Every stablecoin claims to be '$1,' but the structures backing that promise are worlds apart across the three main types. Fiat-collateralized (USDC, USDT) use real dollars and Treasuries; crypto-overcollateralized (USDS, DAI) use locked ETH and crypto assets; algorithmic (UST) rely on code logic and market confidence. These three design philosophies produce radically different outcomes when markets crash, banks fail, or confidence collapses.
Fiat-collateralized is the simplest structure and the most institutionally accepted: you deposit $1 → issuer puts it in cash or short-term US Treasuries → issues you 1 stablecoin. You can always send the stablecoin back and get $1 (minus fees). Advantages: 100% capital efficiency, deepest secondary market liquidity, highest institutional trust. Disadvantages: highly centralized — the issuer controls all reserves and has the technical ability to freeze any address (USDC's blacklist has been used multiple times); transparency depends on the issuer's audit willingness (USDC monthly Deloitte vs USDT attestations); all Treasury interest goes to the issuer, nothing to holders. Main failure modes: fraudulent reserve misrepresentation (Tether's 2021 NYAG settlement) or temporarily inaccessible reserves (USDC/SVB 2023 — a liquidity crisis, not a solvency crisis, resolved in three days).
Crypto-overcollateralized lets you use crypto assets (mainly ETH) as collateral to borrow stablecoins, but must overcollateralize — typically locking over 150% of the borrowed amount. With Sky's USDS: lock $150 of ETH, borrow up to 100 USDS. If ETH falls enough that CR hits the liquidation threshold (~130%), the system auto-sells your ETH to repay the debt. Advantages: high decentralization — code is open, no institution can freeze your position; reserve transparency (all vault states queryable on-chain); some RWA reserves (Sky's US Treasury position) generate real yield distributed to sUSDS holders via SSR. Disadvantages: low capital efficiency (150% collateral for 100% stablecoin); downside crypto exposure (all collateral shrinks in a market crash); oracle manipulation and smart contract bugs are real risks. Main failure mode: extreme crypto crashes (2022's largest daily drop exceeded 40%) triggering mass liquidations — but protocols with sufficient buffer (Sky's 150% min + 130% threshold) have historically survived.
Algorithmic types try to 'replace reserves with code': holding no external assets, using mint/burn mechanisms and arbitrage incentives to automatically maintain the $1 peg. UST/LUNA is the largest case and most persuasive failure: in May 2022, once confidence reversed, a death spiral took UST from $1 to near-zero in days, wiping out tens of billions in market cap. Core problem: the algorithmic stablecoin's '$1 floor' was never an external asset — it was market confidence. When confidence collapsed, the floor vanished. Hybrid types (like Frax) try to reduce risk with 'partial real reserves plus partial algo mechanism,' more robust than pure-algo but still facing the 'can the algo half hold up in a panic?' question. Another hybrid is Ethena USDe: using a delta-neutral derivatives strategy (spot ETH long + perpetual short) to simulate 1:1 dollar backing without traditional banks, but introducing CEX counterparty risk and funding rate market risk.
Choosing a stablecoin is fundamentally choosing 'which risk you can most accept.' Fiat-backed (USDC/USDT): you trust the issuer and regulators — you get maximum liquidity and institutional trust, at the cost of centralization and zero yield to holders. Crypto-overcollateralized (USDS): you trust code and crypto market health — you get decentralization and partial yield sharing, at the cost of capital inefficiency and market volatility exposure. Algorithmic: you're holding confidence itself — potentially higher yield, but if confidence breaks, the floor disappears. Practical guidance: let use case determine choice. Frequent trading → deepest liquidity in USDC/USDT; long-term holding with yield → sUSDS (stable yield) or sUSDe (market-linked yield); censorship resistance matters → USDS over USDC; avoid purely algorithmic types unless you fully understand the mechanism and risks.