You know how USDT holds $1 — dollar deposits and short-term Treasuries behind it. You know how DAI holds $1 — overcollateralized ETH. Algorithmic stablecoins are different: there is no external asset behind them whatsoever. They rely purely on a mechanism of automatic minting or burning when the price drifts, trying to hold $1 through code logic alone. The mechanism works when markets are calm, but it contains a fatal design flaw that produced one of the largest collapses in crypto history in 2022.
Algorithmic stablecoins have two main designs. The first is the dual-token model (example: UST and LUNA): the system has two tokens, one stablecoin (UST) and one governance token (LUNA). The rules are simple: when UST rises above $1, anyone can burn $1-worth of LUNA to mint one new UST and sell it, pushing UST back to $1; when UST falls below $1, anyone can burn one UST in exchange for $1-worth of LUNA, buying UST back toward $1. The second is the elastic supply model (example: Ampleforth AMPL): only one token, but the protocol adjusts every holder's balance daily — when price is high it prints coins to dilute it, when low it reclaims coins from holders. Your balance floats daily, though theoretically your share is unchanged.
The dual-token model has one premise: arbitrageurs, when UST falls below $1, are willing to burn UST and accept an equivalent of LUNA. That willingness depends on believing “I can sell that LUNA for $1 later.” But in panic, the problem emerges: mass UST burns → system mints massive new LUNA → LUNA supply surges, price collapses → the LUNA received for burning one UST is worth less and less → arbitrage loses its appeal → no one keeps burning → UST keeps falling. This positive-feedback death spiral played out in the Terra collapse of May 2022. Within days, UST fell from $1 to near zero; LUNA's market cap evaporated from a peak of roughly $40B; hundreds of thousands of people were wiped out. The root cause was single: the algorithmic stablecoin's “$1 floor” was never an external asset — it was people's willingness to believe it would return to $1. When confidence broke, the floor vanished.
The failure of pure-algo models pushed the industry toward hybrids. Frax's design uses partial fiat collateral plus a partial algorithmic mechanism, automatically adjusting the collateral ratio based on market confidence in the protocol — less collateral when confidence is high, more when it's low. This is far more robust than pure-algo because there's always some real-asset buffer. But even a hybrid doesn't fully escape the circular dependency of “the collateral ratio itself depends on market confidence.” Industry consensus has converged: a purely algorithmic stablecoin relying on no external assets has no capacity for self-repair in large-scale panic.
To tell whether a stablecoin is algorithmic, ask one question: if everyone tried to redeem tomorrow, are there enough external assets to give every holder real $1? If the answer is “it's backed by the market cap of another token,” it's algorithmic, and the death-spiral risk is real. No matter how high the yield or APY, you're pricing confidence risk. This doesn't mean every algorithmic design is a scam, but after Terra, the healthy question is: “what is actually behind your $1?” — not “what's your APY?”