Celestia introduces Voyager blockchain for agent payments
Celestia introduced Voyager, a blockchain architecture it says is designed for hyperscale payments in an agentic internet, and said the system sustained 21 million transactions per second in a benchmark meant to test onchain payments between machines. The announcement, published on October 29, 2026, matters because the company is positioning Voyager as a payments layer for AI agents that need fast, low-cost settlement for content, data and services on demand.
Evidence and context
In the company’s post, Celestia said its first full benchmark showed Voyager processing a sustained 21 million transactions per second across 100 million accounts over five hours. The same post said throughput averaged 21.2 million transactions per second over the run and peaked at 21.6 million, with about 400 billion transfers included in one hour of the benchmark. Celestia said the system kept memory usage at about 48 GB, ran execution on a single CPU core at 96% to 100% utilisation, and moved about 4.8 GB/s through both the journal and the Fibre publisher.
The architecture described by Celestia splits payment processing into separate threads so balance updates are not blocked by other work. Transactions pass through an ingress layer that checks signatures and batches payments before they reach the core node, while a ring-buffer design distributes work across the journal, executor and storage components. Celestia said Voyager is built specifically for payments rather than a general-purpose virtual machine, uses a flat table for state rather than a Merkle tree, and publishes every transaction to Celestia Fibre so others can replay and verify the payments.
The company framed Voyager as part of a broader push toward what it calls the agentic internet, in which machines rather than humans increasingly call APIs, scrape pages and pay for services. Celestia argued that existing payment rails are too slow or too expensive for per-use charges, citing card-network fees, confirmation times and transaction-capacity limits as constraints for micropayments. Those comparisons were presented by Celestia in support of the case for blockchain-based payment rails, not as independently verified market measurements.
There are also material limits to the benchmark as described. The reported test measured a single Voyager node and, according to the company, did not yet compute state roots, replicate data to other machines, or serve real-time transaction-status queries at that volume. Celestia said the system does produce receipts for every payment and that replication, state roots and queries are next. The retrieved coverage therefore supports a high-throughput transaction-processing claim, but not production readiness or end-to-end settlement capacity at the same scale.
Background coverage retrieved the same day also described Voyager as a Celestia Labs product aimed at high-volume payments between AI agents and online services. Celestia said it will publish more technical detail in a forthcoming deep dive and is taking Voyager toward production, but the timing of any launch was not provided.