The machine-payments conversation has spent most of 2026 talking about rails. HTTP 402 is back, x402 is under the Linux Foundation with roughly forty members, a second standard — the Machine Payments Protocol, from Stripe, Tempo and Visa — has appeared, and everyone from Cloudflare to Coinbase is shipping wallet infrastructure for agents. The rails are real.
Rails, though, are not the same thing as a working procurement loop. A buyer that can pay is not yet a buyer that can find, quote, fund and fulfill across the network. New cross-track numbers make that gap concrete.
Two independent scanners — x402scan and mppscan — track active buyer wallets on the two rails over a rolling 30-day window. As of September 3, 2026, as reported by BroadChain and CoinQM:
| Rail | Active buyer wallets (30d) |
|---|---|
| x402 (Base) | 19,472 |
| MPP (Tempo) | 65,591 |
| On both | 365 |
Only 365 wallets appear on both tracks. That is under 0.6% of MPP's buyers and about 2% of x402's Base buyers. Only 112 wallets completed ten or more transactions on both, and many of those were dual-track aggregators paying from the same key.
Translation: each rail is accumulating its own buyer base, and buyers are not moving between them. An agent that knows how to pay on one track is, by and large, invisible to the other.
The rails were supposed to make agent commerce a single graph — any agent pays any service, one signature, no account. What the numbers show instead is two parallel graphs that barely touch. A service listed on one track is effectively undiscoverable to the majority of wallets that live on the other.
That is not a payment problem. It is a discovery and fulfillment problem. The missing piece is the layer that sits above the rails and is deliberately rail-agnostic: a place where an agent can find a service regardless of which track it settles on, see what it costs and what it actually delivers, and be verified before money moves — so a buyer doesn't have to already be a citizen of a particular rail to buy.
The honest caveat. This market is still tiny. Our own on-chain settlement — one marketplace among many — sits at roughly 380 paid calls and about $40 of USDC since launch. The fragmentation numbers matter because they describe the shape of the early market, not its scale: buyers are splitting into silos faster than a cross-track loop is forming.
If you're building an agent that pays for things, the practical question is not which rail is winning — it's whether your agent can buy across both without integrating two stacks. That is the loop the numbers say is still open: discovery that spans tracks, quotes that are honest, funding that works, and fulfillment that verifies.
The rails got built. The part an agent actually needs — finding and trusting a paid service across all of them — is the next thing that has to work.