---
title: "How ten years of failures selected this market"
author: Mariusz Szyma
date: 2026-09-15
lang: en
canonical: https://szyma.co/en/blog/stablecoins/history/
series: "Stablecoins: Past, Present and Future"
series_part: 2/11
series_url: https://szyma.co/en/blog/stablecoins/
data_as_of: 4–10 Sep 2026
---

# How ten years of failures selected this market

Today's dollar tokens are leftovers of a selection process, not the first designs anyone tried. This chapter walks that process from BitUSD in 2014 to the first crash-free contraction in 2026, and the rule that keeps repeating: the reserve and the loop have to be independent of the token they defend.

## What you'll learn

- Walk ten phases from 2014 to 2026 and name, for each one, what was built, what it fixed, and what broke.
- Tell a depeg that came back from one that did not, using the state of the redemption loop rather than the depth of the print.
- Keep two loss figures for Terra and two for Black Thursday, because the sources measure different things.
- Place the May 2026 peak as three disagreeing readings rather than one settled all-time high.

## Before Tether won

Crypto markets in 2014 had no low-volatility asset to settle between trades. Three projects tried to supply one in the same year.

**BitUSD**, on BitShares, launched on 21 July 2014 as a crypto-collateralized token, posted against the chain's own BTS token at 2:1. **NuBits**, presented on 23 September 2014, used a two-token seigniorage design: NuBits as the dollar token, NuShares as the governance token that was supposed to absorb supply. **Realcoin** launched on 8 July 2014 in California and renamed itself Tether on 20 November 2014. The first 100 USDT were minted on the Omni layer over Bitcoin on 6 October 2014; trading started on 25 February 2015. Tether is the design that is still here.

BitUSD held a dollar peg for months in a quiet market. So did early DAI, later, through a year in which ether fell more than 80%. The failure was elsewhere. BTS was not liquid independently of the token it was supposed to secure. When BTS fell hard in 2018, BitUSD went into an emergency procedure called global settlement on 25 November 2018, because the collateral was no longer enough. It never recovered the peg. NuBits broke twice: a three-month gap from May and June 2016, and a second break in March 2018 that printed 0.25 USD and ended the project.

> **My take:** Collateral that moves with the token it secures is not collateral. It is a feedback loop waiting for a bad day. The market did not take that lesson. Eight years later Terra rebuilt the same loop two orders of magnitude larger.

![Ten phases of stablecoin history, 2014–2026, sorted into fiat-backed, crypto-backed, algorithmic and regulatory lanes, from BitUSD on 21 Jul 2014 to the GENIUS Act on 18 Jul 2025.](https://szyma.co/blog/img/stablecoin-timeline-2014-2026.e24ea832.svg)

## Tether scaled without an audit

Crypto exchanges in 2015 and 2016 could not hold ordinary dollar accounts. Banks closed them. USDT gave traders a dollar proxy that moved on a chain, twenty-four hours a day, without a SWIFT ticket on the trader's side. A listing on Bitfinex gave it a deep book. The sentence "one token, one dollar in reserve" was easy to repeat. That was enough to win, with the worst reserve transparency in the category.

Scale ran ahead of supervision. In January 2017, 10 mn USDT were in circulation. In January 2018 alone Tether issued 850 mn new tokens, more than in any previous month. On 27 January 2018 it ended the relationship with its auditor, Friedman LLP, saying a full audit of a "relatively simple" balance sheet would be "unachievable in a reasonable time." In November 2017 a treasury hack took more than 30 mn USDT and forced a hard fork that added a freeze.

The penalties arrived later and they are countable. In 2021 Tether and Bitfinex paid 18.5 mn USD to settle with the New York Attorney General over misleading statements about USDT's backing. The same year the Commodity Futures Trading Commission fined Tether 41 mn USD. The CFTC's sample, 26 months from 2016 to 2018, found the claim of full fiat backing true on 27.6% of days.

USDT still never lost its peg for good. On 4 September 2026 DefiLlama put it at 58.91% of USD-pegged supply. CoinDesk Data and Artemis, in 2026, put it at about 74% of centralised-exchange volume against that 59% supply share. The market paid for depth and settlement speed, not for an audit. Institutional redemption, per Kaiko on 21 April 2026, needs a verified account and about 100,000 USD. The small holder does not close the loop; desks do.

A regulated rival arrived in 2018 anyway. USDC (then under the Centre consortium), TrueUSD, Paxos Standard and Gemini Dollar came out under New York Department of Financial Services oversight, the first de facto standard for a fiat-backed token that wanted a supervisor in the room. Circle built the more transparent book. Tether kept the volume.

## DAI and the day the auctions filled at zero

MakerDAO's first white paper, December 2017, described Dai as a token minted against ether locked in Collateralized Debt Positions. On 18 November 2019 the system became Multi-Collateral DAI and added the Dai Savings Rate. That spread the collateral risk. It did not fix the auction.

**Black Thursday** is 12 March 2020, the day the liquidation engine ran as written and still wrecked its users. One reading has ether falling from 194 to 111 USD in a day. Another has 197 to 89 USD in under thirty-six hours. Both stay here. Ethereum congested; gas printed above 200 gwei; MakerDAO's price oracle went stale; when the feed finally moved, the reported price dropped more than 20% in one step and hundreds of positions liquidated together.

Liquidation auctions needed a ten-minute window for a competing bid. At that gas price, competing keepers could not get in a block. One bot noticed it could bid 0 DAI and win. In about forty minutes that bot took 8.32 mn USD of ether for free. Of 3,994 liquidation transactions that day, 1,462 (36.6%) filled at a 100% discount. The hole in the system is recorded as 4.5 mn USD of uncovered debt in one source and 5.4 or 5.67 mn USD in others. MakerDAO's own documents, in the material used here, do not pick one. The 8.32 mn USD figure counts ether sold at zero. The 4–6 mn USD figures count uncovered debt. They are different measurements, not a disagreement about one number.

Governance voted down an emergency shutdown and minted 20,980 MKR, auctioned for 5.3 mn DAI, to refill the gap. A compensation vote for the liquidated vaults failed: more than 65% of 38 votes cast said no. Owners of those vaults lost 100% of the collateral, not only the surplus needed to recollateralize.

> **Watch out:** A later chapter takes DAI and USDS apart as a peg mechanism. The lesson this chapter needs is narrower. The code did what the code said. The assumption that competing keepers would always appear was false on the day the network was least able to carry them.

## Terra's death spiral

**A death spiral**, in this market, is a mint-and-burn loop that, once the peg breaks, prints the sister token faster than any buyer can absorb it, so each new unit cheapens the "reserve" behind the next one.

TerraUSD (UST) was an algorithmic dollar, paired with LUNA. When UST traded below a dollar, arbitrageurs were supposed to burn UST and mint LUNA, which would lift UST back. Demand came mostly from Anchor Protocol, which paid about 19.5–20% APY on UST deposits, a rate subsidised from Terraform Labs' treasury rather than from loan demand. At the April 2022 peak, sources put UST at 17.5 bn USD or 18.7 bn USD; about 75% sat in Anchor. LUNA's price is recorded at 117 USD in one source and 119.18 USD in another, with a market value above 40 bn USD. The two UST figures are different read dates, not two methodologies.

On 7 May 2022 two large addresses pulled 375 mn UST from Anchor, and Terraform Labs pulled 150 mn UST from a Curve pool ahead of a new pool. By 9 May about 5 bn USD of 14 bn USD left Anchor in a day, 35% of the deposits. From 6 to 13 May Anchor's TVL fell from 16.75 bn USD to 0.314 bn USD, more than 98%. LUNA's supply went from about 343 million tokens to more than a trillion in 48 hours.

Luna Foundation Guard spent its reserve, 80,394 BTC, about 2.8 bn USD at May prices, between 8 and 12 May. The bitcoin was absorbed. By 13 May LUNA and UST had flatlined. Destroyed value is recorded as more than 40 bn USD in one set of sources and 45–60 bn USD in others that count a wider ecosystem. All three figures stay; they are not averaged.

The CeFi cascade is the transmission, not the cause. Three Arrows Capital went into liquidation with 3.5 bn USD of debt. Voyager Digital had lent it 15,250 BTC and 350 mn USD of USDC. Celsius froze 4.7 bn USD of customer funds and filed on 13 July 2022. Do Kwon, extradited from Montenegro, pleaded guilty in December 2025 and is serving fifteen years.

Panic transmitted the failure. The construction caused it. LUNA had no value independent of UST, the same defect BitUSD had shown in 2018, with a subsidised yield large enough that most of the capital never asked where 20% came from.

## One month, two failures

March 2023 produced two different endings in four weeks. One token depegged and came back. One was ordered to stop.

Silicon Valley Bank, one of Circle's six banking partners, hit a classic deposit run after announcing on Wednesday 8 March 2023 that it needed 2.25 bn USD of new capital. Regulators closed it on 10 March 2023. Shortly after 22:00 ET that day Circle confirmed that 3.3 bn USD of about 40 bn USD of USDC reserves, about 8%, sat at SVB. USDC lost the peg almost immediately. At the time 77% of the reserve (32.4 bn USD) was short-term US Treasury bills. There was no commercial paper of Chinese banks, no gold, no bitcoin in that book. By 02:00 on 11 March the price printed 0.87 USD; CCCAGG has 0.8726 USD. Banks are closed at the weekend. Redemption could not run. The secondary market priced the worst case.

DAI, then about 40–43% collateralised by USDC through its Peg Stability Module, printed 0.88 USD; 563 mn DAI were burned in 24 hours. USDP printed about 0.91 USD, GUSD about 0.96 USD. The module built to make arbitrage easy became the pipe that carried the contagion.

On Sunday 12 March the Fed, the Treasury and the FDIC announced a systemic-risk exception covering SVB deposits above the 250,000 USD cap. On Monday 13 March USDC printed 0.9918 USD, and par in the days after. More than 1 bn USD of USDC had already been redeemed on the Friday. [Chapter 1](/en/blog/stablecoins/what-one-usdc-means/) is the three-layer reading of that weekend. This chapter only needs the selection result: a fully reserved, mostly Treasury-backed token can still print 0.87 USD if the bank that holds the cash is closed.

The other failure that month was not a depeg. Paxos issued BUSD for Binance. On 13 February 2023 the New York Department of Financial Services told Paxos to wind the product down, over due-diligence on the Binance relationship, not over reserve quality; Bloomberg put BUSD at about 16 bn USD then, third-largest in the category. The mint halt took effect on 21 February 2023. Holders could redeem through February 2024 or convert to USDP. A later fine of 26.5 mn USD is in the author's knowledge base and not in the Polish edition or the September 2026 research log, so it sits here as that knowledge-base item rather than as a confirmed public order.

> **Example:** USDC's 13% gap lasted about three days because the remaining 36 bn USD of reserve was still there and the desk reopened on Monday. BUSD did not depeg on a print. A supervisor closed the mint. Depth of a price chart would have predicted neither ending.

## What the depth of a depeg does not tell you

**A depeg** is a market price that leaves the target by enough, and for long enough, that the event gets a date. Depth versus shallowness is the wrong split. Return versus no return is the one that maps onto the loop.

![Twelve depeg episodes by maximum depth below $1. UST, deUSD, xUSD and BitUSD never returned; USDC's 13% gap in Mar 2023 did, because the redemption loop reopened. Ranges plotted at the high end; DAI's Mar 2020 reverse depeg omitted.](https://szyma.co/blog/img/depeg-episodes.572b4be7.svg)

USDC fell 13% and was back in about three days. UST fell 98–99% and never came back. deUSD fell 97–98.5% and never came back. USDe on one exchange fell 35%, deeper than USDC, and recovered in hours because the on-chain peg had not actually broken; the same episode is about 1% on-chain. USDT, in the UST contagion of May 2022, is recorded as 4–8% depending on the venue, and was back in hours to a day. Tether did not close redemptions in any contagion episode in this material. Elixir and Stream did, once they had established that the backing was gone.

The variable that decides is whether, at the moment of stress, a liquid reserve existed independently of the token, and whether the redemption channel was still operating, even if it was gated. USDC had 3.3 bn USD frozen and more than 36 bn USD still liquid. That was enough. UST had nothing independent of LUNA. deUSD and xUSD held each other: Elixir's deUSD had about 65% of its collateral (about 68 mn USD of about 150 mn USD of capitalisation) lent to Stream, and Stream held about 90% of remaining deUSD supply. A loss at Stream of 93 mn USD, disclosed on 4 November 2025, closed that circle. xUSD printed as low as 0.10–0.20 USD; deUSD printed 0.015–0.03 USD; USDX followed below 0.60 USD. None of the three recovered a peg in the period this material covers. About 285 mn USD of linked debt sat in the cluster, and about 1 bn USD left DeFi yield products in a week.

November 2025 is BitUSD 2018 rebuilt inside DeFi, small enough that it did not make the Terra headlines and large enough to show the defect had not been retired.

## A peak, then a contraction with no crash

PYUSD launched on 7 August 2023. Ethena's USDe launched on 19 February 2024, reached 1 bn USD of supply in ten weeks, and peaked above 14 bn USD in 2025; the Polish edition puts the October 2025 peak at about 14.7 bn USD and the 4 September 2026 DefiLlama reading at 4.28 bn USD. MakerDAO rebranded to Sky on 27 August 2024 (MKR to SKY at 24,000:1, DAI to USDS at 1:1). USDS is recorded at 11.7 bn USD by the end of the first quarter of 2026; USDS plus DAI together were 11.4 bn USD on DefiLlama on 4 September 2026.

MiCA's e-money-token and asset-referenced-token rules applied from 30 June 2024, CASP rules from 30 December 2024. USDT was delisted from major regulated EU exchanges. Circle received a French electronic-money-institution licence in July 2024, the first global issuer the Polish edition records as MiCA-compliant. The GENIUS Act was signed on 18 July 2025. The regime-by-regime reading of both statutes sits in a later chapter.

On 17 May 2026 total stablecoin capitalisation is recorded three ways, and they are not reconciled. The author's DefiLlama read, used in the Polish edition, is 322.4 bn USD (322.43 bn USD in the knowledge base). DefiLlama's own historical chart, read on 10 September 2026, shows 320.87 bn USD on that same 17 May date. A KuCoin report citing CoinDesk puts about 322 bn USD on 26 May 2026, a different day. All three stay.

By 13 August 2026 the total was 308.0 bn USD, 4.5% below the author's May peak and still up 14.3% year on year. On 4 September 2026 DefiLlama put all pegs at 312.0 bn USD, of which 311.2 bn USD (99.7%) were USD-pegged, 3.3% below that May reading. Earlier contractions in this chapter have a dated crash underneath them. This one does not, in the sources. Three hypotheses circulate (policy rates versus unyielding payment tokens, DeFi outflows, rotation into tokenized Treasuries). None is sourced as the cause. They are correlations in time.

![Five dated readings of total stablecoin supply, from 4.171 bn USD in Jan 2020 to the author's 322.4 bn USD on 17 May 2026 (DefiLlama's chart: 320.87 bn USD that day), then 312.0 bn USD on 4 Sep 2026. Not the daily series.](https://szyma.co/blog/img/stablecoin-supply-2019-2026.116d63da.svg)

On 4 September 2026 the mechanism split was: fiat-backed 90 tokens and 285.04 bn USD; crypto-backed 221 tokens and 25.50 bn USD; algorithmic 23 tokens and 0.67 bn USD, under 0.3% of the category. The market punished the design that went to zero. It did not, on this record, reward the design that published the cleaner reserve page.

## Key takeaways

1. Three 2014 designs tried to make a dollar on a chain; BitUSD and NuBits died on self-referential collateral, and Tether's fiat-reserve model is the one that still dominates.
2. Tether issued 850 mn USDT in January 2018, ended its auditor on 27 January 2018, paid 18.5 mn USD to the NYAG and 41 mn USD to the CFTC in 2021, and on a 2016–2018 CFTC sample was fully fiat-backed on 27.6% of days; it did not permanently depeg.
3. Black Thursday, 12 March 2020, liquidated 8.32 mn USD of ether at a zero bid; uncovered debt is recorded as 4.5 to 5.67 mn USD. Both measurements stand.
4. UST's April 2022 peak is 17.5 or 18.7 bn USD depending on the read date; the May collapse is recorded as more than 40 bn USD or 45–60 bn USD depending on scope. LFG spent 80,394 BTC and did not hold the peg.
5. In March 2023 USDC printed about 0.87 USD with 3.3 bn USD stuck at SVB and a 77% Treasury reserve still intact; BUSD was halted by NYDFS over the Binance relationship, at about 16 bn USD of supply, not over a hole in the reserve.
6. Depth of a depeg does not predict return. USDC's 13% gap reversed; UST's 98–99% and deUSD's 97–98.5% did not. The open, independent reserve and the open desk did.
7. The May 2026 peak is 322.4 bn USD (author), 320.87 bn USD (DefiLlama chart, same date), or about 322 bn USD (KuCoin/CoinDesk, 26 May). The 3.3% contraction to 312.0 bn USD on 4 September 2026 has no documented single-issuer crash underneath it, and no sourced cause.

## Glossary

- **Depeg**: a market price that leaves the target by enough, and for long enough, to get a date. Return is decided by the loop and the reserve, not by how far the print went.
- **Black Thursday**: 12 March 2020, when MakerDAO's liquidation auctions filled at zero DAI because the network could not carry competing keepers.
- **Death spiral**: a mint-and-burn loop that, once broken, prints the sister token faster than the market can absorb it. UST/LUNA in May 2022 is the type case.
- **Global settlement**: BitUSD's emergency close on 25 November 2018, after BTS collateral was no longer enough. The peg did not return.
- **Orderly wind-down**: a supervisor or issuer stopping mint and running down supply on a timetable, as with BUSD after the NYDFS halt of 21 February 2023. Contrasted here with a collapse that has no timetable.
- **MiCA**: the EU's Markets in Crypto-Assets regime. EMT and ART rules from 30 June 2024; CASP from 30 December 2024. Expanded in a later chapter.
- **GENIUS Act**: the US payment-stablecoin statute, signed 18 July 2025. Defined in [chapter 1](/en/blog/stablecoins/what-one-usdc-means/); the regime map sits later.
- **Seigniorage stablecoin**: a two-token design that tries to hold a peg by minting and burning a sister token rather than by holding an independent reserve. NuBits and UST are this family.

## Go deeper

- [What one USDC actually equals](/en/blog/stablecoins/what-one-usdc-means/): the three layers and the three questions about the loop that this chapter's episodes keep answering.
- [Regulation: GENIUS, MiCA and the rest of the map](/en/blog/stablecoins/regulation/): the statutes that close this timeline as licence regimes rather than as press releases.
- 🟢 CFTC.gov, order against Tether Holdings Limited, 15 October 2021: the 41 mn USD penalty and the 27.6% finding.
- 🟢 Circle press releases, 11–12 March 2023: 3.3 bn USD at SVB, about 8% of then-reserves.
- 🟢 DefiLlama API, read 4 September 2026, and DefiLlama historical chart read 10 September 2026: the two May 2026 peak figures.

## Sources

- 🟢 Author's knowledge base, module 2 (ten-phase table, Tether scale-up, 2018 regulated cohort, MakerDAO dates, DeFi Summer supplies, Terra chronology, March 2023 pair, 2023–2026 products and GENIUS Act) and module 6 (depeg-episode table).
- 🟢 Polish research edition, `content/blog/stablecoins/wojny.md` and `content/blog/stablecoins/gracze.md` (BUSD), `content/blog/stablecoins/legislacja.md` (MiCA dates, Circle's French EMI licence).
- 🟢 `content/blog/en/research/state-2026-09.md` §(e): DefiLlama chart 320.87 bn USD on 17 May 2026; KuCoin/CoinDesk about 322 bn USD on 26 May 2026.
- 🟢 `content/blog/en/research/companies.md`: BUSD mint halt effective 21 February 2023; about 16 bn USD at the NYDFS action; redemption through February 2024.
- 🟢 CFTC.gov, 15 October 2021; NYAG settlement coverage as cited in the Polish edition (Banking Dive, 24 February 2021).
- 🟢 Bitcoin Magazine, 8 February 2018: 850 mn USDT in January 2018; Friedman LLP ended 27 January 2018.
- 🟢 MakerDAO white paper; PR Newswire, 18 November 2019, Multi-Collateral DAI.
- 🟡 Glassnode Research, 20 March 2020; Whiterabbit, 17 February 2023 (8.32 mn USD at 0 DAI); CoinDesk, 25 March 2020 (36.6% of liquidations at a 100% discount).
- 🟡 Bitstamp, 2022; Chainalysis, 9 April 2026 and 16 March 2023; Harvard Law School Corporate Governance Blog, 22 May 2023; LedgerMind, 8 June 2026.
- 🟡 Circle press releases, 11–12 March 2023; CoinDesk Data, 13 March 2023; Fed FEDS Notes, 17 December 2025.
- 🟡 Kaiko, 21 April 2026: Tether institutional redemption about 100,000 USD.
- 🟡 CoinDesk Data / Artemis, 2026: USDT about 74% of CEX volume.
- 🟡 Pharos, 2 July 2026; BeInCrypto / CryptoRank / CoinNews, 6–7 November 2025: deUSD / xUSD / USDX cluster.
- 🟡 Eco.com and Circle, July 2024: French EMI licence.
- 🔴 The 26.5 mn USD BUSD fine is a knowledge-base item not independently re-read for this chapter.

## Caveats

The 17 May 2026 all-time-high is not a single number. The author's DefiLlama read, DefiLlama's own chart, and a KuCoin/CoinDesk figure on a different day are all in circulation. Nothing here picks a winner.

Terra's and Black Thursday's losses are recorded on more than one basis. Stating a single figure would hide that.

A current PYUSD supply is not stated. The knowledge base flags an unreconciled gap between a March 2026 figure and a September 2026 snapshot; the brief for this chapter forbids picking one.

The 26.5 mn USD BUSD fine is in the knowledge base and not in the Polish edition or the September 2026 research log. The mint halt, the 16 bn USD scale, and the due-diligence rationale are the confirmed pieces.

Kaiko's 100,000 USD Tether minimum is April 2026 secondary reporting. Chapter 1 left Circle's current minimum unlocated; this chapter does not fill that gap with the Tether figure.

USDT's 4–8% May 2022 depth, USDC's 0.87 versus 0.8726 USD, DAI's 12–15% in March 2023, xUSD's 74–90%, and deUSD's 97–98.5% are ranges. The figure plots the high end of each range and this caveat names them as ranges.

The five-point supply figure is not the daily DefiLlama series of 2,804 observations cited in the Polish caption. That CSV is not in this repository. The points are the dated readings the sources actually give.

The cause of the 2026 contraction is not established. Three hypotheses are named as hypotheses.

Do Kwon's sentence and the CeFi cascade figures come through the Polish edition's secondary sources (Yahoo Finance, 21 May 2026; CNBC; Decrypt; Zircuit) and are not re-litigated here.