I measured the x402 economy. It's $11,748 a month

A developer's measurement of the x402 economy, a protocol revived by Coinbase from an unused 1997 HTTP status code, found total 30-day gross merchandise value of just $11,748 across 14,128 registered services, with only 287 services (2.0%) receiving at least 100 paid calls. Coinbase's own internal API usage accounts for 58.7% of all-time settlement volume, and roughly half of x402 transactions are artificial, including self-dealing and wash trading, while the top independent operator grosses about $872 per month.

There is a lot of writing about AI agents paying each other for services. The protocol backing most of it — x402, revived by Coinbase from an HTTP status code reserved in 1997 and never used — has real infrastructure, real SDKs, and headline numbers in the hundreds of millions. I wanted to build a business on it, so I went looking for the demand rather than the headlines. What I found changed what I built. Everything below is measured, and the raw data and scripts are linked at the bottom so you can check me. On 7 August 2026 I pulled the complete Coinbase CDP x402 Bazaar discovery catalog — every service registered as discoverable. The Bazaar publishes, per service, the 30-day call count and the number of unique payers. That is an unusually honest thing for a marketplace to expose, and it makes the whole economy countable. Registered discoverable services 14,128 with ≥100 paid calls in 30 days 287 2.0% with ≥1,000 paid calls in 30 days 26 0.18% Total 30-day GMV, every seller combined $11,748 Eleven thousand, seven hundred and forty-eight dollars. That is the entire independent x402 seller economy, worldwide, for a month. The best-performing independent operator grosses about $872/month . Tavily's x402 endpoint — the most widely adopted single service by unique payers — grosses about $554/month . They do, and they aren't lying exactly. They're just measuring something else. Coinbase is 58.7% of all-time settlement volume. That is CDP's own internal API usage, not a third-party marketplace. Roughly half of x402 transactions are artificial — self-dealing, where the same wallet is buyer and seller, and wash trading, where a seller funds the buyer wallet that immediately returns the funds. Much of the 2025 spike was a meme coin. PING's "pay-to-mint" mechanic drove transactions up over 10,000% in a week. Wallet retention then fell from ~87% to ~5%. Strip those out and what remains for someone building a service is the number above. Look at the shape of one well-built service — 25 clean Ethereum RPC endpoints: 641 unique payers → 1,018 total calls 1.6 calls per payer Six hundred and forty-one different agents found it, paid once, and never came back. That is not a customer base. That is a tasting menu. Now compare it to the few services with actual demand: x402.tavily.com 422 payers → 55,372 calls 131 per payer stableenrich/exa 277 payers → 12,085 calls 44 per payer Only 32 services out of 14,128 clear the bar of "≥20 unique payers AND ≥10 calls per payer." That is 0.23% of the catalog. Twenty-two services have 300+ payers but fewer than 2 calls each — pure tourism. If you build here, calls-per-payer is the only metric worth watching. Revenue will flatter you; repeat usage won't. Almost all of them resell premium APIs that agents cannot otherwise buy. The top independent operator proxies People Data Labs, FullEnrich, Exa, Firecrawl, and Clado on a per-call basis. Others resell flight search, or neural search. The actual business being done on x402 is arbitrage on account-creation friction . An autonomous agent has a wallet but no credit card, no legal entity, and no ability to sign a SaaS contract. Selling it per-call access to something that normally requires an account is the one model that demonstrably works. It is also, in most cases, a breach of the upstream provider's terms of service. I ruled it out for that reason and I'd encourage you to as well. I probed 350 registered endpoints directly: 73.4% return HTTP 402 correctly 9.1% 405 — usually POST-only, probably fine 7.4% 404 — dead route 4.3% DNS / TLS failure — service gone 2.9% 200 OK with no paywall — accidentally free 1.1% 500 So about 13% are hard-broken, and another 3% are giving their service away by mistake. To check I wasn't just describing "new marketplace is small", I ran the same measurement on Apify — a mature scraping marketplace where solo developers publicly report $1,000–$10,000/month. I pulled 11,348 actors with 30-day user counts, ratings, and pricing. Two findings were worth the effort: 1. The most attractive-looking gaps are the ones you can't legally take. Sorting by "high demand, weak incumbent" surfaces Instagram at 3.39★ with 9,113 monthly users, LinkedIn at 2.93★, Facebook at 2.33★. Tempting — until you notice the bad ratings are the anti-bot difficulty, and those platforms prohibit scraping. Filtering them removes 356 of 578 high-traction actors. 62% of the visible opportunity, gone. 2. The best legitimate weak incumbents are free. Of the 578 actors with ≥100 monthly users, 15 are free — and they include the poorly-rated ones you'd most want to displace. apify/screenshot-url : 901 users, 95,641 runs, 3.69★, $0. Users tolerate a mediocre free tool. A paid competitor needs to be dramatically better, not marginally better. That second one is invisible unless you check the pricing field. I nearly built against it. x402 is a good protocol with almost no demand behind it yet. Implementing it costs a few days and near-zero ongoing money, so it's cheap optionality. Building a business that depends on x402 revenue in 2026 means competing for a pool smaller than one senior engineer's salary. Measure the marketplace before you build for it. Both of these datasets were a few hours of work against public APIs. Both changed my plan. Watch repeat usage, not revenue. 641 payers and 1,018 calls is a worse business than 44 payers and 12,085 calls, and only one of those looks good in a screenshot. I did build something in the end — an MCP server that tells coding agents what broke between two versions of a dependency, and a job-listing actor that reads ATS APIs instead of scraping HTML. Whether anyone wants either is still an open question, and I'll write that up when I know. But the measuring was the useful part, and nobody else seemed to have published it. Before posting this I re-ran the same collector against the same API on 23 August 2026 , using the identical method first accepts .amount , prices capped at $500 to exclude misconfigured listings . The catalog grew. The economy did not. 7 Aug 23 Aug Registered services 14,128 15,309 +8.4% 30-day GMV $11,748 $10,200 -13.2% Services with real demand 32 16 -50% Two individual services worth naming, because I named them above: Tavily , the most widely adopted single service by unique payers, fell from ~55,000 calls and ~$554/month to~~9,700 calls and ~~$97/month — an 82% decline in sixteen days. stableenrich is now the largest independent operator at roughly $1,100/month across 19 endpoints. So the catalog is growing about 8% a month while the money in it shrinks. If anything the original conclusion was too generous. Both snapshots are in the repo, so you can run the comparison yourself. Everything is reproducible. Raw data, collectors and analysis are in the repo: github.com/dhughes6071/driftwatch https://github.com/dhughes6071/driftwatch — 14,128 services with call counts, unique payers, pricing x402 bazaar slim 2026-08-07.json — the 16-day re-measurement, 15,309 services x402 bazaar slim 2026-08-23.json — 11,348 actors with usage, ratings, pricing model apify store 2026-08-07.json — the collectors, the analysis, and the endpoint probe research/data/ Both catalogs come from public, unauthenticated APIs: api.cdp.coinbase.com/platform/v2/x402/discovery/resources and api.apify.com/v2/store . Figures are snapshots of 7 and 23 August 2026 and will keep drifting. The direction of travel over those sixteen days was down.