On-chain lending carried roughly $58 billion through a 52% decline in its collateral without disorderly liquidation. In 2022 a comparable decline produced a chain of insolvencies across the lending sector. The difference is a design choice, and it is worth being precise about which one.
Digital asset lending failed comprehensively in 2022. Celsius, Voyager, BlockFi and Genesis all entered bankruptcy, and Three Arrows Capital's collapse propagated through the lenders that had financed it. The mechanism was consistent: undercollateralised or reputationally collateralised loans, opaque rehypothecation, and no enforceable liquidation until it was far too late.
2026 delivered a decline of comparable severity to the collateral backing on-chain credit. The outcome was materially different.
The measurements
Lending protocols held approximately $58 billion in total value locked, with Aave v3 on Ethereum accounting for around 79% of it, or roughly $46 billion. Utilisation ran at 35–36% against approximately $21 billion borrowed. Seven-day liquidation volumes remained minimal to near-zero despite significant price volatility, and analysts characterised the position as showing no stress indicators.1
Two of those figures deserve emphasis. Utilisation in the mid-thirties means roughly two thirds of deposited capital was not lent out — there was substantial unused capacity, and no scramble for liquidity. And near-zero liquidations through a period when collateral values were falling sharply means positions were, in aggregate, collateralised well above their liquidation thresholds.
Why the structure behaved differently
The design difference is overcollateralisation enforced by code rather than by credit judgment.
An on-chain lending position requires collateral worth more than the loan — typically substantially more — deposited before funds are released. If the collateral ratio falls below a threshold, anyone can trigger liquidation and receive a fee for doing so. There is no negotiation, no forbearance and no relationship consideration. The position closes.
This is a worse deal for borrowers than the 2022 alternative, which frequently extended credit against reputation or against assets the lender could not seize. It is a considerably better arrangement for the system, because it removes the discretion that allows an underwater position to persist and grow.
Overcollateralised lending is capital-inefficient by design. 2026 is the year that inefficiency was demonstrably the point.
The second difference is transparency. Positions, collateral ratios and liquidation thresholds are publicly observable in real time. In 2022 the extent of Three Arrows Capital's borrowing across multiple lenders was not visible to any individual lender until the positions failed. On-chain, aggregate exposure is a matter of public record, which allows participants to observe accumulating risk rather than discovering it.
The liquidation mechanism under stress
Automated liquidation has a known failure mode: if collateral cannot be sold fast enough, or if the liquidation itself moves the price, cascading liquidations can generate the very insolvency the mechanism exists to prevent. This has occurred in earlier episodes when on-chain liquidity was thinner.
It did not occur in 2026, and the reason connects to broader market structure. Order book depth within 200 basis points measured $614.1 million for bitcoin and $475.5 million for ether, with sub-basis-point spreads maintained across venues.1 Liquidations required to clear had somewhere deep to clear into. The credit system's resilience was partly a function of the trading market's depth, which is a dependency worth noting rather than assuming.
Centralised derivatives venues did experience substantial forced liquidation during the same period, including $1.86 billion in a single day on 3 June and a $3.2 billion single-day cascade over the cycle.2 Those are leveraged directional positions rather than overcollateralised loans, and the distinction matters: the derivatives liquidations were the mechanism transmitting the decline, while the lending protocols largely absorbed it.
What has not been tested
Three qualifications are important.
Concentration is high. Aave v3 on Ethereum represented approximately 79% of lending TVL.1 A protocol-level failure — a smart contract exploit, an oracle manipulation, a governance failure — in the dominant venue would be a different event from the market stress observed this year, and smart contract risk is not diversified by the fact that collateral is sound.
Oracle dependency remains. Liquidation triggers depend on price feeds. A feed that fails, lags or is manipulated during exactly the conditions when liquidations are needed is the single most plausible route to a systemic failure in this design, and it has not been stressed at the current scale.
And 2026 was an orderly decline. It was large, but it unfolded over eight months with functioning markets throughout. A faster dislocation — a 40% move in days rather than months, with venues degraded — would test the liquidation machinery in a way this year did not.
The broader observation
The strongest available statement is narrow but real: a transparent, overcollateralised credit system carrying tens of billions of dollars processed a halving of its collateral base without producing an insolvency, at a scale where the previous generation of opaque, undercollateralised lending produced a sector-wide failure.
That comparison is the most direct evidence available that the structural changes made after 2022 addressed the actual failure mode rather than its symptoms. It does not establish that on-chain credit is safe, and the untested risks above are genuine. It does establish that the specific mechanism that broke the last cycle has been engineered around, which is a more concrete claim than most made about progress in this sector.