On 29 December 2023 the last fixed-rate series of Yield Protocol matured, and the project closed its doors two days later. Its team had announced the wind-down in October, saying there was "currently not sustainable demand for fixed-rate borrowing" on its platform, and pointing to unfavourable rules in the United States, Europe and the United Kingdom. Less than three years later, on 21 July 2026, Morpho opened Morpho Midnight on Base, a protocol built for fixed-rate DeFi lending. Two months later Coinbase added USDC loans against bitcoin whose rate and repayment date are set when the loan is opened, with Midnight as the lending protocol underneath.
Few ideas in the sector are older, and few have fared worse. Teams have tried to build fixed-rate DeFi lending since 2020; most of them are closed, wound down or tiny. This lesson explains why a fixed rate is hard to build on a blockchain, how the four main designs work (maturity buckets, sealed-bid auctions, principal tokens and Midnight’s offer-based markets), what happened to the protocols that tried, and who uses the survivors. All figures are dated 29 October 2026 unless stated.
The previous lesson, how flash loans work, dealt with a loan that lasts one transaction. This one deals with the other end of the time scale: a loan whose price is agreed for weeks or months. You need the smart contract holding reserves of two or more tokens, supplied by users, that a decentralized exchange uses to execute trades automatically at a price set by a formula.">pool model from our lesson on crypto lending and borrowing; it is recalled in one paragraph below, not repeated.
Why fixed-rate DeFi lending is hard on-chain
A quick reminder. On Aave, Compound or a Morpho Blue market, every lender of an asset sits in the same pot and every borrower draws from it. The interest rate is a function of utilisation, the share of deposits that is lent out: when borrowers take more, the rate climbs along a curve, and when they repay, it falls. The rate moves at every block in which someone deposits, borrows or repays. Nobody signs a price for a period of time.
The authors of the Midnight whitepaper sum up the link in one line: "Pool-based architectures go together with variable rates." The paper adds two consequences. A borrower who needs a predictable funding cost faces interest rate risk, and a lender has to watch utilisation constantly to keep the return they wanted. In a small market, a modest flow of deposits or loans can move the rate a long way.
Banks solved the problem centuries ago with term deposits, bonds and fixed-rate mortgages. On a blockchain, four obstacles stand in the way, and each design below is an answer to one or more of them.
A fixed rate needs someone who takes the other side
In a pool, the pool itself is the counterparty: you can deposit or borrow any amount at the rate of the moment, because the curve adjusts. A fixed-rate loan for three months at 5% needs a lender who agrees to that exact rate, for that exact date, for at least part of that amount. If nobody is there, there is no trade. Someone also has to absorb the risk that rates move after the loan is signed: either the protocol keeps that risk (and can lose money), or a market finds a price for it.
Every maturity splits the liquidity
A pool has one rate and one pot. A fixed-rate system needs a date. If a protocol offers 1-month, 3-month and 6-month loans, it has three separate markets, each thinner than the single pool would be. Multiply by collateral types and the depth spreads further. The Midnight paper names this directly: in isolated markets, liquidity can fragment between collateral configurations and maturities. Thin markets mean wide gaps between what a lender asks and what a borrower pays.
Leaving before the date costs money
A variable-rate deposit can be withdrawn whenever liquidity allows. A fixed-rate position is a claim on money at a future date. To exit early, you sell that claim to someone else at today’s market price, which may be lower than what you paid if rates have risen since. On a borrowing position, closing early means buying back the debt, at a price that may have moved against you. Without a secondary market, the position is simply locked until maturity.
Aave’s stable rate was never a fixed rate
Aave offered a "stable" borrowing rate from its early versions until 2024, and many users read it as a fixed rate. It was not. Governance could rebalance it, and in its forum post of 31 January 2024 Aave’s developer team BGD reminded users that rebalancing had existed since Aave v2 was launched. After a white-hat reported a critical bug in the stable-rate logic through Immunefi on 4 November 2023, new stable borrowing was halted. BGD then proposed to move every remaining stable position to variable, over the objections of some borrowers who had locked a rate below the variable one. The v3.2 upgrade, announced on 13 September 2024, removed the deprecated stable-rate logic entirely; by 8 October 2024 BGD reported that all Aave v3 instances ran v3.2. Our lending and borrowing lesson tells that story from the user’s side.
The lesson from Aave: a fixed rate bolted onto a variable-rate pool is fragile. The pool must still balance itself, and someone pays when the fixed rate drifts far from the floating one.
Reading a fixed rate: a price today for one unit at maturity
Every serious fixed-rate design in DeFi borrows the logic of a zero-coupon bond or a Treasury bill. You do not receive interest along the way. You buy, today, a claim worth exactly one unit of the loan asset (one USDC, one ETH) on a set date, and you pay less than one for it. The discount is the interest.
The Midnight whitepaper gives the formula in its simplest form: for a traded price P, the simple rate over the remaining term is r = 1/P – 1. Pendle’s principal tokens, Notional’s fCash and Term Finance’s repo tokens all rest on the same idea, even though their mechanics differ.
A worked example, with made-up prices chosen to show the arithmetic, not market quotes. A claim on 1 USDC payable in six months costs 0.97 USDC today. The rate over the term is 1/0.97 – 1, or 3.09%. Annualised on a simple basis (multiplied by two), that is about 6.19%. If the same claim cost 0.99, the six-month rate would be 1.01% (about 2.02% a year).
| Price today (example) | Time to maturity | Rate over the term | Simple annual rate |
|---|---|---|---|
| 0.99 | 6 months | 1.01% | about 2.02% |
| 0.97 | 6 months | 3.09% | about 6.19% |
| 0.95 | 6 months | 5.26% | about 10.53% |
| 0.97 | 3 months | 3.09% | about 12.37% |
Two things follow. First, a lender who buys at 0.97 and holds to maturity earns 3.09%, whatever happens to rates in between, provided the borrower or the protocol pays. Second, the price of that claim moves every day before maturity. If market rates rise, new claims trade cheaper, and an existing holder who sells early gets less than expected. A fixed rate is fixed only for whoever holds to the end.
Design one: maturity buckets in a pool, and why Notional and Yield Protocol ended
How Notional’s fCash worked
Notional was for years the best-known fixed-rate lending protocol on Ethereum. Its building block was fCash. According to Notional’s documentation, an fCash token is defined by a currency and a maturity date and can be redeemed for one unit of that currency at maturity, much like a zero-coupon bond. A lender held positive fCash, a borrower negative fCash.
To lend at a fixed rate, a user swapped cash for fCash in one of Notional’s own liquidity pools, one per maturity. The exchange rate fixed the interest. Notional used continuous compounding: the docs’ own example turns 100 USDC lent for six months at 5% into 102.53 fCash, claimable as 102.53 USDC at maturity. To borrow, a user deposited collateral, minted a pair of fCash tokens (one positive, one negative), sold the positive one in the pool for cash and kept the negative one as the debt due at maturity.
Early exit meant selling fCash back to the pool at the market rate of the moment. Notional’s docs show the cost: lend 1,000 USDC for a year at 3.75% with a 0.3% fee, sell the same day, and you get back about 994 USDC. After maturity, in version 3, fCash converted into Notional’s variable-rate "Prime Cash", so lenders kept earning a floating rate until they withdrew, and late borrowers paid a floating rate on what they still owed.
The design answered the counterparty problem with liquidity providers who took the other side in each pool and earned the trading fees. It did not escape fragmentation: each currency and each maturity was its own pool.
The end of Notional, in three dates
Notional V2 was deprecated in 2024 (DefiLlama marks it inactive from 22 August 2024). Notional V3 peaked at about 23.7 million dollars of deposits on 30 July 2024, per DefiLlama’s series, a small figure next to the billions in variable-rate pools.
On 3 November 2025 a smart-contract hack of Balancer V2 drove the collateral in five of Notional V3’s Balancer-based leveraged vaults to zero: rETH/WETH, staking means depositing a proof of stake coin with a protocol that stakes it for you and issues a transferable token (such as Lido stETH or rETH) representing the deposit plus rewards, which you can use or sell while the coins stay staked.">wstETH/WETH and rsETH/WETH on Arbitrum, ezETH/WETH and rsETH/WETH on Ethereum mainnet. Notional paused the protocol. In its post of 20 November 2025 the team wrote that "it would not be possible to return Notional V3 to a functioning state in production" and chose to wind it down and hand back what was left.
| Notional V3 after the Balancer hack (blog, 20 Nov 2025) | Ethereum mainnet | Arbitrum |
|---|---|---|
| Bad debt | 632.8 ETH | 80.2 ETH |
| Haircut on ETH lenders | 56.019% | 19.244% |
| ETH distributed to users | 414.272 ETH | 93.466 ETH |
| of which bought with protocol reserves | 66.571 ETH | 51.867 ETH |
ETH lenders who had picked a fixed rate for its predictability lost more than half of their deposits on mainnet because of leveraged vaults built on a third-party protocol. Cross-currency borrowers were moved to Aave, each through a personal Safe holding the position. It is a textbook case of the dependency chains described in our lesson on DeFi composability and its cascades.
The third date came on 3 September 2026. An attacker drained about 1.65 million USDC and 69,000 DAI from the escrow contract of Notional V1, deprecated since January 2022 but still open for withdrawals. Notional’s post-mortem of 8 September 2026 traces it to an integer overflow linked to the mintfCashPair function, which let the attacker create unbacked collateral and borrow against it. The team said it was working with US law enforcement and the incident response firm Zero Shadow; no recovery had been announced in that post, which also stated that no funds in any other Notional product were at risk. On 29 October 2026 DefiLlama shows Notional V3 with practically nothing left.
Yield Protocol, the first zero-coupon design
Yield Protocol came from a 2020 paper by Dan Robinson and Allan Niemerg, which the Midnight whitepaper cites as earlier work on fixed rates. Lenders bought discounted tokens redeemable for one unit at a set date, and a dedicated AMM priced them. Its deposits peaked at about 22.4 million dollars on 31 March 2022, according to DefiLlama. The team announced in October 2023 that the December 2023 series, maturing on 29 December, would be the last; the March 2024 series was cancelled and all lending and borrowing ended by 31 December 2023. About 0.25 million dollars still sits in its contracts on 29 October 2026, per DefiLlama, unclaimed by users.
Morpho’s co-founder Paul Frambot gave his reading of that first wave when Midnight launched, telling The Block that "in past attempts, fixed rates were built on top of variable rates, which was imperfect". That is his view, not a settled verdict; the figures above show at least that demand never reached the size of variable-rate lending.
Design two: sealed-bid auctions in repo style (Term Finance)
Term Finance takes its model from the repo market, where banks and funds lend cash against collateral for a set term. Its vocabulary says so: a borrower "sells" collateral tokens to lenders at a purchase price and repurchases them at a higher repurchase price on the repurchase date; the gap is the interest, set by the repo rate. The repurchase window typically lasts 12 to 24 hours, per Term’s terminology page.
The price is not set by a curve or an AMM but by an auction. According to Term’s documentation, each auction is a "sealed-bid, second-price, single-shot, single-price" double auction, also called a call market: lenders submit sealed offers to lend and borrowers submit sealed bids to borrow, and they are matched together at a set time. The protocol computes one clearing rate. Lenders who asked for less than that rate lend, borrowers who offered at least that rate borrow, and everyone trades at the clearing rate. Orders on the right side of the price may still go unfilled if the other side runs out.
The auction answers the counterparty problem in a direct way: borrowers and lenders meet at one moment, so liquidity concentrates instead of trickling in. Lenders receive Term Repo Tokens, ERC-20 tokens worth the repurchase price due to them, which convert one for one into the loan token when burned at maturity; if, after repurchases and liquidations, there is not enough cash to redeem them in full, the haircut applies equally to every lender in that repo. A lender cannot redeem early but can sell the tokens to anyone willing to buy them, and a borrower cannot repay early but can close the position by acquiring and burning the same amount of repo tokens. Borrowers must keep their collateral above a maintenance margin or face liquidation.
Term is alive but small. As of 29 October 2026 DefiLlama shows Term Finance’s lending product at 3.53 million dollars of deposits, all on Ethereum, with 2.96 million borrowed, down from a peak of 68.3 million on 17 December 2024. We found no public explanation for the decline. A newer fixed-rate AMM, TermMax (a separate product listed under its own name on DefiLlama), shows 0.70 million dollars against a peak of 40.7 million on 29 October 2025.
Design three: buying the principal at a discount (Pendle PT)
Pendle is the one fixed-rate venue with size. It does not lend money. It splits a yield-bearing token, such as staked ether or a savings-dollar token, into two parts until a maturity date: a Principal Token (PT) and a Yield Token (YT). The YT holder receives the variable yield and any points; the PT holder gets neither but buys at a discount. Pendle’s documentation says PT trades at a discount to its accounting asset and redeems one for one for it at maturity; that appreciation sets the fixed yield. Our project page on Pendle covers the protocol and its token.
The arithmetic is the zero-coupon formula above. On 29 October 2026, at about 10:00 UTC, Pendle’s public API showed these implied fixed yields for Ethereum markets: 4.73% a year on PT-sUSDe maturing on 26 November 2026, 5.63% on PT-sUSDS for the same date, and 2.2% on PT-wstETH maturing on 30 December 2027. Each of those pools held roughly 4.7 to 5.1 million dollars of liquidity. These are market prices of that moment, not promises; they change with every trade.
The fixed yield comes with conditions. First, it goes to the buyer of PT, the lender side. Pendle does not give a borrower a fixed borrowing cost. Second, a PT is only as sound as the token it is cut from: if sUSDe or sUSDS loses value, the PT redeems for less than expected. Our lesson on yield-bearing stablecoins explains where those underlying yields come from. Third, a high implied yield on a little-known token usually prices risk, not a bargain.
PT also feeds lending markets. Pendle’s docs give Morpho, Silo and Euler as examples of the money markets that accept PT as collateral, because its value at maturity is predictable. That is how many looping strategies are built, with the risks described in our lesson on borrowing against crypto. Pendle’s total value locked was about 1.06 billion dollars on 29 October 2026, per DefiLlama, against a peak of 13.39 billion on 19 September 2025: a fall of about 92%.
Design four: offers and zero-coupon units (Morpho Midnight)
Midnight is the newest attempt, and Morpho pitches it as fixed rates built into the base layer rather than added on top of a variable pool. It is a separate protocol from Morpho Blue, the variable-rate market layer covered on our Morpho project page, with its own code in the morpho-org/midnight repository (main contract under the BUSL-1.1 licence, other files MIT). The whitepaper, published by the Morpho Association in May 2026, describes isolated, immutable markets that anyone can create, each defined by a loan token, a maturity and a set of accepted collateral with their parameters.
Inside a market, positions are counted in units. One debt unit is an obligation to repay one loan token before maturity; one credit unit is a claim on those repaid tokens. Buying units adds credit (you lend), selling units adds debt (you borrow). The rate comes from the discount, with the r = 1/P – 1 formula shown earlier.
Offers that do not lock capital
Midnight has no pool and no canonical order book. Makers sign offers, a price and a maximum size for a market, and publish them through any channel, on-chain or off. A taker picks an offer and submits it to the contract, which settles the trade if the funds are there at that moment. Finding the best offers, called routing, happens outside the protocol and can be done by anyone.
Callbacks are the part that addresses the cost of waiting. An offer can carry code that runs when it is taken, so a lender can keep money in a Morpho Blue market, earning the variable rate, while quoting a fixed rate on Midnight; the funds are withdrawn only if the offer is filled. The same mechanism lets a borrower roll debt into a later maturity in one transaction. To fight fragmentation, offers can share a common budget across many markets, and one signature can approve a whole set of offers through a Merkle root. Prices sit on a tick grid: each step starts as a 2% relative change in implied return and can be narrowed to 1% or 0.5% as a market deepens.
Same date, same market: early exit
Markets mature on fixed calendar dates, not on a term counted from each loan. Every position with the same maturity therefore lives in the same market and is interchangeable with the others. A lender who wants out sells credit units to a new lender; a borrower who wants to close buys units back, or repays one loan token per debt unit at any time. The exit price is the market price of the day, so the mark-to-market logic of the previous section applies.
Liquidations before and after maturity
Each collateral has its own oracle and its own liquidation loan-to-value (LLTV), drawn from the same list as Morpho Blue, with 0% removed and 100% added. A position is healthy while the sum of collateral values weighted by their LLTVs stays above the debt; below that, a liquidator can repay part of the debt and seize collateral at a discount. The thresholds work like the health factor of a variable-rate market, and the mechanics follow our lesson on how liquidations work.
The maximum discount comes from a formula, LIF = 1 / (1 – γ × (1 – LLTV)), where the cursor γ can only take 0.3 at launch, the value used across all Morpho Blue markets. Our own calculation: for a collateral with an LLTV of 86%, that gives about 1.044, a liquidator bonus of roughly 4.4%; at 91.5% it falls to about 2.6%. The repaid amount is capped at what restores the position’s health, except for dust positions.
Maturity adds a new trigger. After the date, any unpaid debt can be liquidated even if the position is healthy, so that lenders can collect. To avoid handing too much to liquidators for a merely late borrower, the incentive starts at 1 (no bonus) at maturity and rises in a straight line to the maximum over 60 minutes, a mechanism the paper likens to a Dutch auction. If a liquidation leaves a shortfall, the bad debt is booked at once by cutting every lender’s credit in that market in proportion.
Fees and access gates
Two fees are possible, both capped in the code: a settlement fee on each trade, whose implied annual rate cannot exceed 50 basis points, and a continuous fee on lenders’ outstanding credit, capped at 1% a year. A market can also name up to two gate contracts at creation, one restricting who may enter and one restricting who may liquidate. The enter gate never blocks exits: a participant can always withdraw credit, repay debt or take back collateral. Gates are how a market can be limited to verified institutions while the base protocol stays open.
Where Midnight stands on 29 October 2026
Midnight went public on Base on 21 July 2026, the day The Block reported its launch, scoped to one network, one pair (cbBTC and USDC) and a limited set of maturities, according to Morpho’s launch post. USDC markets backed by WBTC and cbBTC followed on Ethereum in September 2026, as reported by The Defiant; DefiLlama’s Ethereum series for Midnight starts on 15 September.
The first large distributor arrived fast. On 22 September 2026 The Block reported that Coinbase had added fixed-rate loans in its app: users borrow USDC against their bitcoin, with the rate and the repayment date set at the start. Coinbase runs the app, Morpho provides the protocol and transactions settle on Base. The loans fall due at the end of the current month or of the next one (the last Friday of the month), and a Coinbase spokesperson declined to share the rates offered.
Morpho’s vaults, the largest source of deposits on Morpho Blue, cannot yet lend on Midnight. In its post on connecting vaults to Midnight, Morpho says the adapter will go live in Q4 2026. Curators will then set the rate at which a vault is willing to lend and can cap the share lent for longer periods, for example beyond 90 days; caps can be cut at once, while raising one is timelocked. Each position will be valued at its amount due at maturity, discounted at its entry rate, and depositors will be able to leave with their share of the Midnight position itself (an in-kind redemption) when the vault has no idle cash.
DefiLlama put Midnight at 120.4 million dollars of deposits on 29 October 2026 (113.1 million on Ethereum, 7.2 million on Base) and 62.7 million borrowed. The series rose from 15.9 million on 22 October to 113.9 million on 24 October, almost all on Ethereum. We found no public explanation for that jump. Even after it, Midnight holds about 1% of Morpho Blue’s 11.19 billion dollars.
Who uses fixed-rate lending, and how much money is in it
Four groups show up in the designs and the launch material. Borrowers who need a known cost, such as a business or a fund financing a position for a set period, are the group Midnight and Coinbase aim at: the cost is set when the loan is opened. Institutions used to repo and term funding come next; Morpho’s launch material and The Block cite tokenised real-world assets, structured credit and repo-style trades as target uses, and Term’s design copies repo vocabulary on purpose. DAOs and treasuries that hold stablecoins and want a set return for a set period fit the same logic, though we could not verify named examples on these protocols.
The largest group by size is fixed-yield seekers on Pendle: holders who buy PT to lock the yield of a staked or savings token, and leveraged users who post PT as collateral on money markets. That use is on the lending side only.
| Protocol | Design | Status (29 Oct 2026) | TVL, DefiLlama, 29 Oct 2026 | Peak (DefiLlama) |
|---|---|---|---|---|
| Pendle | PT/YT split traded on an AMM | live | about $1.06B | $13.39B, 19 Sep 2025 |
| Morpho Midnight | offers on fixed-maturity zero-coupon units | live on Base and Ethereum | $120.4M ($62.7M borrowed) | current level |
| Term Finance (lend) | sealed-bid auctions, repo style | live | $3.53M ($2.96M borrowed) | $68.3M, 17 Dec 2024 |
| TermMax | fixed-rate AMM | live | $0.70M | $40.7M, 29 Oct 2025 |
| Yield Protocol | zero-coupon tokens on an AMM | closed 31 Dec 2023 | $0.25M left | $22.4M, 31 Mar 2022 |
| Notional V3 | fCash maturity pools | wound down after 3 Nov 2025 | near zero | $23.7M, 30 Jul 2024 |
| Morpho Blue (variable, for scale) | isolated variable-rate markets | live | $11.19B | |
| Aave V3 (variable, for scale) | shared pool | live | $16.26B |
Put side by side, the numbers are small. Fixed-rate lending in the strict sense, where a borrower pays a set rate, held well under 200 million dollars across these protocols on 29 October 2026, against tens of billions in variable-rate pools. Deposits and borrowed amounts tell different stories and are best read separately.
Risks that come with a fixed date
A fixed rate removes one risk, the floating cost, and adds others. The list below applies to every design, with the details varying.
- The maturity cliff. A borrower must repay or roll the loan by the date. On Midnight, unpaid debt becomes liquidatable after maturity even if the position is healthy, with the bonus rising over 60 minutes.
- Refinancing risk. Rolling into a new maturity happens at the rate of that day, which may be much higher than the old one.
- Mark-to-market losses on early exit. Selling a fixed-rate claim before maturity after rates have risen returns less than its face value; Notional’s own example shows a same-day exit losing about 0.6% to fees alone.
- Dependency risk. ETH lenders on Notional V3 mainnet lost 56% through vaults built on Balancer, not through a flaw in the fixed-rate logic itself.
- Legacy contracts. Notional V1, deprecated in 2022, was drained in 2026 because its contracts were still live for withdrawals.
- Thin liquidity. A market with a few million dollars can have no buyer at a fair price on the day you need one.
- Collateral and oracle risk. A fixed rate does not protect a borrower from a price crash; see our glossary entry on liquidation.
No design removes smart-contract risk. Midnight’s repository lists 15 audit reports by Spearbit, Blackthorn, TrustSec, Stermi and a Cantina competition, dated 7 April to 24 September 2026, plus formal verification with the Certora Prover. An audit lowers the odds of a bug; it does not rule one out.
Taxes and rules for US and UK readers
The legal label of these positions is unsettled. The Midnight whitepaper’s disclaimer insists that terms such as credit, debt, maturity or zero-coupon only describe how the contracts behave and do not denote a security, a bond, a deposit or a loan in the regulatory sense. Whether a US regulator such as the SEC or the CFTC would see a tradable fixed-maturity claim as a security, a swap or neither has not been tested for these protocols in any ruling we could find. Coinbase’s Midnight product reaches users through a regulated company’s app, which is a different setting from a self-custody wallet talking to the contract.
For US taxes, the IRS treats crypto as property, and we found no IRS guidance written specifically for fixed-rate DeFi positions. Swapping USDC for a PT or a repo token, and redeeming it later, may each be argued as a disposal; the interest earned is in principle taxable. Our guide to crypto taxes in the United States sets out the general rules; a tax professional should look at any sizeable position.
In the United Kingdom, HMRC’s Cryptoassets Manual has a section on DeFi lending and staking. Its page on making a DeFi loan (updated 28 November 2025) says there is no single operating model for DeFi lending platforms, that the platform’s operating model and its terms decide whether the lender’s return is taxable income or a chargeable gain, and that where no trade is carried on, lending the tokens may be a disposal for capital gains purposes. The general UK rules on crypto gains apply to the rest.
The date to watch is Morpho’s vault adapter, promised for Q4 2026: it would let billions of dollars in Morpho vaults reach Midnight markets, and it will show whether fixed-rate DeFi lending grows beyond its current corner. The next lesson of the DeFi path moves to liquid staking tokens such as stETH and rETH.
