deNodes

Introducing SnapsVault, an open snapshot marketplace for node operators Every snapshot from every provider in one place and one format, ready to compare and verify before you download. Now live for @celestia, with 8 contributors providing snapshots for mainnet and testnets. snapsvault.xyz

21 million payments a second, with execution running on a single CPU core. That's the first full benchmark of Voyager, a new payments architecture from @Celestia It's built for AI agents, which create a new kind of demand - they pay per request, for every API call, page or piece of data. Those payments are tiny, need to clear instantly and come in huge volumes. Card rails weren't designed for that, with ~30-cent fees, 1-3 second confirmations and Visa at ~65k TPS. What makes it fast: - Signature checks run on separate ingress machines, so verification scales by adding hardware - Stages work on different batches in parallel through an LMAX-style ring buffer, and the executor prefetches the state it needs - No VM and no Merkle tree: transfer logic is native and balances live in a flat table In the benchmark, Celestia Labs ran one Voyager node for 5 hours on random transfers between 100 million accounts. About 400 billion transfers went through at 21.2M TPS on average, peaking at 21.6M, with memory flat at 48 GB. The average sitting that close to the peak means throughput held steady for the whole run, and memory didn't grow with load. There are no state roots or replication yet, so this is a ceiling for execution rather than a live network figure. As a starting point, it's well past anything payments run on today. Every transaction is published to Fibre. During the run, the publisher moved ~4.8 GB/s, about 32x Fibre's 150 MB/s launch setting. Voyager chains pay for that blockspace in TIA, so payment volume on Voyager turns directly into Fibre usage. It's the first workload that would push Fibre well past its launch capacity. Voyager is also the first product to come out of the Sovereign Labs (@sovereignxyz) acquisition in July. With Fibre only recently live on mainnet, Celestia now pairs its DA layer with an execution design built to fill it. It's a big step toward Celestia's goal of bringing every market onchain and powering agentic payments, with rails built for the scale agents need. We're keen to see how quickly Voyager moves toward production.

Fibre is live on @Celestia mainnet, and the more interesting part is the encoding underneath ZODA, the new data encoding scheme behind Fibre, processes data 881x faster than KZG-based designs. Because of that speedup, Fibre launches at 150 MB/s, which is roughly 800k TPS and 15x Celestia's previous capacity. The same protocol has been benchmarked at 3 Tb/s. So the launch figure is a starting setting, and capacity can be raised as usage grows without redesigning the protocol. Two years of research went into this release. We're glad to keep validating Celestia through it 🟣

Proud to have supported @EspressoSys from testnet to mainnet as a genesis validator From running infra to building tooling around the network, it's been great to see how far Espresso has come - with the latest upgrades bringing subsecond finality and 100+ operators now helping decentralize the network. Still early. Lots more ahead ☕️