The strangest feature of on-chain credit is that you must already have the money to borrow money. That is not a design failure or a temporary limitation, it is the only arrangement that works when the lender cannot identify the borrower, and everything else about these systems follows from it.
A conventional loan works because the lender can find the borrower. Identity, a legal system, a credit history and the ability to pursue somebody who does not repay are what allow money to be lent against a promise. Remove all four and the promise is worth nothing, because nothing happens to somebody who breaks it.
On-chain lending removes all four by design. The borrower is an address, the lender is a contract, and there is no mechanism to pursue anybody. The only thing the system can rely on is what it already holds, which is why the borrower must deposit assets worth more than the loan before receiving anything.
That constraint is what produces the apparent absurdity of needing money to borrow money, and once the reasoning is visible the absurdity disappears. The system is not lending against your creditworthiness, because it cannot assess it. It is lending against an asset it already controls, which is a pawnbroker's arrangement expressed in code.
Nobody is assessing whether you will repay. The system holds something worth more than the loan and does not care, which is the only design that works without identity.
Every position has a collateral ratio: the value of what you deposited against the value of what you borrowed. Protocols set a maximum borrowing capacity below the full value of the collateral, and the gap between those two is the buffer that protects the system from the collateral falling before it can be sold.
The size of that buffer varies by asset and the variation is informative. A volatile or thinly traded asset gets a smaller borrowing capacity, because a larger fall is plausible before liquidation completes and a thin market makes the sale itself uncertain. Reading the borrowing capacities a protocol assigns is reading its own assessment of which assets are risky, published as a number.
For the borrower, the ratio is the whole of the position's risk. It moves with the price of the collateral, with the price of the borrowed asset, and with accrued interest, all three continuously and none under your control. A position opened comfortably above the threshold approaches it over time even if nothing moves, because interest accrues.
When a position's ratio falls below the required level, anybody can repay part of the loan and take a corresponding amount of the collateral plus a bonus. That bonus is the incentive that makes liquidation somebody's business, and it is paid by the borrower on top of losing the collateral.
The design is deliberate and it works. Rather than the protocol needing to sell collateral itself, which would require it to trade and to have somewhere to trade, it offers a profit to anybody who does. Competition among liquidators means positions are closed within seconds of becoming eligible, which is what keeps the system solvent.
The consequence for a borrower is that liquidation is not a warning followed by a grace period. It is immediate, automated, executed by a stranger who profits from it, and it happens at the moment the price crosses a threshold rather than at a moment convenient to anybody. There is no call, no negotiation and no discretion.
Rates on these systems are not set by anybody. They are computed from utilisation, meaning the fraction of deposited funds currently borrowed, along a curve defined in the contract. Low utilisation produces a low rate to encourage borrowing; high utilisation produces a sharply higher one to encourage repayment and attract deposits.
The curve is usually gentle up to a target utilisation and then very steep beyond it, which is a deliberate design. The steep section exists to make it expensive to push utilisation toward the point where lenders cannot withdraw, and it does so by making borrowing prohibitive rather than by refusing it.
For a depositor this means the advertised rate is a snapshot of current utilisation rather than a promise, and it can change substantially within hours. For a borrower it means the cost of a position is not fixed, and a rate that was affordable when the position was opened can become the reason it is closed.
Deposits are lent out, so the funds available for withdrawal are only the portion not currently borrowed. When utilisation approaches its maximum, depositors who want to withdraw find that they cannot, not because anything has failed but because the money is out on loan.
The steep interest curve is the mechanism intended to resolve this: as utilisation rises, borrowing becomes expensive enough that some borrowers repay and some new depositors arrive, and liquidity returns. It usually works within hours, and usually is doing real work in that sentence.
What it means practically is that a deposit in one of these systems is not equivalent to holding the asset. It is a claim that is normally redeemable on demand and occasionally is not, at moments that correlate with everybody wanting the same thing. Anybody sizing a position on the assumption of instant withdrawal has assumed away the one condition under which it matters.
Your deposit is lent out. It is available on demand except when everybody wants it, which is the same limitation every fractional system has ever had.
The question people ask is why a person holding an asset would deposit it to borrow less than it is worth, and there are three coherent answers. The first is liquidity without sale: somebody who wants cash but does not want to dispose of a holding can borrow against it, which in many jurisdictions is also a different event for tax purposes than selling, though that varies entirely and is a question for somebody qualified where you live.
The second is leverage. Borrowing against an asset to buy more of the same asset produces a larger position than the capital alone would support, which is the same mechanism as margin trading arranged differently. The third is to obtain a specific asset temporarily, for a trade or an operation, without acquiring it outright.
All three are legitimate and the second is the one that dominates by volume. That matters because it means a substantial share of the borrowing in these systems is directional exposure rather than financing, and it responds to price the way leveraged positions do rather than the way loans do.
The leverage use has an extreme form that is worth naming because it is common and its failure mode is specific. Deposit an asset, borrow against it, buy more of the same asset, deposit that, borrow again, and repeat. Each cycle adds exposure, and the arrangement can be repeated until the marginal borrowing capacity approaches nothing.
The resulting position has a leverage multiple determined by the borrowing capacity and the number of cycles, and it is exposed to the same asset at every level simultaneously. A fall in the price reduces the collateral value and the borrowed asset's value in the same direction, which is the opposite of a hedge.
The failure is fast and it is collective. Because many participants build the same structure on the same assets, a decline triggers liquidations that sell collateral, which pushes the price lower, which triggers the next tier of liquidations. The mechanism is identical to a cascade on a leveraged exchange and it happens on a public ledger where the accumulated positions are visible in advance to anybody who looks.
Liquidation depends on the collateral being sellable for more than the debt, and in a sufficiently fast fall it is not. When a position is liquidated for less than it owes, the shortfall is bad debt: a loan that will not be repaid, held by a system that has already given out the money.
Protocols handle this in different ways and the difference is worth knowing before depositing. Some maintain an insurance reserve funded from fees, which absorbs shortfalls up to its size. Some socialise losses across depositors, meaning your deposit is reduced. Some do both in sequence, and some have simply had a hole in their balance sheet that governance then had to decide what to do about.
That last case is not hypothetical and it is the one that determines what a deposit really is. A depositor is not a lender to a diversified book with a capital cushion; they are exposed to the specific collateral types the protocol accepts, in proportion to how badly those behave in a fall.
Five things, all public and none requiring technical skill. What assets are accepted as collateral, because that is the list of things you are exposed to as a depositor. What borrowing capacities each is given, which is the protocol's own assessment of their risk. How the oracle that prices the collateral works, since liquidations run on that price rather than on any market you can see.
Then the two that people skip. What the protocol does when a liquidation leaves a shortfall, and whether the reserve for it is large relative to the loans outstanding. And what governance can change without notice, since borrowing capacities and accepted collateral are usually parameters rather than constants, and a change to either alters the risk of a position you already hold.
As infrastructure it does something genuinely new: it lends without anybody assessing anybody, at any hour, to any address, with terms that are identical for everybody and visible in advance. That is a real capability and it did not previously exist in any form.
What it does not do is extend credit, in the sense that word carries everywhere else. Credit means somebody assesses a borrower's ability to repay and lends against it, which is what allows a person without assets to buy a house or a business without capital to start. Overcollateralised lending explicitly cannot do that, and the attempts to add identity-based credit on top have so far been small.
The honest framing is that this is a very efficient pawnbroker with no opening hours and no discretion, which is useful and is not the thing the vocabulary suggests. Anybody expecting it to democratise access to credit has misread which problem the design solves, and the design solves its actual problem well.
Because the lender cannot identify you, cannot assess your creditworthiness, and cannot pursue you if you do not repay. With no recourse, the only thing the system can rely on is what it already holds, which has to be worth more than the loan for it to be safe.
When your collateral ratio falls below the required level, anybody can repay part of your loan and take a corresponding amount of collateral plus a bonus. Competition among liquidators closes positions within seconds. There is no warning, no grace period, no call and no discretion.
Because accrued interest is added to the debt, so a position opened comfortably above the threshold approaches it over time even if no price moves. The ratio depends on collateral value, borrowed value and accrued interest, and only the first two get watched.
Nobody. They are computed from utilisation, the fraction of deposits currently borrowed, along a curve written into the contract. The curve is gentle up to a target and very steep beyond it, which makes it expensive to push utilisation toward the point where lenders cannot withdraw.
Because deposits are lent out and only the unborrowed portion is available. When utilisation approaches its maximum, withdrawal is unavailable until borrowers repay or new deposits arrive, which the steep rate curve is designed to cause. It usually resolves within hours.
Depositing an asset, borrowing against it, buying more of the same asset, depositing that, and repeating. It multiplies exposure to one asset at every level simultaneously, so a fall reduces collateral value and borrowed asset value together, and liquidations cascade because many participants build the same structure.
The position is liquidated for less than it owes and the shortfall becomes bad debt. Protocols handle it differently: an insurance reserve funded from fees, losses socialised across depositors, or a hole that governance has to decide about. Which one applies is worth knowing before depositing.
No. Credit means somebody assesses a borrower's ability to repay and lends against it, which is what lets a person without assets buy a house. Overcollateralised lending explicitly cannot do that. It is a very efficient pawnbroker with no opening hours, which is useful and is a different thing.