Payment Channels: 1 Million Payments Per Second
AI will not wait at a checkout screen. The future of internet payments needs to be built for unprecedented scale. But agentic payments today have a friction problem. Every call an agent makes needs a signature approval and settlement. That works fine for one payment. But it gets more complicated when agents are making hundreds of small paid calls in a loop, and human approvals slow down the completion of tasks.
Payment channels are a new Solana primitive built to solve this across x402 and MPP payment protocols. It lets an agent authorize a spending limit once, spend against the limit, and settle the payment amount once. Think of it like a bar tab or a prepaid meter: you put money down up front, run up usage without paying for each transaction, and settle payment once when you leave.
Where per-call payments break
While pay-per-request models simplify paid tool use, certain kinds of agentic spending patterns are a bad fit for settling every request onchain:
- Single calls with unknown costs: certain requests have variable costs. For example an LLM completion billed per token, or a job billed per byte or per second of compute.
- Many small deliveries: a token-by-token streaming response, or a burst of hundreds of cheap calls in a short time period.
A payment channel amortizes the onchain work down to one open and one settle, no matter how much metering happens in between.
How it works under the hood
- Open: deposit a ceiling amount into onchain escrow. Funds stay non-custodial, held by the program, not the operator.
- Meter off-chain: the agent authorizes spending with signed messages, not transactions. Every call is a signature and incurs no transaction costs.
- Settle: the actual amount consumed is recorded onchain in a single transaction.
- Distribute: funds distributed to recipient(s), any unused deposit returns to the primary wallet.
Why this matters
The per-call model breaks agent autonomy and economics:
- A human ends up back in the loop, approving payments one at a time.
- Every payment carries settlement cost and latency. In a world with fully autonomous agent spend where millions of transactions need to be settled, costs add up fast.
- The workarounds are custodial. Prepaid credits mean your balance is gone on the first call and tracked in someone else's database.
Payment channels remove all three. You set a cap. Your agent spends against it. Settlement happens in batches. Funds stay non-custodial the entire time.
Benchmark
In testing, we ran 100,000 unique wallets through a payment channel proxy, issuing more than 1 million payments every second. That is enough capacity to support more than 80 billion payments in 24 hours, at an estimated processing cost of $0.000000000776 per payment. To try it yourself, deploy the open-source template and reproduce the benchmark using the included load-testing tools.
Built open, works everywhere
Payment channels map cleanly onto the open agentic payment standards live on Solana today.
- x402 upto: authorize a ceiling for a single metered call. Your handler runs, the operator settles the one actual amount, and refunds the difference.
- x402 batch-settlement: many deliveries, settled together.
- MPP session: open a channel and stream many metered deliveries, each authorized by a running cumulative voucher, and settle once when the session idle-closes.
The spec is public and the program is open source. To get started with payment channels:
- Program: github.com/solana-foundation/payment-channels
- Concept + docs: pay.sh/docs/building-with-pay/payment-channels/concept
- Toolkit + playground: github.com/solana-foundation/pay-kit
- Benchmark: https://solana.com/developers/templates/pay-high-throughput-proxy
- Spec: paymentauth.org/draft-solana-session-00.html
Live at launch with Alibaba Cloud
Inference has been one of the most in demand use cases across agentic payment protocols. We are excited to be launching payment channels with @alibaba_cloud. Their API endpoints are live today with Payment Channels. Your agent authorizes once and consumes these APIs at scale. No account setup, no per-call approvals.
Live at launch with Alibaba Cloud
Inference has been one of the most in-demand use cases across agentic payment protocols. We are excited to be launching Payment Channels with Alibaba Cloud, and their API endpoints are live today. Your agent authorizes once and consumes these APIs at scale. No account setup, no per-call approvals.