How DeFi Interest Rates Work in 2026: Utilization and the Kink

On September 1, 2026, Morpho’s outstanding loans crossed 5 billion dollars for the first time, and roughly 95 percent of that credit sat in stablecoins, with USDC alone accounting for about 62 percent, according to on-chain figures reported by Cryptopolitan. Aave, still the largest venue, carried about 12.7 billion dollars in active loans against 17.7 billion in deposits the same week. Those headline totals get the attention, but the number that actually runs the machine is smaller and far less discussed: the interest rate, and almost nobody sets it by hand.

In a bank, a committee decides what savers earn and what borrowers pay. On-chain, a smart contract does it, block by block, from a single live input. Move that input and every borrower’s bill and every lender’s yield shift within seconds, with no memo and no notice. The mechanism is not a mystery, but it is genuinely different from anything in traditional finance, and reading it correctly is the difference between understanding a DeFi lending market and gambling on one.

This guide explains how on-chain lending rates are set in 2026: what utilization is, why the kink in the curve exists, where a lender’s yield actually comes from, and how the newest models (Morpho’s adaptive curve, Fraxlend’s time-weighted rate) discard the old fixed curve entirely. It is the companion to our explainer on how DeFi liquidations work, because interest rates and liquidations are the two engines that keep an overcollateralized pool solvent without a bank standing behind it.

Rates without a central bank

On-chain lending uses a peer-to-pool design. Suppliers deposit an asset (say USDC) into a shared smart-contract pool; borrowers draw from that pool against collateral they have locked; and an interest rate model, usually called the IRM, computes the borrow rate every time anyone supplies, borrows, repays or withdraws. There is no loan officer weighing an application, no credit committee, and no central bank publishing a target rate. The cost of credit is a pure, public function of one thing: how much of the pool is currently borrowed.

That has two consequences worth sitting with. First, the rate is transparent and identical for everyone in the same market at the same moment; there is no relationship pricing and no special deal for the big client. Second, there is no institution absorbing losses. A DeFi lending pool is not a bank: there is no deposit insurance, no lender of last resort, and if a loan goes bad the shortfall lands on suppliers. The rate you are quoted is compensation for taking that risk directly, and the model that sets it is trying to keep the pool both busy and solvent at the same time.

Utilization, the one number that matters

Utilization is the ratio of borrowed funds to supplied funds in a given market. If a USDC pool holds 100 million dollars and 70 million is borrowed, utilization is 70 percent. It moves between 0 and 100 percent, and it is the single variable almost every interest rate model reads before anything else.

Utilization is really a liquidity gauge in disguise. At 0 percent, every deposited dollar sits idle and earns nothing. At 100 percent, every deposited dollar has been lent out, which means no supplier can withdraw until a borrower repays or someone new deposits. A healthy pool needs to keep a cushion of idle liquidity so that lenders can exit on demand, yet idle liquidity is wasted capital that drags down yield. The protocol therefore wants utilization high enough to be efficient but low enough to stay liquid, and it uses one tool to steer toward that balance: price. When the pool gets too full, the model raises the borrow rate; when it empties out, the model lowers it. Everything else in this article is a variation on that single idea.

The kink, and the curve everyone copied

The dominant model in 2026 is still the one Compound introduced in 2019 and Aave popularized: a kinked, or jump-rate, curve. It is a piecewise linear function with a single bend, called the kink or the optimal utilization point. Below the kink, the borrow rate rises gently along a shallow slope (slope1). Above the kink, it rises steeply along a much harsher slope (slope2, often called the jump multiplier).

The math is simple. Below optimal utilization, the borrow rate is the base rate plus slope1 scaled by how far utilization has traveled toward the kink. Above it, the rate is the base plus all of slope1 plus slope2 scaled by how far utilization has pushed past the kink toward 100 percent. As RareSkills lays it out, that is base + slope1 times (U divided by U_optimal) below the kink, and base + slope1 + slope2 times (U minus U_optimal) divided by (1 minus U_optimal) above it.

Numbers make it concrete. Take a high-liquidity stablecoin market with a base rate of 0, slope1 of 4 percent, slope2 of 75 percent and an 80 percent kink, a configuration documented by risk analysts at LlamaRisk. At 40 percent utilization the borrow rate is 4 percent times (40 divided by 80), or 2 percent. At the 80 percent kink it reaches 4 percent. Push to 90 percent utilization and it jumps to 4 percent plus 75 percent times (10 divided by 20), or 41.5 percent. At 100 percent utilization it hits 79 percent. The last stretch is punishing by design: cheap borrowing keeps capital productive below the kink, and a wall of rate above it forces repayment before the pool runs dry.

The shape is not arbitrary. Compound’s original designers wanted a rule that needed no human hand: a curve that made capital cheap when a pool was underused and prohibitively expensive as it approached exhaustion, so that the market, not a committee, rationed the last available dollars. A borrower who opens a position at 70 percent utilization and a comfortable 3 percent rate can watch that rate triple in an afternoon if a few large borrowers crowd in behind them, because every new loan pushes utilization toward the kink and drags the whole market up the curve. The rate you sign up for is never guaranteed to be the rate you keep.

Those exact values are not universal constants; they are configured per market and moved by governance, which is why the same asset can carry different curves on different chains. Notice the pattern down the table: the more volatile the collateral, the lower the kink and the steeper the cliff, because a volatile market must protect its liquidity earlier. What stays constant is the shape, a gentle ramp to keep borrowing attractive, then a cliff to defend the pool.

Borrow rate versus supply rate: where your yield comes from

A common beginner mistake is to assume the rate borrowers pay is the rate lenders earn. It is not. Suppliers earn the borrow interest, but only on the fraction of the pool that is actually lent out, and only after the protocol takes its cut. The standard relationship, again per RareSkills, is supply rate equals borrow rate times utilization times (1 minus the reserve factor).

Work an example. If the borrow rate is 10 percent, utilization is 50 percent and the reserve factor is 20 percent, the supply yield is 10 percent times 0.5 times 0.8, or 4 percent. Half the pool is idle and earns the depositor nothing, and a fifth of the interest that is earned is skimmed by the protocol. This is why a quiet market with a respectable headline borrow rate can still pay lenders very little, and why supply APYs leap when a pool fills up: two levers move a lender’s yield, the borrow rate (set by the curve) and utilization (how full the pool is), and near the kink both climb together. When a dashboard shows a supply APY that seems too good, back-solve it with this formula before trusting it.

The table below shows how the same 80 percent kink curve turns utilization into a lender’s yield, assuming a 10 percent reserve factor. Notice how little suppliers earn in a quiet pool and how violently that yield climbs once borrowing pushes past the kink.

The reserve factor, and who gets paid

The reserve factor is the slice of borrower interest the protocol keeps rather than passing through to suppliers. It funds a treasury, a safety buffer, or straight protocol revenue. Aave routes its share toward AAVE buybacks under its Aavenomics framework; Morpho Blue takes the opposite stance at its core, charging a 0 percent base fee and leaving the economics to curators and an optional, governance-controlled fee switch. The market treats both engines as valuable in their own right: MORPHO trades near 2.66 dollars for a valuation under 2 billion dollars per CoinGecko, while AAVE sits around 129 dollars per CoinGecko, the two governance tokens of the sector’s leading credit venues.

The reserve factor is also a policy dial. Raise it and the protocol earns more while lenders earn less, which quietly nudges suppliers toward competing markets; lower it and you subsidize deposits at the treasury’s expense. It sits alongside slope1, slope2 and the kink as one of the levers a DAO can pull, and pulling any of them changes both the user experience and the protocol’s income at once. When Aave’s stewards raised stablecoin slopes in August 2026, the stated side effect was roughly 2.55 million dollars a year in additional DAO revenue, a reminder that rate policy and revenue policy are the same conversation.

Reading a live market: the self-correcting loop

The kinked curve is not just a pricing table; it is a feedback loop. Picture borrowing demand surging in a USDC market. Borrowers pile in, utilization climbs past the kink, slope2 engages and the borrow rate spikes. Two forces respond almost immediately. Marginal borrowers, now facing an expensive rate, repay or close positions. At the same time, the higher borrow rate feeds a higher supply yield, which pulls fresh deposits into the pool. Both forces push utilization back down toward the kink, and the rate settles. The curve behaves like a thermostat, with price as the actuator.

Concretely: say a market sits at the 80 percent kink paying borrowers 4 percent. A large borrower opens a position that lifts utilization to 92 percent. The rate does not rise by a proportional sliver; it vaults past 45 percent, because the position has crossed into the slope2 region. Within minutes, arbitrageurs and yield-seekers notice the double-digit supply APY, deposit fresh liquidity, and the borrower who triggered the spike may repay rather than bleed. An hour later the market can be back near the kink as if nothing happened. The curve did its job, and no one decided anything.

This is why on-chain rates are volatile within a single day and reprice every block: each market is running a live auction for its own remaining liquidity. It is also why a rate you saw an hour ago may be meaningless now. The curve does not care about your plans; it cares only about the pool, and it will happily quote you 40 percent one minute and 5 percent the next if the flows demand it.

The 100 percent utilization trap

The reason slope2 is so steep becomes obvious at the extreme. When utilization reaches 100 percent, the pool is fully lent and no supplier can withdraw until a borrower repays or a new deposit arrives. The steep slope above the kink exists precisely to make that state rare and expensive, so that the rate screams long before the pool locks up.

It is not a hypothetical. During the April 2026 KelpDAO exploit, borrowed WETH pushed several Aave markets to full utilization, freezing supplier withdrawals across multiple deployments while the resulting bad debt was worked out, as NewsBTC reported. The episode is a reminder that a fat supply APY at very high utilization can be a trap rather than a gift: the market is paying you well precisely because you might not be able to leave. Anyone lending into a market above 95 percent utilization is being compensated for withdrawal risk, whether they realize it or not, a dynamic we unpack further in our guide to how DeFi liquidations work.

Morpho’s adaptive curve: rates that learn

Morpho Blue ships a different default model, the AdaptiveCurveIRM, and it represents the clearest break from the kinked curve in years. It combines two mechanisms. The first is a curve, much like a conventional kinked model, centered on a target utilization of 90 percent rather than 80: per Morpho’s documentation, at the 90 percent target with a 4 percent base rate the curve quadruples to about 16 percent at 100 percent utilization and quarters to roughly 1 percent at 0 percent. The second, and the novel part, is an adaptive mechanism that shifts the entire curve up or down over time.

When utilization sits above the 90 percent target, the whole curve drifts upward, making borrowing steadily more expensive until demand cools. When utilization sits below target, the curve drifts downward to coax borrowers back. Crucially, the speed of that drift scales with distance from target: Morpho’s docs give examples where the target rate roughly doubles over ten days at 95 percent utilization, doubles in about five days at 100 percent, and halves over ten days at 45 percent. Instead of a fixed curve that governance must nudge by hand, the market discovers its own rate over time, and it automatically tracks external conditions, since if Morpho is cheaper than rival venues, borrowers flood in, utilization stays above target, and the curve climbs until equilibrium returns.

Under the hood this is a controller borrowed from control theory. As Morpho described when it introduced the model, the rate depends not only on current utilization but on the accumulated history of the gap between actual and target utilization, closer in spirit to a PID controller than to a lookup table. The payoff is efficiency. Because Morpho markets are isolated and do not rehypothecate collateral, they can safely run hotter, which is why the target is 90 percent rather than the 80 percent typical of pooled designs; more of every deposited dollar is working, for the same collateral.

A design space, not one curve

The kinked and adaptive models are the two most common, but they are not the only options, and in 2026 the interest rate model has become a design surface that protocols compete on rather than an afterthought. Fraxlend uses a time-weighted variable rate with no fixed curve at all: it sets a half-life (around 12 hours) so that when utilization sits at 100 percent the rate doubles every 12 hours, and when it sits at 0 percent the rate halves every 12 hours, drifting continuously toward a target, as Flywheel DeFi details. Curve’s crvUSD and LlamaLend markets pair a monetary-policy-style rate tied to the stablecoin’s peg with soft liquidations, and Euler V2 lets each vault choose its own IRM, whether kinked, adaptive or custom.

The trend is unmistakable. On modular platforms like Morpho and Euler, whoever creates a market chooses the interest rate model the same way they choose the collateral, the oracle and the loan-to-value ceiling. The curve is no longer a protocol-wide constant; it is a parameter, and picking a bad one is now a way to build a bad market, just as a bad oracle is. The IRM has graduated from plumbing to product.

August 2026: governance tunes the curves in real time

The clearest proof that on-chain rates are a policy choice rather than a market given is to watch a DAO adjust them. In August 2026, Aave’s Risk Stewards, operated by the service provider TokenLogic, pushed a batch of stablecoin curve changes documented on the Aave governance forum. Slope1 rose by 50 basis points on 22 stablecoin reserves (25 basis points for Base USDC), on the logic that persistent borrow demand could support higher rates. The projected effect was around 2.55 million dollars a year in extra revenue for the Aave DAO.

The batch went further. For USDe, Ethena’s synthetic dollar, the base rate was lifted to 5.25 percent APR with a slope1 of just 0.25 percent, an explicit move to stop traders from recycling borrowed USDe back into staked sUSDe at an artificially cheap cost; the change aligns the borrow rate with Ethena’s native staking yield, which the stewards estimated would lift the staking rate by roughly 50 basis points. Elsewhere, slope2 on X Layer was cut from 40 percent to 20 percent, and optimal utilization was nudged up by 2 percent on some Ethereum and Plasma stablecoin reserves, freeing roughly 116 million dollars of borrowing capacity and pulling peak borrow rates down from the mid-40s to the mid-20s in percentage terms.

The rationale, in the stewards’ own words, was blunt. TokenLogic wrote that “stablecoin rates need to reflect both sides of each market. Borrow demand can support a higher rate when users continue to take on debt.” This is the on-chain equivalent of a rate-setting meeting, except the agenda is a public forum thread and the decision executes as an on-chain transaction. Adaptive models like Morpho’s automate much of this reactivity; kinked models like Aave’s still lean on human governance to move the curve, which is a feature (accountability) and a bug (latency) at the same time.

Floating rates and the overnight-rate problem

Every model discussed so far produces a floating rate that reprices continuously. Tom Wan, head of data at Entropy Advisors, summarized the state of the market to Markets Media: DeFi “has lived entirely at the overnight point, floating rates that reprice every block.” That is ideal for a lender during a demand squeeze and miserable for anyone who needs to know their cost of capital next month.

The answer is fixed-rate, fixed-term credit, which reached real scale in 2026 through Morpho Midnight, Pendle’s principal and yield tokens, and auction-based venues like Term Finance. Paul Frambot, Morpho’s founder, has argued to The Block that fixed-rate lending “is fundamental to how global credit markets operate. Without it, onchain markets remain incomplete.” Fixed rates lock a cost at origination and only reprice at rollover, giving borrowers certainty at the price of flexibility; most on-chain credit is still floating, but the term structure is filling in.

Floating on-chain rates also do not exist in a vacuum. They loosely track the wider dollar money market through arbitrage: when short-term Treasuries yield around 4 to 5 percent, stablecoin supply rates gravitate toward that level, because capital leaves any on-chain market paying less than a tokenized T-bill can. That link has tightened as institutions have arrived, a shift we chart in our coverage of crypto ETF approvals and the broader year-end macro outlook; the Federal Reserve’s policy path now shows up, indirectly, in what a USDC pool pays.

How on-chain rates get gamed

Because the rate is mechanical, it can be manipulated, and the attack surface is worth knowing. The simplest exploit is utilization manipulation: in a thin market, a whale can borrow a large share of the pool to spike utilization and the rate, or dump supply to crush it, profiting from a related position elsewhere. Deep, liquid markets resist this; small or newly created ones do not, which is one more reason to check the size of a pool before trusting its quoted rate.

A second vector is the interplay between rates and oracles. In August 2026 an attacker manipulated a Pendle-based price feed used by a third-party Morpho market to force liquidations, netting a profit while the underlying asset itself was unharmed; it was not an attack on the IRM directly, but it shows how the machinery around the rate can be turned against borrowers. A third is looping, or recycling: borrowing an asset to redeposit a correlated, yield-bearing version of it, which extracts subsidized yield whenever the borrow rate is set below the staking yield. The USDe adjustment Aave’s stewards made in August was a direct response to exactly this pattern.

Finally, when rates spike, liquidations follow, and searcher bots compete to capture the liquidation bonuses; that is the same maximal extractable value supply chain we describe in our piece on MEV strategies. And the interest rate model is itself a smart contract, carrying the same risk as any other on-chain code: a mispriced or buggy curve can be as dangerous as a bad oracle, which is why risk service providers such as Chaos Labs, LlamaRisk and Gauntlet monitor them continuously, and why the audited-yet-exploited pattern we examined in CertiK’s August ledger applies to rate models too.

Who regulates on-chain interest rates?

The short answer is that no authority sets them, and that is the entire point. Unlike a bank deposit rate, which a firm administers and a regulator can scrutinize, an on-chain lending rate is output by a public contract that anyone can read and no one controls. In the United States, the SEC under Chair Paul Atkins has adopted a friendlier posture toward the sector through its Project Crypto initiative and has floated a safe harbor for DeFi, while the CLARITY Act would divide oversight between the SEC and the CFTC and carve out self-custodial, non-intermediated DeFi. None of that turns a DeFi lender into a bank.

The practical consequences are the ones lenders should internalize. There is no federal deposit insurance on an on-chain balance, no lender of last resort, and no regulator guaranteeing that a quoted rate is sustainable; when a loan goes bad, the loss is socialized among suppliers or absorbed by a protocol backstop, not covered by the government. A double-digit stablecoin APY is not a free lunch or a mispricing to arbitrage forever; it is the market’s price for smart-contract risk, liquidity risk and the absence of any safety net. The curve tells you what you are paid. It does not tell you what you are risking.

That gap between yield and safety is precisely why the rate can look generous. A bank can offer a low savings rate because the deposit is insured and the institution absorbs first losses; a DeFi pool offers a higher one because it hands those functions back to the depositor. When a protocol does build a backstop, such as Aave’s staking module that can be slashed to cover a shortfall, that cushion is funded by other users rather than a government, and it has hard limits. Reading a rate honestly means pricing in the possibility, however small, that the number goes to zero.

How to read a rate before you lend or borrow

A short checklist turns the mechanics above into a pre-trade routine.

• Check current utilization and where the kink sits. A market near 95 percent can spike or freeze; one near 20 percent will pay lenders a thin yield no matter how attractive the headline borrow rate looks.
• Identify the model. A kinked curve moves only when governance moves it; an adaptive curve drifts on its own; a time-weighted rate can double or halve on a schedule.
• Read the reserve factor. It is the wedge between what borrowers pay and what suppliers earn, and it is a lever the DAO can change without warning.
• Borrowers should model the rate above the kink, not just today’s number. Slope2 can multiply a borrowing cost within hours if utilization climbs.
• Lenders should treat a rich APY at very high utilization as a warning as much as a reward, because it often signals you may not be able to withdraw on demand.
• Note who tunes the curve and how fast. A protocol with active Risk Stewards can react in days; a passive one cannot.
• Back-solve the supply yield yourself: borrow rate times utilization times one minus the reserve factor. If a dashboard’s number does not reconcile, find out why before depositing.

Interest rate models are the quiet half of DeFi lending. Liquidations get the headlines because they are violent and visible, but the curve is what runs every second in between, deciding who borrows, who supplies and how close a market drifts toward the edge. In 2026, with on-chain credit crossing into the tens of billions and institutions plugging in through products like Robinhood’s Earn, the humble utilization curve has become one of the most consequential pieces of plumbing in crypto. It is worth knowing how to read.

Frequently Asked Questions

How are DeFi interest rates set?

They are set algorithmically by an interest rate model, a smart contract that computes the borrow rate from a market’s utilization (borrowed funds divided by supplied funds). There is no central bank or loan officer; when utilization rises, the rate rises, and when it falls, the rate falls. Lenders earn the borrow rate scaled by utilization and reduced by a protocol reserve factor.

What is utilization in DeFi lending?

Utilization is the share of a pool’s deposits that is currently borrowed. It drives both the interest rate and your ability to withdraw: at 100 percent utilization the pool is fully lent, so suppliers cannot exit until a borrower repays or new deposits arrive. Most models aim to keep utilization near an optimal target, often 80 to 90 percent.

Why did my DeFi supply APY suddenly jump?

Almost always because utilization rose, often crossing the kink where the steep part of the curve begins. Supply APY equals the borrow rate times utilization times one minus the reserve factor, so when a pool fills up both the borrow rate and utilization climb together and lender yield can rise sharply, sometimes within a single block.

What is the kink or optimal utilization point?

The kink is the utilization level where a lending curve bends from a gentle slope to a steep one. Below it, borrowing is kept cheap to keep capital productive; above it, the rate climbs quickly to force repayment and protect the pool’s liquidity, so that suppliers can still withdraw. Optimal utilization is commonly set around 80 to 90 percent depending on the asset.

Are DeFi lending rates fixed or variable?

Most are variable and floating, repricing every block as utilization changes. Fixed-rate, fixed-term products do exist in 2026, through venues like Morpho Midnight, Pendle and Term Finance, and they lock a rate at origination; but the majority of on-chain credit still charges a floating rate driven by the utilization curve.

By Yuki Tanaka, senior DeFi correspondent at HOGE Wire.