Borrowing Against Bitcoin Is a Custody Decision Before It Is a Credit Decision

On September 21, Circle began letting institutional clients borrow USDC against bitcoin without selling it. The customer's bitcoin is turned into a token called cirBTC, and that token is pledged in the Morpho lending market. The launch itself is covered in our news report.

The product is new, the promise is not. Borrowing against bitcoin instead of selling it has been offered for years, collapsed spectacularly in 2022, and has been rebuilt since. In our reading, what Circle's launch makes visible is that the rebuilt market has split into models that answer one question very differently: who can move the bitcoin while the loan is open? That answer decides where a borrower can lose the collateral. The interest rate comes second.

Four answers to one question

Every bitcoin-backed loan works the same way on the surface. The borrower locks bitcoin worth more than the loan, receives dollars, and gets the bitcoin back after repaying. If the bitcoin price falls far enough that the collateral no longer covers the debt with a margin, the collateral is sold. Underneath, four different arrangements are in use.

Model	Who can move the collateral	Examples
Custodial	The lender, its custodian or its funding partner	Ledn, Strike, Nexo, Xapo Bank, bank credit lines
Wrapped bitcoin in an onchain market	The wrapper's custodian holds the bitcoin, a smart contract holds the token	Coinbase with cbBTC, Circle with cirBTC, Aave
Multisig escrow on Bitcoin	Several parties together, none alone	Firefish, Debifi, Unchained
Script-enforced vaults	A Bitcoin script, triggered by external proofs or price reports	Discreet log contracts, Babylon's vaults (proposed)

Custodial. The borrower sends the bitcoin to the lender. Since mid-2025, Ledn keeps collateral in custody with Ledn or its funding partners and no longer lends client assets out to generate yield. That removes one of the practices that sank the 2022 lenders. It does not remove the lender, or the funding partner, as a counterparty.

Wrapped bitcoin in an onchain market. The borrower's bitcoin goes to a custodian, which issues a token on another blockchain. The token sits in a smart contract that liquidates automatically. Coinbase has run this model for retail customers since January 2025, and Circle's launch applies it to institutions.

Multisig escrow. The bitcoin stays on Bitcoin, in an address that needs several signatures to spend. The mechanics of multisig are explained in our knowledge base. Debifi uses three of four keys, held by the borrower, the lender, Debifi and a fourth institution. Firefish goes further. The borrower pre-signs every possible outcome of the loan, repayment, default, liquidation and cancellation, and then discards the escrow key. After that, the bitcoin can only move along those pre-signed paths. If the other parties stop responding, the borrower can recover the collateral alone one month after the loan's maturity date.

Script-enforced vaults. The loan terms are enforced by the Bitcoin script itself, with an oracle or a cryptographic proof deciding which branch executes. This is the least mature model. Babylon posted a proposal in May 2026 to bring such vaults to Aave. A joint launch had originally been targeted for April.

What the price difference buys

Custody choices show up in the rate. Coinbase advertises loans from 5.1%, with variable rates set by the lending market. Unchained disclosed rates of 14% to 16% in June 2026, according to Silicon Valley Bank, which puts the sector-wide range at 7.5% to 16% a year. Debifi reported average rates just above 10% in 2025.

The cheapest loans are the ones where the collateral leaves Bitcoin. Onchain markets pool liquidity from many lenders and liquidate by code, which keeps their costs low. Multisig lenders match individual lenders to borrowers and handle each loan with more manual work. A borrower choosing between 5% and 12% is therefore not only choosing a rate. The borrower is choosing between trusting a wrapper's custodian and a smart contract, or trusting the parties that sign in a multisig.

The drawdown as a stress test

Bitcoin set a record of 126,080 dollars on October 6, 2025, and fell to roughly 58,300 dollars by June 7, 2026, according to crypto.news. Depending on the price feed, the decline measures about 54%. It was a severe test for a lending market that had been rebuilt after 2022.

Liquidation follows a simple rule. A loan is sold when its loan-to-value ratio, the debt divided by the collateral's value, reaches the liquidation threshold. The price decline that triggers it is therefore one minus the starting ratio divided by the threshold. The table applies this to three providers whose thresholds are published by the provider or stated by its chief executive.

Provider	Starting ratio	Liquidation at	Decline that triggers liquidation	Highest starting ratio that survives a 53.8% decline
Coinbase via Morpho, at the maximum	75.2%	86%	12.6%	39.7%
Coinbase via Morpho	50%	86%	41.9%	39.7%
Ledn	50%	80%	37.5%	36.9%
Firefish	50%	95%	47.4%	43.8%

Our calculation shows that at a starting ratio of 50%, none of these loans would have survived the full decline without the borrower adding collateral or repaying part of the debt. A Coinbase loan opened at its maximum, the 133% minimum collateral, is sold after a decline of 12.6%. To ride out the drawdown untouched, a borrower would have needed to start below roughly 40% at Coinbase and below 37% at Ledn.

Three limits apply to the calculation. It assumes the loan was opened at the peak, which is the worst case. It ignores accrued interest, which pushes the ratio up over time. Coinbase explicitly counts interest toward its threshold. And it ignores margin calls, which give custodial borrowers a chance to add collateral before the sale.

Firefish survives longest because it liquidates only at 95%. That shifts the risk rather than removing it. At 95%, the collateral covers the loan by just over 5%, which leaves the lender a thin cushion if the sale itself happens at a falling price.

Lenders passed, borrowers did not. Ledn told Silicon Valley Bank that it recorded no losses on its consumer loans during the drawdown and has never lost money on a loan. According to reports on the launch of its newer product, many borrowers of Strike's first loan product were liquidated during the same decline. Both statements can be true at once. Overcollateralization protects the lender precisely by selling the borrower's bitcoin near the bottom.

Where borrowers get burned

Each model fails in its own way. The record of the last four years shows at least six failure points.

The custodian is the counterparty. In 2022, Celsius, BlockFi and Genesis collapsed, and Silicon Valley Bank estimates that roughly 11 billion dollars in customer funds were wiped out. The court-appointed examiner in the Celsius case found that the company had used customer funds to pay other customers' withdrawals, cover operating costs and fill holes in its balance sheet. Today's custodial lenders promise segregation and no reuse of collateral. Those promises are contractual. They hold only as long as the contract, the audits and the company behind them hold.

The funding is concentrated. Galaxy Research counted 14.6 billion dollars of open loans at Tether at the end of September 2025, 60% of the centralized crypto lending market. Silicon Valley Bank names Tether among Ledn's funding partners, and Tether extended a 2.1 billion dollar credit facility to Strike. The largest single source of money in this market is a stablecoin issuer, a development we traced from the regulatory side in our analysis of stablecoins and the dollar.

The wrapper can change hands. A wrapped token is only as good as the custodian behind it. In 2024, BitGo, the custodian of the widely used wrapped token WBTC, announced it would move key control into a joint venture involving Justin Sun. It later revised the structure so that BitGo entities keep two of three keys. MakerDAO and Aave, two of the largest lending protocols, considered dropping the token as collateral and tightened its risk parameters instead. Token holders had no mechanism to stop the change. A borrower whose collateral is cbBTC or cirBTC carries the same exposure to Coinbase or Circle.

Code only knows what its price feed tells it. Onchain markets liquidate the moment an oracle, the price feed a contract relies on, reports a threshold crossing. Centralized venues depend on price feeds as well. During the crash of October 10, 2025, the exchange Binance reimbursed users 283 million dollars after its pricing system misvalued collateral and triggered erroneous liquidations. The failure can also run the other way. In November 2025, the synthetic dollar xUSD lost its peg after Stream Finance disclosed a 93 million dollar loss, and lending markets that priced it at a fixed one dollar could not liquidate positions in time. The losses fell on the lenders in those markets.

Multisig still needs someone to decide. In a multisig escrow, no single party can take the bitcoin. Someone still has to confirm that a loan was repaid or that the price crossed the threshold. At Firefish, both of those oracles are currently operated by Firefish itself, which the company lists among the drawbacks of its own design. The borrower's protection is that the bitcoin can only go to predefined addresses, not that no one has to be trusted.

The model can change under the borrower. In November 2025, the lending app Lava moved from discreet log contracts to a custodial setup. Users criticized that the update did not make clear it was a fundamental change to how their collateral was held.

Where bitcoin lending is heading

In the author's reading, the market is developing along four lines at once, and they pull in different directions.

Credit markets are pricing bitcoin loans like other loans. In February 2026, Ledn closed a 188 million dollar securitization of bitcoin-backed loans. S&P Global rated its senior tranche BBB, the first investment-grade rating for such a portfolio. The deal used an established custodian and a backup servicer that can take over margin calls and liquidations if the lender fails, a role that did not exist in this market before. Silicon Valley Bank expects that more institutional capital will push rates toward those of loans against securities.

Liquidation is becoming a product feature, not a law of nature. Since July 2026, Strike offers loans that the bitcoin price cannot liquidate as long as payments are made on time. The borrower pays for it with a maximum starting ratio of 45%, a term of six months, a rate premium of 2.95 percentage points and no access to the collateral during the term. Strike says the premium funds hedging. Who carries that hedge in the next drawdown is not public.

Custody is moving back onto Bitcoin, slowly. Babylon's proposal for Aave would lock bitcoin in a Taproot output on Bitcoin and represent it on Ethereum. Even that design settles liquidations in WBTC, according to the proposal as reported, so the wrapper returns at the moment it matters most. Silicon Valley Bank sees the Lightning Network as a way to post collateral and answer margin calls within seconds, but notes that lenders have yet to build for it.

Banks are taking the top of the market. JPMorgan, Wells Fargo, Citi, Charles Schwab and Morgan Stanley already lend against bitcoin products, mostly exchange-traded funds, according to Silicon Valley Bank. Circle's launch belongs to the same movement. Both bring borrowing against bitcoin into institutions that know how to lend, and both do it by keeping the bitcoin with a custodian.

The strongest case against this reading

The best counterargument is that custody is the wrong focus. For most borrowers the main risk is leverage, not the key. A borrower who opens a Coinbase loan at the maximum is liquidated after a decline of 12.6%, and a custodian with the same threshold would sell at exactly the same point. A borrower at 30% survives almost any drawdown in any model. On this reading, a regulated custodian with audits and a public track record is safer for most people than a multisig setup they do not fully understand, and the losses of 2022 came from fraud and hidden reuse of collateral, not from custody as such.

That argument is strong, and the calculation above supports half of it. The starting ratio decides whether a loan survives a drawdown. What custody decides is whether the collateral is still there when the loan ends, and that is a separate risk that no starting ratio can offset.

What would prove this reading wrong

This analysis would be weakened if, in the next severe drawdown, custodial lenders again come through without losses while multisig and script-based models fail on oracle errors, setup mistakes or lost recovery paths. In that case the risk of the non-custodial models would have proven larger in practice than the custodial risk they are designed to avoid.

It would also be weakened if wrapped bitcoin tokens keep their redemption at one to one through a custodian failure or a forced change of custody. So far, no major wrapper has been tested that way.