When a Lending Pool Runs Dry: A Plain Guide to DeFi Utilization Spikes
Last week a major stablecoin market briefly refused withdrawals, and depositors panicked. Here is what actually happened, why it is not a hack, and how to read the warning signs.
On Thursday morning, thousands of users of Marrow Finance, one of the larger on-chain lending protocols, tried to withdraw stablecoins and received an error. Social media filled with the word “insolvent.” Within six hours, the problem had resolved itself and nobody had lost money. The episode is a useful lesson in how decentralised lending works, and where it can frighten you.
How a lending pool works
Imagine a communal pot. Depositors add dollars-pegged tokens, and borrowers take them out, posting other assets such as ETH as collateral. Borrowers pay interest, depositors receive most of it, and the protocol keeps a small cut.
Crucially, the pot is not a vault holding everyone’s balance. Most of the money has been lent out. If depositors put in 100 million dollars and borrowers have taken 80 million, only 20 million remains available for withdrawals. This ratio, borrowed divided by deposited, is called utilization.
At 80% utilization, the pool is working normally. At 100%, the pool is empty: every unit of liquidity is on loan, and a depositor who wants out must wait until someone repays.
Why the rate curve exists
Protocols defend against this with an interest-rate curve. Below a target utilization, usually around 85% to 90%, borrowing rates rise slowly. Beyond that point, they spike steeply. Marrow’s curve lifts the borrow rate from about 6% at its kink to a maximum of 120% annually at full utilization.
The logic is behavioural. High rates push borrowers to repay and attract new depositors, which restores liquidity. Done right, the pool heals without anyone intervening.
That is precisely what occurred on Thursday. A large depositor, identified on-chain only by a wallet address, withdrew 140 million dollars-equivalent of USDC to move it into a new yield strategy. Utilization jumped from 87% to 99.4% in two blocks. Withdrawals above the remaining liquidity failed. Borrow rates climbed above 90%, and by mid-afternoon, borrowers had repaid about 110 million, pulling utilization back to 91%.
A frozen withdrawal is a liquidity problem. A loss is a solvency problem. They are not the same thing.

Liquidity versus solvency
This distinction is the heart of the matter. A frozen withdrawal is a liquidity problem: the money exists, but it is lent out. A loss is a solvency problem: the money is gone, perhaps because collateral lost value faster than liquidators could sell it.
Marrow’s collateral coverage during the episode stayed above 142%, meaning every dollar borrowed was backed by about a dollar and forty-two cents of assets. The protocol’s risk dashboard, which Marrow publishes in real time, showed no bad debt at any point.
“People see a failed transaction and assume the worst,” said Ilse Fernhollow, a risk analyst at the independent research collective Tallowmark. “But the system did what it was designed to do. The uncomfortable part is that it was designed to make you wait.”
How to read the warning signs
Utilization spikes are not always benign. They can signal a bank-run dynamic, particularly if the collateral backing the loans is volatile or illiquid. Depositors considering their exposure can watch a few indicators:
- Utilization above 90% for more than a day. One-off spikes tend to resolve; sustained ones suggest structural imbalance.
- Concentration of depositors. If three wallets supply half the pool, one exit can drain it.
- Collateral quality. Pools accepting thinly traded tokens as collateral carry real solvency risk, not just timing risk.
- Oracle and liquidation health. Check how quickly the protocol can liquidate risky positions during volatile markets.
There is also a practical step: keep funds in pools with deep liquidity and diversified depositors, and treat any yield above the broader market rate as payment for taking one of these risks.
Marrow’s governance forum is now debating changes, including a lower kink at 80% and a withdrawal queue that would prioritise smaller depositors during a crunch. Neither fully eliminates the problem. Any system that lends out deposits will, at some point, be unable to return them all at once, and that is true of banks as much as of smart contracts.
The difference is transparency. On Thursday, every depositor could see exactly where the money was, why they could not reach it, and when it would come back. That is a better disclosure than most traditional institutions have ever offered in a run.
Sana tracks crypto regulation across the EU and UK, drawing on years of reporting on financial compliance. She is based in Dublin.
Conversation (…)
One briefing. Every morning. Free.
Join readers who start the day with Crypto Beacon: the essential Crypto & web3 stories, curated by our editors.