The CeFi lending collapses and what creditors actually got back

Most accounts of the 2022 crypto lending collapse stop at the moment the withdrawals stopped. That is the dramatic part, and it is also the least useful part, because it tells you nothing about what a creditor of one of these companies ended up holding. The mechanical detail that decides that outcome is a single date in each case: the petition date. On that day, a customer stops owning bitcoin and starts owning a US dollar claim against an estate, measured at that day's prices. Everything afterwards — every distribution, every settlement, every announcement of "100% recovery" — is denominated in those frozen dollars.

Celsius filed on 13 July 2022, near the bottom of that cycle. BlockFi filed in November 2022. Both estates later made good on very large fractions of their claims. Both sets of customers still ended up materially worse off than if they had simply held the coins, because the coins appreciated and the claims did not. Gemini Earn users were the exception, and the reason is structural rather than lucky. This page sets out the sequence, the enforcement outcomes with their correct labels, and what did and did not change afterwards.

The petition date is the whole story

When a company files for Chapter 11, the court needs a single, fixed measure of what every creditor is owed. For a crypto lender that means converting each account into US dollars at the value shown on the filing date. Celsius's own distribution documentation is explicit about this: claims were valued in US dollars at the assets showing in the account on 13 July 2022.

Consider what that does arithmetically. A customer who deposited bitcoin when it was expensive, watched the platform freeze during a drawdown, and then received a high percentage of their petition-date dollar claim two or three years later has been repaid at the worst price in the sequence. Bitcoin and ether both rose substantially after mid-2022. A 100% dollar recovery on a claim struck at the bottom buys back materially fewer coins than were deposited. The recovery percentage in the headline and the loss the customer experienced are measuring two different things, in two different units.

What each estate actually returned

The table below compares the four principal cases plus Abra on the single axis that matters most: what the customer ended up holding, and in what denomination. Figures are taken from regulatory releases, court-related distribution material and company announcements, and every line where a final number could not be verified is marked as such rather than estimated.

Four products, four different structures

The four failures are usually described as one event, and in market terms they were. Structurally they were not, and the differences explain a great deal about how each resolution went.

Celsius ran the deposits on its own balance sheet. Customer assets funded proprietary trading, uncollateralised institutional loans and activity in the CEL token, which the company also issued. When the trading losses arrived there was no separation between the customer liability and the failed strategy, which is why the estate had so little left to distribute against such a large claim pool and why the recovery had to be built up through litigation and asset sales over several years.

BlockFi also held deposits on its own balance sheet, but its losses came from exposure to identifiable counterparties — FTX, Alameda Research and Three Arrows Capital — rather than from internal trading. That distinction turned out to matter enormously, because a claim against a failed counterparty is an asset. BlockFi settled $874.5m of claims against FTX and Alameda in March 2024 and sold them to third parties at a premium. The estate did not recover crypto; it monetised litigation, which is what made a full dollar recovery possible.

Voyager failed for the same reason as BlockFi — the Three Arrows Capital default — but with a marketing problem attached that the others did not have. The representation that deposits were "FDIC-insured" and "as safe with us as a bank" became the centre of the FTC's case. Roughly $484m was recovered, including a $450m settlement with FTX.

Gemini Earn was the only one of the four where the customer-facing brand and the borrower were different companies. Gemini distributed the product; Genesis Global Capital borrowed the assets. That separation meant the recovery ran through the Genesis estate and through the New York Attorney General rather than through a single insolvent balance sheet — and it is the structural reason Earn users ended up receiving coins rather than a dollar claim.

Getting the enforcement labels right

These cases are frequently summarised in a way that blurs three legally distinct things. It is worth separating them, because they say different things about what was proven.

• A settlement resolves a claim without a contested finding. BlockFi's $100m settlement with the SEC and 32 states in February 2022 is one. So is the FTC's resolution with Celsius in July 2023, which carried a $4.7bn judgment suspended to allow remaining assets to be returned, alongside a permanent ban on handling consumers' assets. So is the June 2025 FTC settlement with Voyager's former chief executive Stephen Ehrlich: a $2.8m payment made jointly with his wife, a lifetime ban on marketing or selling crypto products to retail customers, and bans on misrepresentation and on obtaining consumer financial information deceptively.
• A civil judgment or order is a court determination in a non-criminal case. A Celsius co-founder was ordered to pay $10m in April 2026 as that case neared its end.
• A criminal conviction and sentence is a different order of finding altogether. Alex Mashinsky was sentenced to 12 years in prison on 8 May 2025 for securities and commodities fraud and ordered to forfeit approximately $48m, with his own crypto holdings redirected to creditors.

One further distinction belongs here. The SEC dismissed its case concerning Gemini and Genesiswith prejudice on 23 January 2026, following full investor recovery. Dismissal with prejudice means those claims cannot be brought again. It is the opposite of an unresolved allegation, and any summary that leaves the case open-ended is out of date.

Why Gemini Earn users were different

The Gemini Earn structure separated the distributor from the borrower. Gemini marketed the product; the assets were lent to Genesis Global Capital. When Genesis failed, Earn users were exposed to Genesis, not to Gemini's balance sheet, and the resolution ran through a combination of the Genesis estate and pressure from the New York Attorney General. The outcome Gemini announced in May 2024 was the return of Earn users' digital assets in full and in kind — over $2bn of actual coins.

That is why this one case looks so different in retrospect. An in-kind return means the customer participates in whatever the asset does between the failure and the distribution. A dollar claim means they do not. It is not a reward for choosing a better platform; it is the consequence of how that particular estate was resolved. A reader should draw exactly one general lesson from it: the form of the recovery is at least as important as its percentage, and neither is knowable in advance.

What changed afterwards, and what did not

Three things genuinely changed. First, the enforcement position on staking-as-a-service hardened and then softened: Kraken paid $30m and shut its US staking programme in February 2023, then relaunched a US staking product on 30 January 2025. Second, Nexo left the United States in December 2022 and settled with the SEC and state regulators for $45m in January 2023, before re-entering the US market in April 2025 through an infrastructure partnership. Third — and most directly responsive to the failures — Ledn discontinued its bitcoin and ether growth accounts on 1 July 2025 and moved to a fully custodied model, stating that it would stop reusing client assets to create leverage. That is a business deliberately giving up a revenue line, and it is documented in our Ledn review.

What did not change is more important. There is still no deposit guarantee scheme covering crypto anywhere in this research, with the narrow exception of Xapo Bank's US dollar balances under the Gibraltar scheme — and Xapo states plainly that its crypto balances are not covered. Several large venues still run explicit lending products: OKX Simple Earn and Gate Simple Earn both lend deposits to borrowers, a structure described in how crypto lending works. And the disclosure gap persists: Nexo and Crypto.com do not explain on consumer pages how their yield is generated. A product being legal and a product being transparent are separate questions.

The final observation is about time. The Celsius case was still producing orders in April 2026, almost four years after the freeze. Gemini's arbitration in August 2026 cleared the company on one user's claim while leaving interest claims open. If you are weighing a custodial yield product, the realistic downside is not only "how much might I lose" but "for how many years might this be unresolved". That, rather than any rate, is the comparison worth making against the alternatives set out in where to earn crypto.