Solana launched payment channels aimed at AI agents making small, high-frequency payments, and the headline number traveled fast: a test that pushed one million payment requests per second through a proxy, at a claimed cost near nothing. What got lost in the headline is the more durable change — a second settlement model arriving alongside the per-call model x402 has used so far.
The durable news: agent payments now have two settlement models, not one. Per-call settlement — every request triggers its own on-chain settlement, the model x402 built on — and session-based net settlement — approve a spending limit once, sign messages off-chain, and settle only the final net amount. Solana's channels ship both x402 and MPP through the latter.
| Model | How the money moves | Best for |
|---|---|---|
| Per-call (x402) | Each 402 Payment Required challenge is paid individually; a receipt is bound to each settled call. | One-off, verifiable, cross-party calls — pay for exactly what you used, with a receipt per call. |
| Session / channel (MPP, Solana channels) | Approve a limit once; signed off-chain messages accumulate; only the net amount settles on-chain at the end. | High-frequency, same-party, low-value traffic — a "digital bar tab" where per-call on-chain fees would dominate. |
The bar-tab analogy is apt. A per-call settlement for every one of a million tiny requests means a million on-chain settlements — latency and fee overhead every time. A channel collapses that to one opening, one closing, and a net number. The tradeoff is granularity of proof: per-call gives you a receipt per call; a channel gives you a net receipt at the end.
The one-million-per-second figure deserves the same skepticism any benchmark gets. It was a controlled test through a proxy — not current mainnet throughput — and it measures how fast signed messages can be relayed, not how fast real economic value is moving. The useful question is not "how many requests per second can the channel relay" but "how much real money is actually settling, and between whom."
That question has an uncomfortable answer across the whole x402 ecosystem right now. Independent analysis has flagged that a large share of x402 activity — roughly half in some estimates — looks like self-dealing or wash trading rather than genuine buyer-to-seller demand. Reported daily settled volume sits in the tens of thousands of dollars, not the billions the transaction counts imply. Most individual payments are under $0.50, often sub-cent.
This is not a reason to dismiss the infrastructure. It is a reason to separate infrastructure maturity from economic demand. The channels, the wallets, the Linux Foundation governance that now shepherds the x402 spec with roughly forty member organizations — all of that is real and moving fast. The economic demand is real too, but it is still small, concentrated in cheap data and security micro-endpoints, and easy to overstate with raw transaction counts.
None of this is a reason to wait. The rails are consolidating faster than the demand side is forming, which is exactly the window where a service that is honest about real usage, and priced where the money actually clears, has room to grow.