Algorithmic Stablecoins: Why Most Broke Their Peg
Algorithmic stablecoins promise a dollar without a bank account behind it. The mechanism is elegant on paper and fragile in practice. Here is why.
An algorithmic stablecoin is a token that tries to hold a fixed value, usually one US dollar, using software rules and market incentives rather than a reserve of cash or bonds held by an issuer. The design is meant to be capital-efficient: instead of parking a real dollar behind every token, it uses arbitrage, minting and burning to nudge the price back toward the peg. The catch is that this only works while traders believe it will work, and that belief has broken more than once.
The most-studied failure is TerraUSD (UST), whose collapse in May 2022 wiped out enormous value in days. In the U.S. Securities and Exchange Commission's own enforcement action against Terraform Labs, regulators described UST as a stablecoin that was supposed to maintain its peg through an algorithm rather than reserves. That single event reshaped how the entire industry, and its regulators, talk about the category.
Key takeaways
- An algorithmic stablecoin maintains its peg through code and trader incentives, not a dollar-for-dollar reserve of cash and short-term Treasuries.
- The best-known mechanism is a two-token loop: a stablecoin and a volatile "absorber" token that expands and contracts supply to defend the price.
- That loop can enter a death spiral. If the absorber token loses value faster than the system can print it, both tokens can fall toward zero together.
- The dominant stablecoins by supply today (USDT, USDC) are reserve-backed, not algorithmic. Purely algorithmic designs are 0.04% of the market: $0.13B of $313.74B in adjusted onchain stablecoin supply (Allium data, September 23, 2026).
- Reserve requirements in recent US stablecoin legislation are framed around holding reserve assets against tokens in circulation, which a design holding no reserves would not satisfy.
The problem algorithmic designs were trying to solve
A reserve-backed stablecoin is simple to trust and expensive to run. Every token in circulation is backed by a real dollar or equivalent asset sitting in an account, so scaling to billions of tokens means custodying billions of real dollars. That capital does nothing except sit there as collateral.
Algorithmic stablecoins were pitched as the way around that constraint. If code could hold the peg using incentives instead of idle reserves, you could scale a dollar-denominated token without warehousing an equal pile of dollars. On paper that is more capital-efficient and more decentralized, because no single company holds the cash. The failures have shown that removing the reserve also removes the floor.
How the two-token loop works
One major algorithmic design, and the one behind the largest failure, uses two linked tokens: the stablecoin itself and a volatile governance or "seigniorage" token that absorbs price pressure. Terra's system paired UST (the stablecoin) with LUNA (the absorber), a structure documented in Terra's own historical protocol documentation.
The mechanism runs in two directions:
- When the stablecoin trades above $1: the protocol lets anyone burn $1 worth of the absorber token to mint one new stablecoin. Minting more supply pushes the price back down toward the peg, and the arbitrageur pockets the difference.
- When the stablecoin trades below $1: the protocol lets anyone burn one stablecoin to mint $1 worth of the absorber token. Removing supply pushes the price back up, and again the arbitrageur profits from the gap.
In calm markets this arbitrage keeps the price glued to a dollar. The problem lives entirely in the second direction, when the peg is under stress.
How the feedback loop compounds
The fatal weakness is that defending a falling stablecoin means printing more of the absorber token, which dilutes it. If holders are already selling the absorber, more supply drives its price down faster, which forces even more printing to redeem each stablecoin. The table below is a simplified illustration of how that feedback loop compounds. The numbers are hypothetical, chosen only to show the shape of the effect, not drawn from any specific protocol.
| Step | Stablecoin price | Absorber token price | Tokens minted to redeem $1M of stablecoin |
|---|---|---|---|
| 1 (calm) | $1.00 | $100.00 | 10,000 |
| 2 (mild stress) | $0.95 | $60.00 | 16,667 |
| 3 (panic) | $0.80 | $15.00 | 66,667 |
| 4 (spiral) | $0.50 | $1.00 | 1,000,000 |
| 5 (collapse) | $0.10 | $0.01 | 100,000,000 |
Read down the last column. As the absorber token falls, the number of new tokens the protocol must mint to buy back a fixed amount of stablecoin explodes. That hyperinflation of the absorber destroys the confidence that was holding the whole structure up. Terra's UST and LUNA traced almost exactly this path over a few days in May 2022, a sequence later detailed in the SEC's complaint.
Not all "algorithmic" tokens are the same
The label gets stretched across designs with very different risk profiles. It helps to separate three buckets.
| Model | What backs the peg | Example type | Main risk |
|---|---|---|---|
| Reserve-backed (not algorithmic) | Cash and short-term Treasuries, 1:1 | USDC, USDT | Issuer, custody, redemption trust |
| Crypto-overcollateralized | More than $1 of volatile crypto locked per $1 issued | DAI-style vaults | Sharp collateral crashes, liquidation lag |
| Purely algorithmic (uncollateralized) | Nothing but a supply mechanism and market faith | UST / LUNA type | Death spiral, total loss |
MakerDAO's DAI is often lumped in with algorithmic stablecoins, but its structure is different: users lock crypto worth more than the DAI they mint, and the system liquidates positions if collateral falls too far. The rules of that system are set out in Maker's protocol documentation. Overcollateralization gives it a real asset floor that a purely uncollateralized design lacks.
Why this matters even if you never touch one
Two shifts have pushed the algorithmic model to the margins, and both affect how everyone builds with stablecoins now.
Regulation now defines reserves into the rules. The U.S. GENIUS Act, signed in 2025, sets out requirements for payment stablecoins, including backing with high-quality liquid reserves. You can read the framing in the text tracked on Congress.gov. A stablecoin whose entire premise is holding no reserves cannot satisfy a reserve requirement, so purely algorithmic dollar tokens are outside the reserve-backed model that recent US legislation is built around
The money and the usage sit with reserved tokens. The stablecoins actually moving value in payments, trading and settlement are the reserve-backed ones. Industry research on the growth of the stablecoin economy treats reserve-backed stablecoins as the base case for institutional adoption. The practical takeaway is that if you are building a business on stablecoin rails, the reliable primitives are reserved, and algorithmic designs are a risk to screen out rather than a foundation to build on.
What changes when the peg holds versus when it breaks
- Redemption certainty: with a reserved stablecoin, a holder can redeem for a real dollar from the issuer. With a purely algorithmic token, redemption depends on an absorber token that can itself be collapsing, so there may be nothing worth taking.
- Capital efficiency vs. survivability: the algorithmic model frees up the reserve capital, but the same design removes the buffer that lets a token survive a bank-run moment. The saving shows up in good times and the cost shows up all at once.
- Contagion: when a reserved token wobbles, the damage is largely contained to that token. When an algorithmic token spirals, it drags its paired absorber token, its lenders and any protocol that treated it as safe collateral down with it.
Telling reserved from algorithmic in the raw data
The category confusion is not just a naming problem, it is a data problem. To judge whether a token is behaving like a reserved stablecoin or an uncollateralized one, an analyst has to reconstruct minting and burning events, supply changes, the identity of the paired absorber token and its price, and redemption flows, and then line those up consistently across every chain the token trades on. Raw blockchain logs do not label any of that. The same mint event looks like an unstructured internal transaction on Ethereum and something structurally different on a Cosmos-based chain, and the two never agree on field names or formats out of the box.
Resolving each of those records into the same fields (asset, issuer, event type, supply delta, USD value) across many networks is what makes an apples-to-apples comparison possible. Allium ingests and standardizes stablecoin activity across many chains into consistent schemas. For teams that want the underlying tables, the stablecoin datasets expose supply, mint and burn events, and transfers in a normalized form. That normalization is what lets a researcher spot an absorber token inflating before a peg breaks rather than after.
Risks and open questions
The honest state of the field:
- No algorithmic design has survived a real crisis at scale. Several "improved" or "fractional" models launched after 2020 promising to fix the death spiral. None has yet demonstrated that it holds through a sustained loss of confidence, which is the only test that matters.
- Hybrid models blur the line. Some tokens are partially collateralized and partially algorithmic. That makes the reserve requirement question harder, because the algorithmic portion carries the same tail risk regardless of how thin it is.
- Regulatory status is still settling. Frameworks like the GENIUS Act address reserved payment stablecoins directly, but the treatment of decentralized, crypto-collateralized and hybrid designs is less resolved. Anyone relying on one should read the primary rules rather than summaries.
- "Decentralized" does not mean "safe." Removing a single corporate issuer removes one point of failure and adds another: there is no balance sheet to make holders whole when the mechanism fails.
The short version: an algorithmic stablecoin removes the cost of holding reserves, and with it the asset that backs redemption under stress.
Frequently asked questions
What is an algorithmic stablecoin in simple terms?
It is a token designed to hold a fixed value, usually one dollar, using software rules and market incentives to expand and contract its supply, rather than holding a real dollar in reserve behind each token. The peg depends on traders acting on those incentives, which works in calm markets and can break sharply under stress.
Why did TerraUSD (UST) collapse?
UST relied on a paired token, LUNA, to absorb price pressure. When UST fell below its peg, the system minted large amounts of LUNA to redeem it. As LUNA's price crashed, ever more of it had to be printed, hyperinflating LUNA and destroying confidence in both tokens. This feedback loop is known as a death spiral, and it played out over a few days in May 2022.
Is USDT or USDC an algorithmic stablecoin?
No. USDT and USDC are reserve-backed stablecoins: their issuers report holding cash and short-term assets against the tokens in circulation, with composition disclosed in periodic attestations. Their risks relate to issuer trust and reserve quality, not to the supply-mechanism failure that defines algorithmic designs.
Is DAI an algorithmic stablecoin?
Not in the uncollateralized sense. DAI is crypto-overcollateralized: users lock crypto worth more than the DAI they mint, and the system liquidates positions if collateral falls too far. That collateral floor makes it structurally different from a purely algorithmic token that holds no backing.
Are algorithmic stablecoins legal under new US rules?
The GENIUS Act, signed in 2025, sets reserve requirements for payment stablecoins. A purely algorithmic token holds no reserves by design, so a design holding no reserves would not satisfy a reserve requirement from the regulated payment stablecoin category in the US. Read the primary legislative text rather than summaries for the current detail.
How can you tell if a stablecoin is algorithmic?
Check the issuer's own documentation for how the peg is maintained. If it describes maintaining the peg through minting, burning or a paired token rather than through cash and Treasury reserves, it is algorithmic. Onchain data on supply changes, mint and burn events, and any paired absorber token confirms whether behavior matches the stated model.
Interested in learning more about Allium’s stablecoin data? Speak to someone on the team.
Allium provides onchain data infrastructure. Companies named in this article may be Allium customers, prospects or commercial counterparties. This article is informational only and is not investment, legal or tax advice. Data and information last reviewed: September 23, 2026.
