Crypto lending, from the lender's side of the table

Every yield in crypto that is not a protocol reward or a government bond is somebody's borrowing cost. That is the whole of crypto lending, and the reason it is worth understanding from the lender's side is that the rate you are offered is a residual: what is left of the borrower's payment after utilisation, reserves and the platform's cut. Judge the rate without knowing the borrower and you are guessing.

Who is borrowing, and why, turns out to be a short list. Traders wanting leverage, who post volatile collateral and borrow stablecoins. Holders who want liquidity without selling, which is the entire bitcoin-backed loan business. Market makers and funds financing inventory. And, at the institutional end, credit desks borrowing at a spread. There is no fifth category doing something clever, which is why a rate far above what those borrowers can pay is a warning rather than an opportunity.

Two structures, one economic model

Lending yield reaches retail through two entirely different plumbing systems. The distinction is not technological snobbery; it changes what you own, who can take it, and what you do if it goes wrong.

The lending category is large enough that neither side is niche. DefiLlama tracked $50.18 billion across 582 on-chain lending protocols on 16 September 2026, led by Aave at $17.57 billion and Morpho at $10.06 billion, with SparkLend at $5.02 billion, JustLend at $3.65 billion, Maple at $2.94 billion and Compound at $1.51 billion. On the centralised side the businesses are private and the numbers mostly are not published at all — which is itself the single biggest difference between the two.

Where a DeFi supply rate comes from

Pool-based lending markets do not negotiate. They publish a curve. Borrowers pay a variable rate that is a piecewise-linear function of utilisation — the proportion of the pool currently borrowed — with a kink at an optimal point, typically somewhere between 80% and 92% for stablecoins and lower for volatile assets. Past the kink the slope steepens sharply, which makes borrowing expensive enough to force repayment and protects the ability of suppliers to withdraw.

Put real numbers through it and the market becomes readable. Here is Aave v3 on Ethereum, the single largest venue, on 16 September 2026.

The same market, two different numbers

Rates on these venues move intraday, and trackers disagree. Compound V3's Ethereum USDC market illustrates it well: one dashboard showed a 5.76% supply rate against 6.84% borrow at 90.79% utilisation on $362.6 million supplied, while a second tracker showed 4.61% supply on $374.1 million at the same timestamp, with a seven-day range of 3.29% to 5.85% and a note that the displayed figure excludes COMP incentives. Neither is wrong. A utilisation-driven rate is a snapshot of a curve, and quoting a point estimate for a rate that moved 250 basis points inside a week is the mistake, not the discrepancy.

The practical rule: for pooled lending, read the range and the utilisation together. A high supply rate at 95% utilisation is a market close to its kink, which means it pays well and may be slow to withdraw from. A high supply rate on a thin pool is usually an incentive programme with a finite budget. We keep the cross-venue snapshot updated onthe rates page.

Curated vaults, and the person you have quietly hired

Morpho is the second-largest lending protocol and works differently enough to matter. Its base layer, Morpho Blue, is a set of isolated markets — one collateral asset, one loan asset, one oracle, one liquidation threshold each. Retail deposits generally do not go there directly. They go into curated vaults run by firms such as Steakhouse and Gauntlet, which allocate across those markets, set caps, and charge a performance fee. Your yield is a weighted blend of the underlying markets minus that fee.

That introduces a risk with no equivalent in the Aave model: you are trusting a curator's risk judgement, not only the code. Two examples from 16 September 2026 make the point. A Steakhouse USDC vault on Ethereum paid a 4.36% net APY on $68.17 million with a 5% performance fee. A Steakhouse USDC vault on Base — same curator, same asset, same name — paid 3.18% on $131.83 million with a 25% performance fee. Same brand, 118 basis points apart, and a five-fold difference in fee. The Base vault was also 86% concentrated in a single USDC/cbBTC market despite the diversified framing.

Maple sits at the other end of the spectrum, with $2.94 billion in TVL and $1.769 billion of active loans. It is institutional credit rather than pooled overcollateralised lending, which means the yield is a credit spread — you are being paid to take the risk that a named institutional borrower does not repay. That is a legitimate business and a completely different instrument from an Aave deposit, and it is worth noting that Maple's early backers included Alameda Research. Protocol-level averages give a sense of the range on offer: Morpho's 245 tracked pools averaged 4.28%, Maple's three averaged 4.89%, Spark's 22 averaged 2.01% and Kamino's 151 averaged 1.93%.

Borrower demand is a cycle, and it sets your rate

There is one more thing a supply rate is telling you, and it is easy to miss while staring at the number. Borrowing demand in crypto is overwhelmingly demand for leverage, and leverage is pro-cyclical. When prices rise, traders borrow dollars to buy more; utilisation climbs, the curve steepens, and supply rates rise with it. When prices fall, positions close, utilisation drains, and the rate you are being paid falls at precisely the moment you most want compensating. A lending yield is therefore not a defensive asset. It is a long position on the appetite of other people to be long.

That is why the risk-free comparison is unforgiving. A three-month US Treasury bill paid 3.97% on 15 September 2026. Aave's USDC market paid 3.57%, Aave's USDT 3.11%, and the tokenised treasury sector paid an aggregate 3.74% seven-day yield. Mainstream crypto lending yield is currently at or below the rate available from the US government, and it carries contract, oracle, liquidation and — in the centralised case — insolvency risk on top. Anything much above 6% is being paid by token emissions, leverage, credit risk or basis risk, and it is worth making the provider tell you which.

The venues themselves keep changing shape around that reality. Aave v4 went live on Ethereum mainnet on 30 March 2026 with a hub-and-spoke design, in which a central liquidity hub holds assets and individual spokes attach their own collateral types, risk parameters and liquidation rules; it was approved on the tightest governance vote in Aave's history, which is itself a risk worth noting. Spark describes itself as an on-chain capital allocator borrowing from Sky's stablecoin reserves and deploying across DeFi, CeFi and real-world assets — and reported negative protocol revenue of $75,149 over a thirty-day window despite $27.69 million of gross quarterly revenue. Euler, meanwhile, now holds 74.5% of its TVL on Monad rather than Ethereum. None of that changes the formula. It changes where the formula is running and who wrote the parameters.

Who admits to lending your coins

The most useful sentence in crypto earn documentation is the one naming the borrower, and it is rare. OKX prints it: Simple Earn crypto "will be pooled and loaned to borrowers on our platform (which may include borrowers under our loan programmes or margin traders)", with returns distributed hourly based on lending supply and OKX keeping 15% of accrued returns, part of which is routed to internal risk funds. It does not publish how the APR is determined, but the model itself is unambiguous. Gate says the same thing without saying it: its Simple Earn page carries a total-lending figure in USDT, which is only a meaningful metric for a lending book. We take the OKX disclosure apart in the OKX Earn review.

Ledn goes furthest. It states that interest on USDC and USDT Growth Accounts is entirely generated by lending to its overcollateralised bitcoin-backed retail loan book, that no DeFi protocols are used, and that the book has no history of loan losses since inception — a claim it dates to June 2026 and pairs with its own reminder that past performance does not guarantee future results. It pays 6.5% below $100,000 and 8.5% above, accrues daily, pays monthly in kind, and ring-fences accounts through two Cayman special-purpose vehicles. That structure has not been tested in an insolvency and offers EU or UK retail no local recourse, both of which we set out in the Ledn review.

Against those, the silence elsewhere is conspicuous. Nexo's own September 2026 explainer tells readers to check custody arrangements, regulatory authorisation and reserve transparency — and does not explain how Nexo generates yield. Crypto.com does not state it. Wirex says only that deposits are converted to DAI or another stablecoin and deployed into a DeFi protocol, without naming the protocol. You cannot price a credit exposure you cannot see; you can only decide whether you trust the brand, which is a different activity. Our guide tochoosing a venue turns that into a checklist you can run on any product page in about ten minutes.

Overcollateralisation is a mechanism, not a guarantee

The standard defence of crypto lending is that loans are overcollateralised, so default is impossible. That is true in the same sense that a house cannot burn down if the fire brigade arrives instantly. Overcollateralisation converts credit risk into execution risk: the position is safe as long as the liquidation engine can sell the collateral quickly enough, at a price near the mark, in exactly the conditions that made the borrower insolvent.

The 10 and 11 October 2025 cascade is the reference case. Roughly $19 billion was liquidated across about 1.6 million accounts — nine times the previous single-day record. Around 70% of it, about $6.93 billion, landed in the forty minutes between 20:50 and 21:30 UTC, with $3.21 billion inside one sixty-second window at 21:15. Bitcoin fell 14.5%, from $122,574 to $104,782, Ethereum fell 12.2%, and Solana briefly dropped more than 40%. Approximately $350 billion of market capitalisation went with it. Nothing about that event was a failure of the collateral maths. It was the collateral maths working at full speed, and price marks behaving strangely while it did — Ethena's USDe printed $0.65 on Binance alone during the same window while holding its peg everywhere else.

Smart-contract failure is the third layer, and it is well documented. Euler lost $197 million in March 2023, of which $240 million was subsequently returned by the attacker; the Curve Vyper compiler reentrancy bug cost $61.7 million in July 2023. More recently, Tectonic lost $124.47 million to a donation attack on 30 August 2026 and Liquid Network lost $320 million to an unbacked cross-chain mint on 6 September 2026. Aggregate losses are falling even as incident counts rise — Immunefi counted $972 million across 207 incidents in the first half of 2026, while CertiK put the same period at $1.32 billion using a different methodology — but the two largest 2026 events both landed in the weeks before we checked. Ourrisk guide keeps the running tally.

What to ask before you supply anything

The diligence for lending is shorter than for most financial products, because there are only a few things that can go wrong. Who is the borrower, and what have they posted? What is the utilisation, and how fast can I withdraw if it rises? What does the platform keep, and is that disclosed as a number or as a phrase? If this is a vault, who curates it, what is the performance fee, and how concentrated is the largest market? And if this is a company rather than a contract, what is the legal entity, and what does the last comparable failure tell me about what a recovery looks like?

That last question has a documented answer. BlockFi's creditors achieved a full recovery of their US dollar claims valued at the July 2022 petition date, which meant being repaid at 2022 prices and missing everything that followed. Gemini Earn users were the exception: they received their actual coins back, in kind, in May 2024. The difference between those two outcomes is worth more than several percentage points of yield, and it is the subject ofwhat happened to the CeFi lenders.

If you would rather shop products than mechanisms, crypto savings accounts covers the retail wrappers. If you want the dollar-denominated version of this analysis, stablecoin yield takes the four funding engines apart. And if you want to see what happens when the same capital is used to provide liquidity rather than credit, DeFi yield farming is the comparison.