How each kind of stablecoin holds its peg

Mechanism, not the label on the box

Chapter 1 named three surviving pegs and one extinct design. The live market is slightly richer than that, and the dashboards are slightly poorer.

Six constructions answer the stress question. Fiat-backed tokens hold cash and short bills and defend the peg at the issuer's desk. Crypto-backed tokens hold over-posted crypto in a vault and defend it by liquidation. Hybrid tokens mix that vault with tokenized Treasuries and a Peg Stability Module. Synthetic tokens hold a hedged book. Commodity-backed tokens hold metal and do not target one dollar. Algorithmic tokens hold a sister token and, in this material, no longer have a market.

DefiLlama, on 4 September 2026, used three buckets: fiat-backed (90 tokens, 285.04 bn USD), crypto-backed (221 tokens, 25.50 bn USD) and algorithmic (23 tokens, 0.67 bn USD). Hybrid and synthetic sit inside "crypto-backed" on that dashboard. USDe has neither a Maker-style vault nor a Circle-style cash window. A risk reading that stops at the label is unfinished.

The same day's reference state: 423 tokens on 175 chains; 312.0 bn USD across all pegs, 311.2 bn USD of that USD-pegged; 3.3% below the author's 17 May 2026 peak of 322.4 bn USD. Chapter 2 records the competing peak figures.

Fiat-backed: the desk, and the bank behind it

A fiat-backed stablecoin holds cash and short-term Treasury bills against the supply, one for one, and defends the peg by mint and redeem at the issuer. The window is institutional: KYC, a signed desk agreement, and a minimum ticket that the knowledge base records as 100,000 USD and up. Chapter 1 left Circle's current floor unlocated; Tether's is the Kaiko figure from April 2026.

On 4 September 2026 DefiLlama put USDT at 183.34 bn USD (58.91%) and USDC at 74.46 bn USD (23.93%). The newer four in the same family that day: USD1 at 4.24 bn USD, USDG at 3.18 bn USD, PYUSD at 3.00 bn USD, RLUSD at 2.40 bn USD. RLUSD is the smallest of those four despite the press weight on Ripple's name.

The failure point is the bank or custodian that holds the cash, not a market price of the reserve asset. Silicon Valley Bank in March 2023 is the type case. Treasuries in the book did not stop the print.

Circle's reserve, on a Deloitte attestation dated 6 November 2025, was about 13% cash and bank deposits and about 87% the Circle Reserve Fund. Circle's 10-K for year-end 2025, relayed in the Polish edition, puts the fund at about 88% of USDC reserves. Those are different dates and slightly different shares; they are not averaged here.

Tether's BDO attestation for 30 June 2026 put about 115 bn USD in Treasuries, about 18.83 bn USD in gold, and about 5.8 bn USD in 98,932 bitcoin, with the rest in cash, repo, money-market funds and loans. The GENIUS Act's permitted reserve list is cash, T-bills of 93 days or less, repo, money-market funds and Fed balances. Gold and bitcoin, about 24.6 bn USD of Tether's mid-2026 book, sit outside that list. That is the stated reason Tether is building a separate USAT token through Anchorage rather than putting USDT itself through the federal payment-stablecoin box.

Crypto-backed, and the hybrid that added Treasuries

A collateralized debt position, a CDP, is a vault that mints a stablecoin against crypto posted in excess of the amount issued. Typical loan-to-value in this material is 60–80%. If the collateral's price falls through the line, the vault is liquidated automatically.

GHO, Aave's dollar token, grew from about 35 mn USD to about 565–584 mn USD in 2026. crvUSD sat at about 290–293 mn USD and liquidates through LLAMMA, a band mechanism, rather than a single threshold. Both are still a rounding error next to USDT. The failure point is the one Black Thursday showed: an oracle that is late, or a liquidation engine that cannot keep up.

Sky's USDS, with legacy DAI still in circulation, is the hybrid. It keeps crypto vaults and adds tokenized Treasury exposure and a Peg Stability Module that swaps into USDC. USDS peaked at about 12 bn USD in April 2026. On 4 September 2026 the Polish edition's DefiLlama read puts USDS at 6.60 bn USD; the knowledge base records about 8.16 bn USD for September 2026. Both sit here. Legacy DAI is recorded at about 4.47 bn USD in the knowledge base.

USDS has a freeze function that DAI never had. Two tokens that look the same in a DeFi pool are not the same object once an address needs to be stopped.

Synthetic: a hedge instead of a pile

A delta-neutral hedge, in Ethena's design, is a long spot position against a short perpetual futures position sized so that the net exposure to the crypto's price is supposed to stay near zero. The funding rate is the periodic payment between longs and shorts on that perpetual; when it is positive, shorts are paid, and that payment is the yield that sUSDe passes through.

USDe has no fiat reserve and no formal par window at an issuer. The peg lives in the hedge and in the secondary market. Supply peaked above 14 bn USD before October 2025; the Polish edition puts the October peak at about 14.7 bn USD and the 4 September 2026 DefiLlama reading at 4.28 bn USD. The failure point is a funding rate that stays negative, so the hedge costs money instead of paying it.

USDe is not a payment stablecoin under the GENIUS Act definition in chapter 1. Neither is DAI. Calling them that in a US legal sense is a category error.

Gold tokens, and the design that went to zero

A commodity-backed token holds physical metal and redeems into it. PAXG (Paxos) and XAUT (Tether) are the two names in this material. The unit price tracks gold, not 1.00 USD, so they are a poor dollar peg and a different product. No source used for this chapter gives a precise September 2026 supply for either token, so none is stated.

The algorithmic design is in the past tense. UST had no reserve. It held the peg by minting and burning LUNA. Supply was above 18 bn USD in early May 2022 and 0.12 USD by 09:00 on 16 May. Combined value of UST and LUNA is recorded at about 60 bn USD to near zero in a week; other sources count about 40 bn USD depending on scope. Chapter 2 keeps both. The failure is structural: demand for the sister token has to grow without a ceiling.

Four questions cut across every family, and two tokens in the same family can answer them differently.

1. Does yield reach the holder. The GENIUS Act bars a payment-stablecoin issuer from paying it directly; wrappers (sUSDe, sUSDS) sit beside that ban.
2. Is there a formal right to redeem at par, or only a secondary market. USDG is described as fully redeemable 1:1, a stronger retail term than USDT's and USDC's institutional desks. USDe has no formal par window.
3. Who controls the collateral: one legal entity, a DAO, or a book of futures positions at venues.
4. Can the issuer freeze an address. USDT and USDC ship a blacklist in the contract. USDS can freeze. DAI could not.

Proof of reserves, attestation, audit

Three documents get treated as if they were one.

A proof of reserves is a cryptographic or on-chain check that some set of assets existed at a moment. It does not, by itself, say those assets are unencumbered, or that they cover the whole liability, or that they existed yesterday.

An attestation is an accountant's opinion on management's assertion for a stated date, under an AICPA (or, for Tether's BDO reports, ISAE 3000) assurance standard. It is one day's snapshot of what management says the book is. It is not a year-round test of controls.

An audit, in the reserve-specific sense this chapter uses, is a full examination of the reserve over a period, with an opinion on the statements rather than on one day's assertion. As of 2026, neither USDT nor USDC has a published full reserve-specific annual audit in the sources used here. Circle's financial statements as a public company are a different object from a reserve audit of USDC.

USDC's circulating supply differs by method, and the gap is stated as a method gap. Circle's attestation, net of unissued and denied tokens, showed 77.17 bn USD on 30 April 2026. DefiLlama's raw on-chain figure was 73.71 bn USD on 29 June 2026. Different dates, different cuts of the same token. They are not averaged, and they are not evidence that the reserve is short.

USDS, in a March 2026 on-chain breakdown with no named third-party attestor, was about 38% USDC through the Peg Stability Module, about 25% crypto vaults, about 22% RWA loans, about 10% Spark, about 5% other.

Paxos moved PYUSD's attestor to KPMG. That is a change of accountant, not the completed Tether Big Four engagement this chapter is told not to treat as a finished audit.