The x402 ecosystem has a number problem — but not the one you think. The headlines report cumulative protocol volume: Coinbase says 14 million agent payments in 30 days, Cloudflare 160 million transactions, Solana's foundation hundreds of millions. Those are real numbers about the rails.
Nobody had measured the other side: what does the average listing actually earn? So I pulled Coinbase's public x402 discovery index in full — 15,058 listings across 1,602 distinct hosts — and computed the usage distribution at the host level.
The result: the median listing gets roughly one call in thirty days — not one payment, one call. A host with three hundred listings is as "unused" as a host with one.
Listings concentrate. The top 10 hosts hold 27.6% of all listings; the top 50 hold 49.6%. And 751 of 1,602 hosts — nearly half — list exactly one endpoint.
Usage concentrates far harder. Total 30-day call volume across the whole index is roughly 312,000 calls. Two hosts alone — an oracle provider with 3 listings, and a social-data host with 15 — account for ~104,000 calls, a third of the entire index. A single search endpoint carries 12,500 calls by itself.
| Host type | Listings | 30-day calls |
|---|---|---|
| Oracle data (3 endpoints) | 3 | 52,316 |
| Fresh social data (15 endpoints) | 15 | 51,974 |
| Search (1 endpoint) | 1 | 12,523 |
| Enrichment (15 endpoints) | 15 | 26,951 |
| Median host, everywhere else | — | 1 |
The demand head is specific and boring in the best way: oracle data, fresh social data, search, enrichment. Things an agent needs right now, priced in cents, available at request time. Everything else is the long tail.
"14 million payments" and "median one call" are the same ecosystem seen from two altitudes. The cumulative figures are dominated by a small number of high-throughput endpoints. The long tail — thousands of listings across hundreds of hosts — is effectively dormant.
This matters for three reasons:
1. The discovery layer's liveness problem is self-inflicted. If an index ages out listings by call-count, it can't tell "unused" from "unpopular" — because most of what survives is also unused. A median of one call on both sides of the churn line means call-count is the wrong health signal.
2. The growth problem is not "add more listings." It's "get listings into the demand head." The index already has fifteen thousand endpoints; it does not need a sixteen-thousandth novelty tool. It needs more of the few data types agents actually pay for.
3. An honest dashboard is a differentiator. A dashboard that shows the distribution — median one, a third of volume from two hosts — tells a builder where the money is. A dashboard that shows cumulative protocol volume only tells them the rails exist.
Honest framing beats inflated framing. The builders who publish the long-tail reality are the ones a new seller should actually trust — because when they do show growth, it will mean something.