Celestia advertises 625 million transactions
a second. We measured 1.69

Over 30,000 consecutive blocks — twenty-three hours and forty-seven minutes of Celestia mainnet, ending at 06:32 UTC on 2 November 2026 — the chain carried 144,615 transactions. That is 1.6895 a second.

Celestia’s homepage, fetched the same morning, advertises “625M+ TPS” as the “Potential throughput of blockchains built on Celestia”, and says the team “built Fibre to accommodate 1 Tb/s of data throughput (625M+ transactions per second)”. The measured figure is smaller by a factor of roughly 370 million.

The gap is the story, but the fair reading of it beats the unfair one.

The traffic is the right kind

Most chains advertising throughput generate most of it themselves. Celestia does not. We decoded 3,005 transactions sampled across 632 blocks in the measured window. Of those, 97.80 per cent were MsgPayForBlobs — the message a rollup sends when it pays to post data. Everything else, transfers and delegations and IBC packets and reward withdrawals together, came to 2.20 per cent.

That is a good result, and the opposite of what the same test does to most chains. The activity here is paying customers doing what the chain exists for.

The problem is how little of it there is. The blobs are small: median 379 bytes, mean 1,394, the largest we saw 165,779 against a documented 8 MB ceiling. Across the window the chain carried 97,249 bytes of blob data per block against an on-chain capacity of 33,554,432 — 32 MiB, read from the consensus parameters on two independent endpoints. Utilisation was 0.2898 per cent.

Fibre shipped, and four transactions used it

Celestia Fibre, the product the 1 Tb/s figure describes, reached mainnet when app version 10 activated at 03:45 UTC on 20 October. It is not vapour: the module answers on-chain.

We counted what has gone through it. Querying three public endpoints independently returned the same number each time: four. Four MsgPayForFibre transactions in the module’s entire life, at heights 15,129,045, 15,129,339, 15,138,464 and 15,380,778, the most recent on 28 October. None appeared anywhere in our sample.

The benchmarks deserve their own sentence. The 1 Tb/s figure came from a partial test of the networking layer across 498 Google Cloud machines in North America; only the later 3.07 Tb/s run, on 120 AWS machines, was end-to-end. Celestia’s own write-up of it is unusually candid: “Several settings were specific to this benchmark and will not be available in the initial mainnet release. We used 2 GiB blobs (mainnet will start at 128 MiB), 1-second blocks, and a limit of 2,000 Fibre submissions per block, up from the current 200.” The run came from “an experimental performance branch”.

The documentation and the homepage no longer lead with the same product. docs.celestia.org still opens with “a modular data availability network” and carries no throughput figure at all. The homepage still calls Celestia a “modular blockchain” in its meta description, but the copy a visitor actually reads now sells millisecond latency and agentic payments, and the word blob appears on it nowhere.

What it costs to run

Chain fees were $86 over 24 hours and $35,553 over the trailing year. Issuance, measured from timestamped supply snapshots thirty and fifty days apart, runs at 2.27 per cent a year — about 26.8 million TIA, roughly $12.4m at today’s price. Fees cover 0.29 per cent of that. DefiLlama reports protocol revenue of zero.

There is nothing else on the chain. Celestia appears in none of DefiLlama’s 420-chain TVL list, its 209-chain stablecoin list or its 297-chain DEX list. The largest genuine on-chain protocol is Stride liquid staking at about $292,000, against a token worth a great deal more.

The bull case is real and worth stating. Liveness is excellent: across 30,000 consecutive blocks there were no missing heights, the largest gap between two blocks was 8.671 seconds and nothing exceeded thirty. Block time has fallen from 11.9 seconds at launch to the 2.853 we measured, across nine upgrades verified from on-chain activation heights. Inflation has been cut twice by governance, from 8 per cent to 2.27, on dates matching those heights exactly. And at 0.29 per cent utilisation demand could grow roughly 345-fold before capacity binds.

The verdict: a competently run chain with very little on it, selling a datacentre benchmark as a property of the network. The engineering is real, the demand is not there yet, and the homepage now leads with something else entirely.

Throughput, block times, transaction decoding, validator set and supply measured directly from three independent Celestia RPC endpoints (rpc.celestia.pops.one, public-celestia-rpc.numia.xyz, celestia-rpc.publicnode.com) between 06:29 and 08:21 UTC on 2 November 2026; the MsgPayForFibre count was re-verified on all three at 08:20. Capacity from the chain’s own blob and consensus parameters. Marketing copy quoted from celestia.org and docs.celestia.org fetched the same morning; benchmark quotations from blog.celestia.org posts of 13 January and 1 October 2026. Price and market cap from CoinGecko at 08:20 UTC; fees, TVL, stablecoin and DEX figures from DefiLlama the same morning. Contested: DefiLlama’s two fee endpoints disagree ($2,822 against $468 over 30 days) because one counts the chain’s fees and the other protocols deployed on it; CoinGecko’s circulating supply cannot be verified on-chain; the blog dates Fibre’s launch to 19 October where the chain dates the upgrade to 20 October. The 97.80% figure is a sample. Not a recommendation to buy, sell or hold anything.