The whole x402 Bazaar earned between $9,491 and $31,847 last month across 15,174 listings. One gift-card checkout is 98% of that uncertainty.

What I measured

On 2026-08-20 Ramp announced that corporate agents can pay for x402-priced services from a Ramp-provisioned Solana USDC wallet, with ledger attribution and an audit trail. Their public skill file ramp-make-x402-payment tells the agent to discover services at Coinbase's resource endpoint and to accept an offer only if it matches an exact filter: scheme exact, Solana mainnet, the canonical USDC mint, a positive amount, and a feePayer.

That filter is a precise definition of "what a Ramp agent will actually be offered." So I ran it against the entire population.

Full crawl. I paginated api.cdp.coinbase.com/platform/v2/x402/discovery/resources to exhaustion: 15,160 listings (the total grew by 2 during the four-minute crawl). 5,580 of them — 36.8% — pass Ramp's filter.

The menu is enormous and the kitchen is tiny

Those 5,580 payable listings come from 148 domains and 193 receiving wallets.

The largest single operator has 961 listings behind one wallet. Second has 946. These are not 5,580 merchants. They are a few dozen operators publishing near-duplicate endpoints at scale.

But the listings are alive — that part surprised me

My prior was link rot: a directory of dead spam. It isn't. I drew a stratified random sample of 150 payable listings (60 from the top-5 domains, 90 from the long tail) and requested each one three separate times, identically.

Then I read the sellers' wallets

Discovery listings publish their payTo address, which makes the seller side of this economy directly readable on Solana. I took the eight wallets holding the most payable listings — 3,127 listings, 56.0% of everything a Ramp agent can pay — and measured 30 days of inbound USDC to each.

Method: resolve each wallet's USDC associated token account, paginate getSignaturesForAddress back 30 days counting only successful transactions, draw a random sample of up to 35 per wallet, and compute the inbound amount two independent ways — the owner's pre/post token-balance delta, and the parsed transfer instructions targeting that token account. The two paths agreed on 226 of 226 sampled transactions. Per-wallet totals are the sampled mean times the transaction count.

One day later: what the whole catalog earns

Caveat 2 below says it plainly: reading payTo token accounts on Solana cannot see an operator earning on another chain. So on 2026-08-21 I measured the same economy from the opposite end.

Every Bazaar listing carries a quality object with its call count and unique payer count for the last 30 days — the platform's own accounting, across every network it indexes. I pulled all of it (15,174 listings requested, 15,174 returned) and priced each listing's calls at its cheapest payment option.

Why a range and not a number — and why the range is one merchant. 14,700 of the 15,174 listings price themselves purely with the exact scheme, where the advertised amount is the amount charged; those add up to $9,419. A further 241 listings use other fixed-price schemes (batch-settlement, exact_cosmos_authz, onchain and similar) worth $72 — which is why the floor of the range above is $9,491 rather than $9,419; all 14,941 listings without an upto accept total to that figure. The remaining $22,356 comes from the 233 listings carrying an upto accept — a variable-price scheme in which, per Coinbase's own specification, the advertised amount is the maximum the client authorizes and the settled amount “MAY be 0.” A catalog crawl cannot see what was actually settled. So the honest whole-catalog figure is a range: $9,419 + $72 + $0 = $9,491 if every variable-price call settled at zero, $9,419 + $72 + $22,356 = $31,847 if every one settled at its ceiling.

98% of that uncertainty is a single gift-card checkout. api.bitrefill.com/x402/invoice/pay took 22 paid calls from 4 unique payers, each authorized upto $1,000 — so at most $22,000, and $22,000 of the $22,356 in variable-price ceilings across the whole catalog. Remove it and the other 15,173 listings share $9,847 a month at their own ceilings. Note what this means even at the pessimistic end: an average settled ticket above roughly $56 would still make Bitrefill the largest single merchant on this rail, ahead of stableenrich.dev at $1,233. The largest revenue anyone can point to here is still an agent buying goods, not an agent buying data.

Correction, 2026-08-30. This paragraph originally ended “nobody outside Bitrefill can tell from public data which side of $56 it lands on.” You can, and I have. Bitrefill's settled x402 receipts over the trailing thirty days, read from x402scan's facilitator index on 2026-08-30: $774.14 across 634 paid calls from 109 unique buyers — an average ticket of $1.22, about 27× below the advertised upto ceiling and far below $56. Note these are two different counters over two different windows: the catalog's own quality field reported 22 calls from 4 payers for that one endpoint on 08-21, while the settled figure covers every Bitrefill receiving address the facilitators indexed. The average ticket is what answers the question, and it lands on the low side — exactly the direction the upto spec warned about. Caveat 8.

77% of listings have exactly one payer. 11,702 of 15,174 saw a single unique payer in 30 days. Getting indexed requires a settled paid call, and a listing that goes 30 days without a settlement is dropped from the catalog — so a large share of this traffic is sellers paying themselves to get listed and stay listed. One endpoint is named /x402/bazaar-keepalive. It has one call, from one payer, at $190.

The median receiving address earns $0.15 a month. Across 1,463 addresses: median $0.15, p90 $6.50. Twenty-nine clear $100 in thirty days. Four clear $1,000.

One methods note, because it nearly fooled me: joining the two datasets required a case-sensitive match. Lowercasing the payTo addresses — harmless for EVM, fatal for base58 — returned a clean zero for all eight wallets, which looks exactly like two instruments disagreeing rather than like a broken join.

Why this matters more than a settlement count

x402's headline metric is settlements, and settlement counts are enormous. A population-scale study of x402 on Base (arXiv:2607.12575, July 2026) measured 136,708,672 settlements worth $44.1M over 280 days — and found 21.20% fictitious and 63.78% internal settlement inside a linked cluster. Their lower bound for value that demonstrably reaches a nameable independent service: $187,861 over 280 days, about $670 a day for the whole Base ecosystem.

That paper measured Base in depth and Solana coarsely. This is the Solana seller side, on exactly the filter Ramp's agents use, and it points the same way: the supply side is live, enormous, hyper-concentrated, and earning almost nothing.

Two practical consequences.

If you are building on x402: publishing another endpoint into this directory is not a distribution strategy. You would be listing number 15,175 against operators who already have a thousand each and still make less than a coffee a day. The scarce thing here is demand, and no amount of supply manufactures it.

If you are funding an agent wallet: "browse other x402 services" reads like a marketplace with 15,160 vendors. It is a few dozen operators with thousands of near-duplicate endpoints. Discovery is explicitly not an endorsement — Ramp's own skill says so — but the gap between how that menu reads and what it is sits directly underneath the attribution and audit-trail promise. Quality and provenance signal on the merchant side is, as of today, the only unserved layer in this stack.

Caveats, stated plainly

• 1Sampling. The 30-day totals are extrapolations from up to 35 randomly sampled transactions per wallet, not a full sum of every transfer. Small-sample wallets carry wide error.
• 2Scope. I measured inbound USDC to the exact payTo token account published in the listing. An operator earning elsewhere — another chain, another wallet, off-chain invoices — is invisible to this method, by construction.
• 3Not every inbound transfer is a payment. I counted inbound USDC, not verified x402 settlements. If anything this inflates the numbers, which is the direction that makes the conclusion safer.
• 4One day. This is a single-day census, 2026-08-20, extended 2026-08-21. It is a snapshot, and the population is growing.
• 5The catalog numbers are platform-reported. Call and payer counts for the 08-21 section come from the Bazaar's own quality field, not from chain data. I priced each listing's calls at its cheapest advertised option, which understates any seller whose buyers paid a dearer one.
• 6Decimals are not uniform. Stablecoins are 6-decimal on most of these networks but 18-decimal on BNB Chain. Pricing all of them at 6 produced one listing that appeared to earn $430 billion, which is how I found the bug. That row is CoinMarketCap at a real price of $0.01.
• 7Advertised price is not settled price under upto. Added as a correction on 2026-08-23. Caveat 5 named only the direction in which cheapest-option pricing understates. It missed the opposite direction: 233 listings carry an upto accept, where the published amount is a maximum authorization and the settled amount may be anything at or below it, including zero. As first published on 2026-08-21 this page said Bitrefill's checkout “took 22 calls at $1,000 each — $22,000” and led with $31,847 as a point estimate. Both were ceilings stated as measurements. The whole-catalog figure is now given as a range, and the conclusion is unchanged in the direction that matters: correcting this makes the economy smaller, not larger.
• 8The catalog is not the economy — and the settled numbers now exist. Added as a correction on 2026-08-30. This page overstated its own scope: $9,491–$31,847 is the Coinbase Bazaar catalog, not "the entire public x402 seller economy." Measured on x402scan's facilitator index (unauthenticated, tRPC public.stats.overall and public.sellers.bazaar.list, 30-day timeframe, read 2026-08-30): settled x402 volume across 18 indexed facilitators is $1,369,049 over 19,578,477 transactions, 19,376 unique sellers and 21,425 unique buyers. The catalog is roughly 2% of that — a ~46× error in the sentence, not in the conclusion. Three controls: the 18 facilitator rows sum to the reported total within 0.01%; three independent surfaces agree within 0.7% at the one-day timeframe; and the 875 Bazaar merchants resolvable in the seller table sum to $898,871, 65.7% of the total, with zero address overlap after collapsing nested recipient sets. What the settled data does not overturn is the finding: the median Bazaar merchant settles $0.40 a month, p90 excluding the largest is $21.80, the top five merchants take 91.6% of it, and the arXiv figures above — 21.2% fictitious, 63.8% intra-cluster — apply to this volume as well. One caveat on the caveat: I could not independently re-derive these dollars on-chain, because every public Base RPC I tried refused a thirty-day eth_getLogs range. They are x402scan-indexed figures, corroborated three ways internally and once externally against a large merchant's own published call count.

Sources

• Coinbase CDP x402 discovery API (crawled in full, 2026-08-20)
• ramp-public/skills — Ramp's public agent skills, incl. ramp-make-x402-payment (MIT)
• arXiv:2607.12575 — How Agentic Is Agentic Commerce? Population-scale measurement of x402 (Ling, Zhou, Wu, Wang; July 2026)
• x402 protocol