Terra was a blockchain with two tokens: UST (a US$1 stablecoin) and Luna (governance token). They were linked by a mint-burn mechanism: burn US$1 of Luna to create 1 UST; burn 1 UST to redeem US$1 of Luna. Demand for UST came mostly from Anchor Protocol, which paid 19.5% APY — subsidized by the treasury. In May 2022, large withdrawals stressed the mechanism: UST lost its peg, Luna was minted in hyperinflation, price crashed, UST crashed further — death spiral. In one week, US$60 billion evaporated. It is Diamond-Dybvig applied to DeFi.
The mechanism: elegant and fragile
Terra was built by Terraform Labs, founded in 2018 by Do Kwon and Daniel Shin. The proposal: create a stablecoin without depending on fiat reserves (as USDC or USDT do) — a purely algorithmic stablecoin. The mechanics: (i) user wants 1 UST, burns US$1 of Luna, receives 1 UST; (ii) user wants to sell 1 UST, burns 1 UST, receives US$1 of Luna. This created an arbitrage bridge that, in theory, kept UST at peg.
Why it worked in theory: if UST fell to US$0.98, an arbitrageur would buy UST at US$0.98, burn it for US$1 of Luna (2% profit), pushing UST price up. If UST rose to US$1.02, arbitrageur would burn Luna for UST (receiving US$1 in UST), sell at US$1.02 (2% profit), pushing UST price down. The mechanism was "self-regulating" as long as arbitrage remained profitable.
The problem: arbitrage only works in benign equilibrium. In a crisis, when everyone wants to sell UST at the same time, massive UST burning creates massive Luna minting — Luna hyperinflation tanks its price — which corrupts the very redemption mechanism (since UST is still being swapped for a Luna that is worth less and less). Technicians called this the "death spiral". Kwon and defenders argued it could never happen. In May 2022, it did.
Anchor Protocol: the artificial demand engine
An algorithmic stablecoin only makes sense if it has real demand. Terra manufactured that demand artificially via Anchor Protocol, a DeFi app that paid 19.5% annual interest on UST deposits. In a world where US Treasuries paid 0.5% and money markets 0.2%, 19.5% was absurdly generous — and with no apparent risk (dollar peg, smart contract, "DeFi institutionalized").
The outcome was obvious: tens of billions migrated to Anchor. At peak, ~75% of all UST in existence was deposited in Anchor. But because the protocol paid 19.5% while loans generated far less, there was a deficit. The LFG (Luna Foundation Guard) treasury subsidized the gap — indefinitely, in theory.
Economists warned: sustainable interest comes from real economic productivity (productive loans, tax-backed Treasuries, cash-flowing real estate). A subsidized 19.5% is not yield — it is a temporary subsidy. It attracts capital, but that capital flees the moment the subsidy is questioned. Anchor was, in essence, a sophisticated Ponzi scheme — not in a strict legal sense, but economically: sustainability depended on new entrants.
Diamond-Dybvig: the model that predicted everything
Douglas Diamond and Philip Dybvig published in 1983 one of the most important papers in monetary economics: "Bank Runs, Deposit Insurance, and Liquidity". They mathematically formalized why banks are inherently fragile even when solvent. The argument:
- Banks perform liquidity transformation: they take demand deposits (redeemable at any time) and extend long-term loans (not instantly liquidatable)
- If all depositors redeem at the same time, the bank cannot liquidate loans fast enough — and fails
- This is a self-fulfilling bad equilibrium: if I think others will run, I run too — even if the bank is solvent
- Solution: deposit insurance (FDIC in the US), lender of last resort (central bank), redemption suspension in a crisis
Terra/UST reproduced that model exactly — without any of the protections. UST was a "demand deposit" yielding 19.5%. The "reserve" was Luna, an asset whose value depended on UST's very stability (endogenous collateral). There was no FDIC, no Fed, no suspension mechanism. When the run began on May 7, there was no brake.
The run: the anatomy of 7 days
The trigger: concentrated withdrawals from Curve pool
On May 7, US$150 million in UST were withdrawn from Curve's 3pool (a decentralized market where UST traded against USDC/USDT). The pool became unbalanced; UST started trading below US$1. Arbitrage trades tried to restore the peg, but exit volume exceeded capacity. UST fell to US$0.98. In a conventional market, 2% off peg is a trivial correction. In a leveraged system with endogenous collateral, it was the fuse.
LFG tries to defend with Bitcoin
LFG had accumulated ~80,000 BTC as a "strategic reserve" for exactly this scenario. They sold Bitcoin to buy UST on the market and attempt to restore the peg. Initially it worked — UST returned to ~US$0.99. But block-sized BTC sales pressured global BTC price (falling ~15% that week), creating contagion. And LFG exhausted the BTC ammo quickly. By May 10, UST was at US$0.60.
Luna's death spiral
Since UST could be converted to US$1 of Luna via the protocol mechanism, arbitrage would theoretically correct. But that required massive Luna minting — which collapsed Luna from US$80 to US$5, then US$0.05. Users burned UST at US$0.50 for Luna at US$5; then Luna at US$0.50; then essentially nothing. Luna's hyperinflation was historic — daily supply increase above 100%. Luna went from US$30 billion market cap to zero.
Halting, fork and fallout
The Terra network was temporarily halted. Do Kwon proposed a hard fork — a new "Terra 2.0" without UST, with a Luna 2.0 airdrop to old holders. It was implemented but delivered pennies to those who had lost thousands. Anchor collapsed. Contagion spread across the ecosystem: Three Arrows Capital (3AC, hedge fund), Celsius, Voyager — all exposed to UST via yield farming or lending — failed in the following weeks. FTX briefly emerged as a "savior" by lending to some firms. Six months later, FTX itself would collapse.
Three permanent lessons
1. Endogenous collateral is not collateral
An asset is good collateral when its value is independent of the asset it is collateralizing. Real estate for a mortgage; public equity for margin; gold for reserves. Luna as "collateral" for UST failed by design: if UST fails, Luna fails (because demand for Luna came from UST via Anchor). This is called endogenous collateral or reflexivity (Soros), and it is a recipe for circular collapses. Surviving stablecoins (USDC, DAI) use exogenous collateral: Treasuries, ETH, BTC.
2. Yield above market is disguised subsidy
When a product offers yield 10x above the risk-free rate with no obvious risk, ask where the money comes from. Usually: (i) temporary subsidy (Uber subsidized drivers to scale); (ii) hidden leverage; (iii) fraud. In no case is it sustainable indefinitely. Anchor's 19.5% was a subsidy propped up by LFG reserves that could not survive a mass exit. Sophisticated investors knew — and stayed anyway, betting on timing.
3. DeFi needs the protections it took centuries to invent
The traditional banking system learned the hard way (crises of 1907, 1929, 2008) the importance of deposit insurance, lender of last resort, capital ratios, stress tests, regulatory suspensions. DeFi assumed it could reinvent finance from scratch and learned the hard way that those protections exist for real reasons. MiCA regulation in the EU (2023-2024) and GENIUS Act discussions in the US are direct responses to Terra — requiring 1:1 backing for stablecoins. Pure algorithmic design appears dead as a viable model at scale.
Want to take the decision in Do Kwon's place?
At SMX, you are CEO of Terraform Labs at 3 AM on May 9, 2022, with UST falling and Bitcoin running out. Four decisions under extreme pressure. Academic feedback.
Explore SMX →Conclusion: when technical elegance is not enough
Terra did not collapse because Do Kwon was incompetent — on the contrary, the technical design was elegant and defensible in papers. It collapsed because the mechanism was fragile to large-magnitude shocks, and the founders' arrogance in downplaying that risk prevented basic safeguards. Diamond-Dybvig predicted exactly the scenario; DeFi ignored it.
The Terra collapse was educational in the most bitter sense. It exposed: (i) the difference between backed and algorithmic stablecoins; (ii) the importance of exogenous collateral; (iii) the risk of subsidized yields disguised as DeFi-native; (iv) systemic contagion in interconnected markets; (v) the need for proactive regulation. Surviving stablecoins learned: USDC publishes monthly attestations, DAI went hybrid with RWAs, USDT improved transparency. Do Kwon was arrested, extradited and is facing fraud charges. The ecosystem moved on. But US$60 billion were transferred from investors to the collective memory of "don't trust yield that looks too good". It is the most expensive lesson ever charged in DeFi.