Sui’s homepage says it set a record of more than six million transactions per second in July. Measured against a public full node this morning, it sustained about 93 over 24 hours and 79 over 30 days. Both numbers are true. Only one of them describes a blockchain.
The six million figure came from an off-chain experiment: payment channels that settle to the mainnet only when they close, run with a test token and sponsored gas. Sui’s own blog says so. It is a legitimate demonstration and it is not throughput. The Mysticeti paper claims over 200,000 transactions per second in controlled testing, which puts real usage at about one twentieth of one per cent of the lab figure.
That gap is the story. The constraint is demand.
What the chain earns
Sui collected $3,528 of layer-one gas fees in the last 24 hours, $108,052 over the last 30 days and $3.96m over the trailing year. The Foundation publishes its own revenue data, and it corroborates DefiLlama to within 3.5 per cent.
A market capitalisation of £2.46bn is $3.33bn, so the token trades at roughly 840 times the dollars the network charges for its use. Counting every application on Sui, about 100 times. Calendar gas take has fallen 89 per cent in a year, from $911,194 to $98,374.
| SUI price | £0.60 |
| Market capitalisation | £2.46bn |
| Fully diluted valuation | £6.01bn |
| Circulating share of max supply | 40.97% |
| Total value locked | $473m |
| TVL against October 2025 peak | −82% |
| Layer-one fees, trailing year | $3.96m |
| Measured throughput, 30-day average | 79 tx/s |
| Active validators | 126 |
| Nakamoto coefficient | 18 |
| Staking yield | 1.35% |
The object model, honestly
The technical claim is real but narrower than the marketing. Every asset is an object with a declared owner, so a transaction states in advance which objects it touches. Independent transactions can execute in parallel without the read-write guessing optimistic execution needs elsewhere, and those touching only objects you own can skip consensus ordering.
The limit is where the money is. Lending markets, trading pairs, order books and oracles are all shared objects, and shared objects go through consensus and serialise. Sui’s documentation says so. Lending alone accounts for more locked value on Sui than everything else combined — exactly the activity the fast path does not accelerate.
The fast path makes your own transfer quick. It does not make a busy application quick.
The bear case
Beyond the fee collapse: stablecoins on Sui have fallen 46 per cent in a year to about $444m, with the holder count up 0.4 per cent in eight months. Around 11 per cent of user transactions fail. Two critical mainnet incidents landed in nine months — a six-and-a-half hour consensus outage in January, and a settlement failure in May that took roughly 50 hours to resolve.
The supply picture is subtler than usual. Monthly unlocks halved in June to 22.0m SUI, about 0.54 per cent of circulating, with no cliff ahead. But the published schedule ends in May 2030 at 47.83 per cent circulating, leaving over five billion tokens with no timetable at all. And 69.8 per cent of maximum supply is staked against 40.97 per cent circulating, so a great deal of unvested SUI is earning rewards and voting today. No first-party breakdown exists.
Then there is Cetus. The May 2025 exploit took $223m through an overflow bug in a third-party maths library, and validators coordinated within hours to freeze $162m by blocking the attacker’s addresses, with a vote carrying 90.9 per cent of stake approving the reclaim. Users got their money back. The chain also showed that its validator set will censor quickly on command, and did something similar after the Volo exploit in April 2026.
The bull case
The validator set is genuinely well distributed: 126 validators, a Nakamoto coefficient of 18, and Mysten Labs’ own two validators holding 5.63 per cent between them. That is flatter than most layer ones. Mysticeti is real, published and shipped, and Mysten reports an 80 per cent latency reduction on testnet. Locked value has risen 14 per cent off its June low across three months, and 30-day exchange volume of $1.18bn is up 137 per cent.
One consumer application deserves a mention. CARDZ.GAME, a physical trading-card platform, generated $532,060 of fees in 30 days — more than any exchange or lending protocol on Sui, and five times the chain’s own gas take. A real product with paying customers, and the most interesting thing here.
The verdict is not that Sui is broken. It is that a well-engineered chain with a well-distributed validator set is valued at some 840 times its fee income while usage is flat and locked value is down four-fifths. The technology is not the weak point. The demand is.
Sources, retrieved 7 September 2026: CoinGecko for price and supply; the Sui Foundation’s circulation, revenue and transaction APIs; DefiLlama for locked value, fees and stablecoins; Sui’s status-page incident history; docs.sui.io; the Mysticeti paper on arXiv; sui.io for marketing claims, labelled as such. Throughput was measured against a public Sui full node by checkpoint comparison over 24 hours, 7 and 30 days, not taken from a dashboard. Contested: daily transaction counts differ by up to 4.9 times across three sources because each counts something different, so only measured throughput is quoted; stablecoin supply is reported between $431m and $449m; the 2023 allocation split could not be confirmed first-party; the share of staked SUI that is unvested is disclosed nowhere we could find; and Sui’s documentation contradicts itself on whether owned-object transactions still bypass consensus. Not a recommendation to buy, sell or hold anything.