Algorand's homepage advertises "10,000+ transactions per second", and under its finality heading goes further: "In practice, over 10,000 transactions per second can be finalized the moment they enter a block." Over the 250,000 consecutive rounds ending at 18:44 UTC today, the chain settled 9,725,071 transactions in 691,269 seconds. That is 14.07 a second.
The window matters, so here it is: round 65,890,951 at 18:43 UTC on 11 October to round 66,140,951 at 18:44 UTC today, eight days exactly. A narrower 50,000 rounds gives 11.38. A wider million, back to 17 September, gives 13.23. The figure moves a little depending on where you cut it. It does not move by three orders of magnitude.
This is not sampling. Algorand writes a cumulative transaction counter into every block header, so the measurement is one counter subtracted from another over one timestamp subtracted from another. A second independent node returned an identical delta, which is what you would hope for and not always what you get.
Most of the traffic is not people
That counter is generous about what a transaction is. Walking every nested call in 200 blocks drawn at random across the window, 29.86 per cent of the counted total was generated by contracts invoking other contracts rather than submitted by anybody.
A second random sample, this one of 500 blocks and 13,265 top-level transactions, breaks down as 42.90 per cent asset transfers, 41.80 per cent application calls, 13.84 per cent plain payments and the rest asset configuration. Consensus housekeeping barely registered: one key registration and one state proof in the whole sample.
Then the asset identifiers. The three busiest were 127746157, 127745593 and 127746786 — Blitz, Bullet and Rapid chess rating points. Between them, 27.24 per cent of every top-level transaction; 29.05 with World Chess Progress Points added. USDC, the chain's largest stablecoin, managed 3.32 per cent. One application, created four days before the window closed, took 18.94 per cent alone.
None of that is illegitimate, and writing chess ratings to a public ledger is a real use of one. But it is worth being clear what the throughput number counts. The chain is not busy with finance.
What the money says
DefiLlama puts total value locked at $33.2m. Every decentralised exchange on the chain traded $302,892 in the last 24 hours, 54 per cent of it on Tinyman. Chain-wide fees over the same day came to $6,274, and stablecoins total $91.4m.
ALGO trades at 8.16p, for a market capitalisation of £740.6m and a fully diluted value of £816.5m against the 10bn cap. The ledger reports 9,813,587,411 ALGO in existence; CoinGecko counts 9,073,302,566 as circulating, the 740m difference being foundation stock it does not treat as free float. Of the total, 20.07 per cent is registered online for consensus. Four-fifths of the supply is not securing anything.
Fourteen transactions a second is 0.14 per cent of the advertised figure. The block times, meanwhile, held within 2 per cent of their mean for eight days.
The claim that does hold up
Alongside the throughput line, the homepage advertises "0 downtime in 7 years (and counting)". Eight days cannot verify seven years, but the period we measured is a good advertisement for the rest. Mean block time across the 250,000 rounds was 2.765 seconds. Cut into 250 buckets of a thousand rounds, the slowest averaged 2.813 and the fastest 2.723. The longest gap between consecutive blocks inside the slowest bucket was nine seconds, and one protocol version ran throughout. That is a boringly reliable chain, and reliability is the harder engineering problem. The uptime claim is not the one doing the misleading.
| Advertised throughput | over 10,000 TPS, "in practice" |
| Measured, 250,000 rounds | 14.07 TPS |
| Measurement window | round 65,890,951 to 66,140,951, 8.00 days |
| Mean block time | 2.765 s (range 2.723–2.813 by 1,000-round bucket) |
| Longest inter-block gap found | 9 s |
| Contract-generated share of count | 29.86% |
| Chess rating tokens, top-level share | 27.24% (29.05% with World Chess) |
| Total value locked | $33.2m |
| ALGO price, market cap | 8.16p, £740.6m |
Bear and bull
The bear case is capacity nobody wants. A chain settling 14 transactions a second against $33.2m of locked value and $6,274 of daily fees is not constrained by block space but by demand, and spare capacity has no price.
The bull case is the same fact read the other way. The infrastructure works, it held a 2.765-second heartbeat for eight days without stumbling, it costs almost nothing to use, and it would not need re-engineering if demand arrived. Whether it arrives is the entire question, and block headers cannot answer it.
Throughput, block times, transaction types, asset and application identifiers and supply measured from Algorand mainnet via two independent algod nodes (algonode.cloud and 4160.nodely.dev) and the AlgoNode indexer, 19 October 2026, 18:45–18:47 UTC. Marketing claims quoted from algorand.co, retrieved 19 October 2026. TVL, decentralised exchange volume, chain fees and stablecoin totals from DefiLlama; price, market capitalisation and circulating supply from CoinGecko, both 19 October 2026, 18:46 UTC. Contested: the chain's ledger (9,813,587,411 ALGO) and CoinGecko's circulating supply count locked foundation stock differently; both appear above. Whether Algorand's developer documentation qualifies the homepage claim is unverified — it returned no comparable figure to a static fetch. Transaction-type and inner-transaction shares come from random samples of 500 and 200 blocks within the window, not from all 250,000; composition varies block to block and these are estimates. Not a recommendation to buy, sell or hold anything.