Between July 20 and July 21, 2026, the circulating supply of USD.AI (USDai) fell from about 202 million dollars to about 154 million dollars, a drop of roughly 24 percent in 24 hours. Over the same window the token never left a dollar. CoinGecko has it at 1.00 dollars with a 24 hour change of about four hundredths of a percent, and DefiLlama prices it at 0.99996. We read the live supply directly off Arbitrum mainnet before writing this, and it matched the public trackers to within a fraction of a percent.
The combination decides which word is honest. A stablecoin whose price collapses has depegged. A stablecoin whose supply collapses while the price holds has been redeemed or unwound at par. USDai is the second case, not the first. And here the cause is not diffuse. It is one transaction, which we decode below.
This is a companion to our earlier read of a different shrinking synthetic dollar, MegaUSD, which shed more than half its float in a day and held its peg, and to the broader case for reserves you can verify rather than take on faith. The closest cousin by mechanism is our walk through Neutrl's NUSD.
What actually happened
USD.AI is a synthetic dollar issued by the USD.AI protocol, which per its own materials and DefiLlama finances physical AI and GPU compute hardware. It lives almost entirely on Arbitrum. Per DefiLlama's per-chain balances, of the roughly 154 million dollars outstanding, about 152 million is on Arbitrum, with only about 5,700 dollars on Ethereum and about 1.33 million on the Plasma chain. So the whole move is an Arbitrum event, which is where we read it.
The drop is not a slow bleed across many holders. It is one redemption. At Arbitrum block 485,964,642, timestamped July 20, 2026 at 21:17:02 UTC, a single transaction burned 54,404,473.45 USDai and paid out 54,404,473.45 PayPal USD to the account that redeemed. That one event is the overwhelming majority of the entire 24 hour supply change. Everything else on the token over the surrounding day nets to a small remainder.
What the on-chain data shows
Here are the numbers we can stand behind without attribution, because we read them ourselves. The live USDai token on Arbitrum is the contract at 0x0A1a1A107E45b7Ced86833863f482BC5f4ed82EF. Its symbol is USDai and it carries 18 decimals. At Arbitrum block 486,273,517, around 18:44 UTC on July 21, 2026, its totalSupply was 154,447,289.46 USDai. Anyone can reproduce that with a single call:
cast call 0x0A1a1A107E45b7Ced86833863f482BC5f4ed82EF "totalSupply()(uint256)" --rpc-url https://arb1.arbitrum.io/rpc
Divide by ten to the eighteenth for the human figure. Now the redemption. The transaction is 0xbb7dd585ef93b7fce01803e5e85039527629d8b7c570e40a40b72a9ab1632d18. Inside it, two transfers tell the whole story. The USDai contract emits a Transfer of 54,404,473.45 USDai from the redeeming account to the zero address, which is a burn. In the same transaction, the PayPal USD contract at 0x46850ad61c2b7d64d08c9c754f45254596696984 emits a Transfer of 54,404,473.45 PYUSD out of the USDai contract to that same account. The two amounts are identical to the cent. USDai was destroyed, and an equal face value of PayPal USD, a fiat backed dollar, was released from the token contract's reserves. That is a one to one redemption at par, executed on-chain and in full view.
One thing worth surfacing, because it is what a live check catches and a headline does not. If you read only the daily supply number, you would see a 24 percent drop and might reach for the word depeg. If you also read the price, you would see it never moved. And if you read the transaction, you would see the reason: not stress, but one large holder exiting through the front door at a dollar. Three different reads, three levels of confidence, and only the last one tells you what actually happened.
There is a flip side, and it belongs here rather than buried lower. One holder redeeming roughly a quarter of the entire float in a single transaction is a concentration signal in its own right. A first large redemption clearing at par is consistent with an orderly unwind, but it is not proof that the remaining holders will stay, and a large first mover can leave before a crowd, not instead of one. What we can verify is that this exit settled at par on the liquid leg. We cannot verify that the event is over, and we are not going to imply that a 24 percent drop in a day is nothing.
Why a live check would have caught the exit early, and what it could not
This is where verification actually helps, and where it does not, stated plainly. A continuous read of totalSupply, or a watch on transfers to the zero address, would have shown a 54 million dollar burn the moment it landed, hours before any daily tracker rolled over. If you held USDai, or lent against it, or ran a strategy that assumed a deep and stable float, that signal was on-chain and free the entire time. Watching supply and watching for a single dominant redemption is a real, cheap, early warning for exactly this failure mode, a fast exit, in a way that reading a once a day chart is not.
What that same check cannot tell you is whether the remaining backing is sound. A supply read is not a reserve audit. We can verify one strong thing here: the liquid leg that settled this redemption was real, because a fiat backed dollar of equal face value actually moved out of the contract, on-chain, when the holder redeemed. That is more than most redemptions let you see. But it is evidence about this redemption, not proof that every remaining USDai is fully backed, because we did not and cannot recompute the protocol's full reserve from the chain alone. A peg holding through a large exit is consistent with redemptions clearing at par, which is the reassuring reading, but it is not the same as a solvency proof, and we will not dress it up as one.
Where Kerne stands
The honest comparison is about method, not size, and we need to say the size part plainly so the method part is credible. Kerne is tiny. The kUSD in circulation is a few thousand dollars, orders of magnitude smaller than USDai. We are not holding ourselves up as a safer alternative at scale, because we are not at scale. What we can say is that the discipline this event rewards, being able to watch the float and see a redemption settle at par with your own eyes, is one we build on rather than ask you to trust.
Concretely, kUSD is backed one to one by USDC in an on-chain peg module, a backing ratio you can read off the chain and that we also publish, signed hourly, at our proof of reserves endpoint /api/por. That signature is our own key, not a third party's. Our first external audit, by Hexens, delivered its initial report on July 20, 2026, and remediation is underway, so we do not call the attestation independent yet. The delta neutral leg of the design is currently flat rather than actively hedged. We disclose those limits rather than dress them up. None of this is an offer or a solicitation, and nothing here is advice. It is a description of what is on-chain and what is not.
What a holder should do, on any dollar
Not as an accusation, but as a checklist you can run on anything you hold, including us. First, separate the two questions the word depeg blurs together: is the price at a dollar, and is the float stable. Read the price and the token's totalSupply as two different numbers, because this week they told two different stories. Second, when the float moves fast, read the transactions, not just the aggregate. A single large redemption and a broad panic look identical on a supply chart and nothing alike on-chain. Third, ask where the backing lives and whether you can recompute it, or only read someone's statement of it. We built a free tool to run these reads on any token or vault at kerne.fi/verify, and we would rather you point it at us than take our word.
Sources and as-of times. Current USDai supply: our own direct read of totalSupply on the Arbitrum mainnet contract 0x0A1a1A107E45b7Ced86833863f482BC5f4ed82EF at block 486,273,517, 154,447,289.46 USDai, July 21, 2026 around 18:44 UTC. The redemption: our decode of Arbitrum transaction 0xbb7dd585ef93b7fce01803e5e85039527629d8b7c570e40a40b72a9ab1632d18 at block 485,964,642, July 20, 2026 at 21:17:02 UTC, showing a 54,404,473.45 USDai burn and an equal PYUSD payout from the token contract; PayPal USD is the Arbitrum contract 0x46850ad61c2b7d64d08c9c754f45254596696984. Supply history and per-chain split: DefiLlama stablecoin id 309. Price and 24h change: CoinGecko and DefiLlama, July 21, 2026. The design description of USD.AI is attributed to the protocol and DefiLlama and not independently verified by us. The identity of the redeeming account is not independently confirmed and is described by its on-chain behavior only. Kerne figures: our own proof of reserves endpoint. Figures are point in time, and on-chain supply is often revised, so recompute before relying on any number here.
Verify it yourself
Run the same check on any reserve, or have it run for you.
Paste any issuer's signed attestation into the free verify tool and recover the signer, rehash the figures, and check freshness in your own browser. For a machine-signed, point-in-time read of an address you name, delivered on the page in about two minutes, the instant self-serve read is $29; a human-reviewed read is $149. A teardown like this one, commissioned on any target you name, is $499. An independent read of a counterparty you hold or allocate to is $2,500. Attestation tooling, not an audit, and not a solvency opinion.