Bitcoin Lending After Celsius: What Changed
June 2022. Celsius Network freezes withdrawals. $4.7 billion in customer funds, mostly Bitcoin and Ethereum, locked behind a message saying the company needed time to "stabilise liquidity." Most customers never got their full allocation back. Some are still waiting for bankruptcy proceedings to resolve three years later.
I have thought about Celsius a lot while building Ducat. Not because I want to dwell on it, but because it is the clearest possible illustration of why custody structure matters in lending.
What actually happened at Celsius
Celsius presented itself as a savings and lending platform for crypto. You deposited Bitcoin, they paid you yield. They lent that Bitcoin to institutional borrowers and kept the spread. The yield was real. The business model appeared to work.
What was less visible: Celsius was redeploying customer Bitcoin into DeFi protocols, chasing higher yields to fund those customer payouts. When DeFi positions went wrong (the stETH depegging and the collapse of the Luna ecosystem hit Celsius particularly hard), the yield machine broke. They had locked funds in positions they could not exit quickly. When customers wanted their Bitcoin back, Celsius could not provide it.
The June 12, 2022 withdrawal freeze was not a regulatory action or a hack. It was a liquidity crisis caused by a custodial model where customer funds were being used as a working balance in risky strategies.
What changed in the industry
Three things shifted meaningfully after Celsius.
Proof of reserves became expected. Before Celsius, most platforms published nothing about their actual reserve positions. After, the pressure to provide audited or on-chain proof of reserves became significant. Ledn's Armanino audit, Nexo's various compliance reports, and Kraken and Coinbase's reserve disclosures became baseline expectations rather than differentiators.
Non-custodial gained credibility. The philosophical case for non-custodial had always been there: if you do not give someone your Bitcoin, they cannot misuse it. Celsius made that argument viscerally real. Suddenly the abstract principle was illustrated with billions of dollars of customer losses.
Institutional participants got more cautious. Family offices, corporate treasury functions, and high-net-worth individuals who had been comfortable with custodial CeFi became much more sceptical. Many moved to stricter segregated custody arrangements or stopped lending Bitcoin entirely.
The limits of proof of reserves
Proof of reserves is better than no proof of reserves. But it is not sufficient.
Proof of reserves tells you that a company holds assets today. It does not tell you what liabilities those assets are supposed to cover. It does not tell you whether those assets might be locked in DeFi positions that cannot be immediately redeemed. Celsius could have published a proof of reserves the week before they froze withdrawals and it might have shown adequate Bitcoin holdings. The problem was the liabilities and the liquidity mismatch, not the nominal balance.
This is why non-custodial is architecturally more honest than custodial-plus-proof-of-reserves. With non-custodial, there is no company balance sheet to hide a mismatch in. Your Bitcoin is in your vault. The protocol's liabilities are the UNIT you borrowed. The whole thing is on-chain and visible.
Why I built Ducat with non-custodial as the foundation
When I started working on Ducat, there were two design choices that were never really debated within the team.
First: Bitcoin L1. If you are going to build a lending protocol for Bitcoin holders, it should stay on Bitcoin. Bridging to another chain introduces dependencies on entities and systems that are not Bitcoin.
Second: non-custodial. The user's Bitcoin should never be under the control of a company. Not because companies are necessarily bad actors, but because companies are failure points. They can be hacked, they can be mismanaged, they can be regulated out of existence, and they can make mistakes. Protocols that remove the company from the custody chain remove that category of risk.
Celsius would have been structurally impossible to replicate with Ducat's architecture. There is no Ducat treasury that holds customer Bitcoin. There is no Ducat lending desk that redeploys your collateral into yield strategies. When you deposit BTC into a Ducat vault, it sits there, locked by cryptography you control, until you repay the loan or the protocol liquidates it according to transparent on-chain rules.
What the lending market looks like now
The platforms that survived 2022 and are still operating today are, with some exceptions, the more conservatively run ones. Ledn's survival (and their behaviour during the crisis, which was notably better than most) gave them credibility. Nexo survived. Coinbase never really entered the consumer lending market aggressively enough to be at risk.
New entrants like Arch, Strike, Lava, and Ducat entered a market where the survivors and new builders had all absorbed the same lesson: do not commingle customer funds with operational risk.
The market is healthier than it was in 2021. That does not mean risk-free. It means the obvious failure modes have been addressed, at least partly.
For a full comparison of current platforms and their structures, see our . For the mechanics of how non-custodial loans actually work, see .
This piece reflects the author's perspective on industry events and is not financial advice. Celsius bankruptcy proceedings are ongoing; details in this article reflect information available as of April 2026.
