I used to read “depeg” the way most people do. A stablecoin drifts off its dollar, a red line crosses the $1.00 mark, the chart gets ugly for a few hours, and then it snaps back. Price story. Bad afternoon. Buy the dip.
After three of these things, I don’t believe that anymore.
Here’s the sentence I keep coming back to, and it’s the whole spine of this piece: a depeg is not a price event. It is a redemption event. The price is the symptom. The actual failure is in the channel that lets you turn the token back into a real dollar. When that channel is open, the market can scream all it wants and the peg holds. When that channel closes, no amount of “look at our reserves” matters, because you can’t redeem and everyone knows you can’t.
So when someone asks me “what happens when a stablecoin loses its peg,” my answer is not a number. It’s a sequence. And I’ve watched this exact sequence three times now, and every time the same hidden step is where it breaks.
Let me walk the three, one chain at a time.
Chain 1 — Iron Finance (June 2021): the algorithmic one that ran out of buyers
Iron Finance’s IRON was a partly-collateralised algorithmic stablecoin on Polygon. Backed by roughly 75% USDC and 25% TITAN, using something called a Target Collateral Ratio. TITAN was the protocol’s own token — and that detail is where the whole story turns.
The trigger: TITAN ran from about $10 on June 9, 2021 to an all-time high of $64.19 on June 16, 2021, then went to roughly $0 in about two days. The bank run hit on June 16, 2021. Not weeks of decline. Two days.
Who ran first: the big accounts. Honestly, this is the part people wave away and it’s the most important part. The Federal Reserve studied this run with transaction-level data in its FEDS Notes (Adams & Ibert, June 2, 2022), and what they found was blunt — the largest and most sophisticated accounts ran first. Not the retail crowd. The whales. The ones who understood the mechanism best and knew it couldn’t hold.
What was supposed to catch it: the Target Collateral Ratio — a backing that was mostly USDC and partly TITAN. As long as TITAN held value, the collateral ratio held, and IRON held.
Why it didn’t: TITAN was the thing that had to hold, and TITAN was the riskiest asset in the whole system. The mechanism depended on the price of its own token staying up while everyone tried to leave at once.
What actually stopped it — or didn’t: nothing in the mechanism. IRON itself fell roughly 30% almost immediately, trading around $0.70. Losses were estimated at about $2 billion, and it’s widely cited as the first large-scale DeFi “bank run.” A smart-contract malfunction temporarily blocked redemptions during the panic — which is exactly the thing I keep talking about. The channel jammed right when people needed it most. Mark Cuban was an early investor and liquidity provider and publicly discussed being hit. Iron Finance insisted it wasn’t a rug pull but a bank run caused by the algorithmic design. I’ll take them at their word on the “not a rug pull” part. It doesn’t change the outcome.
Chain 2 — TerraUSD / UST (May 2022): the backstop that fed the fire
UST was a fully algorithmic stablecoin, held to $1 by a mint-and-burn arbitrage with LUNA. No hard reserves backing it. The design said: if UST drops below a dollar, arbitrageurs burn UST to mint LUNA, and the price comes back. Elegant on paper.
The trigger: UST began slipping below $1 on May 7, 2022, and over $2 billion of UST was withdrawn or de-staked around May 7–8.
Who ran first: same pattern as Iron. The people closest to the mechanism, moving in size, before the crowd understood there was a problem.
What was supposed to catch it: the Luna Foundation Guard’s Bitcoin war chest. A real, hard reserve. LFG held up to 80,394 BTC, and the plan was to deploy it to defend the peg.
Why it didn’t: watch what actually happened to that “backstop.” LFG sold 52,189 BTC on May 8 and another 33,206 BTC on May 10 — leaving only 313 BTC by May 16, about a 99.6% depletion. It didn’t work. And here’s the part that I think most people still get wrong: the crisis was never a liquidity problem that BTC could solve. It was a confidence problem turning into a reflexive deleveraging. The Bitcoin backstop answered a question nobody was asking.
Worse, the stabilising mechanism did the opposite of its job. Redemptions minted LUNA. LUNA supply went from roughly 343 million tokens on May 7 to about 6.5 trillion by May 14. The coin that had to hold value to keep UST at a dollar was being printed by the very act of defending it. LUNA fell from about $80 to below $1 by May 11 and to fractions of a cent by May 13–14. UST fell to about $0.10 by May 14 (later around $0.01–0.02). The Terra chain was halted twice, around May 10 and May 12.
What actually stopped it — or didn’t: nothing stopped it. Roughly $40 billion of value was destroyed (UST-portion estimates around $18 billion; combined ecosystem losses quoted at $40–60 billion). And in December 2025, Terraform Labs co-founder Do Kwon was sentenced in New York to 15 years in prison for fraud by US District Judge Paul Engelmayer, who called it “a fraud of epic generational scale.” Kwon had pleaded guilty in August 2025 and agreed to forfeit more than $19 million.
So no. Not a mechanism failure. A design that minted more of the very token whose value had to hold.
Chain 3 — USDC (March 2023): the peg that broke while the reserves were fine
This is the one that changed how I think about the whole subject, because it’s the one where the money was never actually lost.
On March 10, 2023, Silicon Valley Bank was closed by California regulators, with the FDIC appointed receiver. Circle disclosed that about $3.3 billion of USDC’s reserves — roughly 8% of the reserve portfolio — sat at SVB and could not be withdrawn.
The trigger wasn’t a bad loan book at Circle. It was a bank failing on a Friday.
Who ran first: here’s the twist. Because it was a weekend, primary-market redemptions effectively stalled — banks aren’t open on Saturday. So the selling pressure had nowhere to go except secondary markets. And secondary markets price fear, not fundamentals.
What was supposed to catch it: the reserves. USDC was supposed to be the boring, fiat-backed, regulated one. It was.
Why it didn’t: the redemption channel closed. Not because the collateral was bad — because the plumbing that turns USDC back into dollars runs through banks, and banks were shut. So USDC fell to a low of roughly $0.87 — a peak deviation of about 12–13% below peg, a low print of about $0.8789 — on March 11, 2023. That was USDC’s largest-ever deviation. The reserves were fine. The channel was closed. That is the entire lesson in one weekend.
What actually stopped it — or didn’t: On March 12, 2023, the Treasury, Federal Reserve and FDIC invoked a systemic-risk exception to make all SVB depositors whole. That’s what stopped it. Not the market finding its footing. A specific policy decision. USDC recovered to about $1.00 by March 13, 2023; Circle said it had redeemed $3.8 billion on the Monday. Contagion hit USDC-collateralised coins like DAI and FRAX (and Curve’s 3pool saw record volume), and it all unwound on the same recovery.
I want to be very careful about the takeaway here, because it’s the one that gets people hurt. The reserves turned out to be fine, yes. But the peg broke anyway, and it was restored by a federal backstop. A federal backstop is not the base case. Do not build your mental model of USDC on the assumption that the Treasury shows up every time.
The chain that repeats
Three different designs — partly-collateralised algorithmic, fully algorithmic, and plain fiat-backed. Three very different reserve situations. What was the same?
Here’s the sequence, and I’d bet it holds for the next one too:
- Confidence cracks. Some event — a token price, an exploit, a bank failing — makes people ask “can I still get my dollar back?” The question alone is enough. Nobody needs proof first.
- The informed run first. The largest and most sophisticated accounts move before anyone else. I’ve now seen the Fed’s own data confirm this on Iron, and the pattern repeat on Terra.
- The redemption channel closes or jams. On Iron, a smart-contract malfunction blocked redemptions. On Terra, the mechanism itself turned redemptions into hyperinflation-by-minting. On USDC, it was literally a weekend and the banks were shut. Three totally different causes, one identical effect: you cannot get out.
- Price gaps. Only now does the chart move. The price isn’t leading the failure. It’s reporting it, late, from the secondary market.
Read that again in order. The price move is step four. Almost everyone watches step four and calls it the event. It isn’t. It’s the echo.
Side by side
| Case | Collateral model | What the price did | What actually broke | What stopped it |
|---|---|---|---|---|
| Iron Finance (June 2021) | Partly-collateralised: ~75% USDC + ~25% TITAN, via a Target Collateral Ratio | IRON fell ~30% almost immediately, to around $0.70; TITAN went from $64.19 on June 16 to ~$0 in about two days | The collateral included TITAN — the system’s own token — so a run on TITAN broke the backing; a smart-contract malfunction temporarily blocked redemptions | Nothing in the mechanism; ~$2 billion in losses |
| TerraUSD / UST (May 2022) | Fully algorithmic via mint-and-burn with LUNA; no hard reserves | UST fell to ~$0.10 by May 14 (later ~$0.01–0.02); LUNA went from ~$80 to below $1 by May 11 | No hard reserves; the BTC “backstop” addressed liquidity while the crisis was confidence; redemptions minted LUNA, ~343 million tokens on May 7 → ~6.5 trillion by May 14 | Nothing; ~$40 billion destroyed; chain halted twice; LFG’s BTC drained from up to 80,394 to 313 by May 16 |
| USDC (March 2023) | Fiat-backed; ~$3.3 billion — roughly 8% of the portfolio — sat at SVB | USDC fell to a low of ~$0.87 (peak deviation ~12–13%; low print ~$0.8789) on March 11, 2023 | The redemption channel: a weekend meant primary redemptions stalled, so selling moved to secondary markets | A policy decision — on March 12, 2023 the Treasury, Fed and FDIC invoked a systemic-risk exception; USDC back to ~$1.00 by March 13 |
Look across that table and the thing that jumps out is that the “what actually broke” column has three completely different answers, and the “why it got bad” is the same answer three times.
A peg holds on the day you can still redeem.
Not on the day the reserves look fine.
What I actually watch now
I stopped checking whether a stablecoin’s reserves “look fine.” They looked fine for USDC on March 11, 2023. What I check instead is all about the channel:
- Can I redeem right now, today, and through whom? If the answer depends on a bank being open, I’ve just learned that my access is a business-hours privilege.
- Who is allowed to redeem directly, and who has to use the secondary market? If the primary channel is a short list of institutions, then “the market price” is the price for people like me, and it can diverge from par without anyone’s reserves being wrong.
- What is the collateral made of, and does any of it depend on the system’s own token? Iron answered this the hard way. Any stablecoin backed partly by its own ecosystem token has a floor made of the same material as its ceiling.
- What breaks first — the asset or the plumbing? Terra broke the asset. USDC broke the plumbing. Both took the peg down.
I’ll say the honest limit out loud: this is my personal read of public records — filings, central-bank research notes, court outcomes and price history — and it’s written to explain how failures happen, not to tell you what to hold. This is not financial advice.
But if you take one thing from three depegs, take this: a peg holds on the day you can still redeem, not on the day the reserves look fine. The failure sequence is always the same — confidence, then redemption, then the channel closes, then the price gaps. And by the time you’re watching the price, you’re watching the last step of a chain that already finished.
Watch the door, not the chart.
Related reading: What Is a Stablecoin? Three Designs Under One Word · Where Stablecoin Yield Comes From
(The End)
