What Is an Algorithmic Stablecoin? How the Designs Work and Why Most Failed

An algorithmic stablecoin is a token that tries to hold a fixed price, usually $1, by using code to change its supply or by swapping against a second token the same project issues. A fiat-backed stablecoin does it differently: it keeps cash and short-term government debt in reserve. When an algorithmic coin trades above its target, the protocol adds supply. When it trades below, the protocol removes supply. The peg holds only while enough people believe it will. The best-known test of that belief was TerraUSD (UST), which collapsed in May 2022. As of 25 October 2026, none of the 15 largest dollar stablecoins listed by DefiLlama is classed as algorithmic. US law now requires payment stablecoins to be backed at least 1 to 1 by reserves.

What changed most recently

• 18 July 2026: a US Treasury study was due. Section 14 of the GENIUS Act requires the Secretary of the Treasury to study "non-payment stablecoins, including endogenously collateralized payment stablecoins". The Treasury must consult the Federal Reserve Board, the OCC, the FDIC, the SEC and the CFTC, and report to Congress within 365 days of enactment. The Act was approved on 18 July 2025, so the report was due around 18 July 2026.
• 11 December 2025: Terraform Labs co-founder Do Kwon was sentenced to 15 years. The US Attorney's Office for the Southern District of New York announced the sentence for what it called a "$40 billion fraud". Kwon had pleaded guilty in August 2025 to wire fraud and conspiracy charges. Judge Paul Engelmayer passed the sentence.
• 18 July 2025: the GENIUS Act became law (Public Law 119-27). A "payment stablecoin" issuer must "maintain identifiable reserves backing the outstanding payment stablecoins" "on an at least 1 to 1 basis". A coin held up only by code or by its own sister token cannot meet that test.
• 25 January 2025: USDD 2.0 went live on Tron. USDD launched in 2022 and was described then as an algorithmic stablecoin. TRON DAO's own release presents version 2.0 as over-collateralised, minted against TRX or USDT. DefiLlama now classes USDD as "crypto-backed", with about $1.78 billion in circulation on 25 October 2026.

How algorithmic stablecoins work

Every design does the same basic job: it widens supply when the price is above target and shrinks it when the price is below. There are three main ways to do that.

Rebase: change every balance (Ampleforth, AMPL)

A rebase token changes how many tokens each wallet holds, rather than moving tokens between wallets. According to Ampleforth's documentation:

• the protocol rebases once a day at 2:00 UTC;
• the target is "the CPI adjusted 2019 USD", not a flat $1;
• supply changes only when the price moves past a deviation threshold, currently 5%;
• every holder's balance changes by the same percentage, so each holder's share of the network stays the same.

What matters in practice is that the token count in your wallet floats while your share stays constant. The dollar value of a position is not protected; only the number of tokens changes.

Two-token seigniorage: swap against a sister token (TerraUSD and LUNA)

UST relied on a smart contract that let holders swap 1 UST for $1 worth of LUNA, and the reverse.

• If UST traded above $1, traders could burn $1 of LUNA to mint 1 UST and sell it at a profit. That added UST supply.
• If UST traded below $1, traders could buy UST cheaply and redeem it for $1 of LUNA. That removed UST supply.

The weak point is that the backstop is a token the same system creates. When UST holders redeem at scale, the protocol mints LUNA to pay them. That pushes LUNA's price down, which means each redemption needs even more LUNA, which weakens confidence in UST further. The SEC described UST as an "algorithmic stablecoin" that supposedly kept its peg by being interchangeable with LUNA.

The GENIUS Act has a name for this design: an "endogenously collateralized" stablecoin. It is one that "relies solely on the value of another digital asset created or maintained by the same originator" to keep its fixed price.

Fractional-algorithmic: part collateral, part code (Frax, original design)

Frax launched on 20 December 2020 using what its documentation calls a fractional-algorithmic mechanism. Each FRAX was backed partly by collateral such as USDC and partly by the protocol's own share token, FXS. At an 85% collateral ratio, redeeming 1 FRAX returned $0.85 in USDC and $0.15 in newly minted FXS. The ratio moved with the price. It fell when FRAX traded above $1 and rose when FRAX traded below.

In February 2023, Frax governance passed proposal FIP-188 to target a 100% collateral ratio. News reports from the time said the ratio had been about 92% before the vote. Frax's v3 documentation now says the protocol "will attempt to keep >=100% CR at all times". By its own description, the largest surviving fractional design is no longer fractional.

Two failures that show the weak point

IRON, June 2021

IRON was about 75% backed by USDC and about 25% by its own token, TITAN, at the time of its run. A Federal Reserve FEDS Note by Austin Adams and Markus Ibert (2 June 2022) records that IRON "significantly broke the peg on June 16 and dropped from $1 to less than the $0.75". The authors concluded that "design flaws in the no-arbitrage mechanism contributed to the failure". They added that "the speed of IRON's failure serves as a useful reminder that some stablecoins are not stable at all."

The lesson is that partial collateral did not stop the run. Once the token part of the backing lost value, it was worth far less than its share of the peg was supposed to be.

TerraUSD, May 2022

• Scale. An NBER working paper by Liu, Makarov and Schoar (April 2023) describes Terra as "the third largest cryptocurrency ecosystem after Bitcoin and Ethereum". It says Terra "collapsed in three days in May 2022".
• Demand was driven by yield. The SEC alleges that UST was marketed as a "yield-bearing" stablecoin, with yields of up to 20% through Anchor Protocol. The NBER authors trace the run to large holders pulling funds from Anchor, as doubts grew about whether the system could last. They found the run was likely not driven by concentrated market manipulation.
• Who lost most. The same paper finds that wealthier, more sophisticated investors exited first and lost less.
• Aftermath. In February 2023 the SEC charged Terraform Labs and Kwon. In June 2024 Terraform and Kwon agreed to a $4.47 billion proposed judgment with the SEC, The Block reported. Kwon was sentenced in December 2025.

Where the rules stand in the US

The GENIUS Act covers payment stablecoins and requires 1 to 1 identifiable reserves. A coin that relies on its own sister token cannot meet the reserve test, so it cannot be a payment stablecoin under the Act. Endogenously collateralized coins are instead the subject of the Section 14 Treasury study, which must analyse eight areas:

1. the categories of non-payment stablecoins, including the benefits and risks of their design features;
2. the participants in non-payment stablecoin arrangements;
3. their use and potential use;
4. the nature of their reserve compositions;
5. the types of algorithms they employ;
6. their governance, including decentralisation;
7. how they are promoted and advertised;
8. the clarity and availability of consumer disclosures.

Questions to ask about any coin that calls itself algorithmic

1. What do you get if you redeem one? Cash or Treasuries, another crypto asset, or the project's own token? If it is the project's own token, the peg depends on that token's price.
2. Who prints the backstop? If the protocol can mint the asset that pays redemptions, a run can feed itself, as it did with LUNA and TITAN.
3. Where does the demand come from? If most holders are there for a subsidised yield, demand can disappear as soon as the yield is cut or doubted. Anchor offered up to 20%, and TRON DAO subsidised a 20% APY tier on USDD 2.0 staking at launch, according to press reports.
4. What does the price feed use? The Fed note found that IRON's mechanism used a time-averaged price, and that switching to the spot price would have opened it to manipulation.
5. Does the label match the design today? FRAX and USDD both began as algorithmic or partly algorithmic and now describe themselves as fully or over-collateralised. Check the current documentation, not the launch-era name.