative mints. Chainalysis describes PING, a pay-to-mint token where a user queried a URL, received a 402, and paid 1 USDC to mint. It processed over 150,000 transactions in its first month, and Base's near-zero gas let users repeat the loop many times. Those wallets went dormant within about a week.

Testing. Chainalysis uses a send-to-self transaction as its proxy for a tester and tracks conversion from tester to payer, which tells you testing is a visible share of activity. The send-to-self proxy is from its June 2026 post; the October report calls agentic commerce an experimentation phase and describes early activity as predominantly faddish.

Self-dealing and wash activity. Artemis Analytics classifies gamed x402 activity in two buckets: self-dealing, where one wallet is on both sides, and wash trading, where the seller funds the buyer, who pays the seller back. In March 2026, according to reports republished by KuCoin and StartupHub, Bloomberg cited x402.org data showing about $24 million in agent payments over 30 days, Allium Labs measured about $3 million for the same window, and an Artemis analyst's filter for self-transactions and fund cycling cut it to about $1.6 million.

Autonomy is assumed. Chainalysis identified x402 transactions by their onchain signature, wallets interacting with the protocol. Neither of its posts claims to verify that each wallet was an autonomous agent rather than a script or a person.

How to read the next headline

Six questions separate a useful number from a marketing one.

Count or value? If only a count is given, the value is probably small.

Cumulative or windowed? A cumulative total only goes up. A 30-day window can fall: CryptoSlate, citing Artemis data, reported adjusted dollar volume down about 77% from a November 2025 peak as of May 2026, and other outlets have reported daily transaction counts down far more from a December 2025 peak. Both are old enough now that you should pull current figures rather than quote these.

Which chains and which facilitators? Chainalysis calls Base the most popular chain for agentic commerce, and the methodology for identifying x402 traffic differs by tracker.

Filtered how? Ask whether self-dealing, wash cycles, mints and send-to-self tests were removed, and by what rule.

Who is reporting? A protocol's own dashboard, an independent analytics firm, and a news summary of either are three different levels of evidence.

Is there repeat behavior? Chainalysis's stickiness data is more telling than its count: the share of wallets funded a second time within 30 days rose from 70.6% for the Q3 2025 cohort to 89.3% for Q2 2026, and the share making a first x402 payment within an hour of funding rose from 4% to 67%. Repeated payment to several counterparties over several days looks like use. A single burst looks like a test or a farm.

What it means if you build on x402

Do not size anything on ecosystem counts. The number that matters is what your own endpoints settled, from how many distinct payers, and how often they came back. That is measurable on your own server.

For agent operators, the small payment size is the design constraint. When a typical payment is a few cents, a per-payment human approval costs more than the purchase. The workable answer is a bound set in advance. CryptoCadet's rail does this on Base: a non-custodial session-key bounded USDC float, held under OS-keychain custody, so an agent can make many small payments without each one needing sign-off and without access to more than the float.