Aave on Ethereum: 4.3% of its USDe is borrowed, 93.6% of its USDT

On 1-Oct-2026, the USDe market on Aave v3 Ethereum Core had 557.46 million dollars deposited and 24.21 million borrowed: a utilization rate (the portion of deposits currently borrowed) of 4.3%. On the same day, USDT in the same market was 93.6% borrowed. Both assets share a commonality decided by an Aave vote: the protocol values them using the same price aggregator, meaning that for collateral calculations, one unit of one is worth the same as one unit of the other. In all other respects, the risk committee treats USDe as the riskier of the two — 0% LTV outside of e-mode, meaning no borrowing power against it except in the efficiency mode that groups correlated assets; a 25% reserve factor (the portion of interest kept by the protocol) compared to 10%; and a borrow cap equivalent to 7.1% of its supply cap. Between 28-Aug and 29-Sep-2026, the base borrow rate rose from 0 to 6.60% in eight steps, anchoring it to the yield that USDe itself pays when staking, and debt fell from 442.20 million on 27-Aug-2026 to 24.21. Three alternative explanations do not fit the data: the borrow cap was 700 million throughout the entire ladder and was never reached, the same pool was lending 73.9% of its USDe on 1-Aug-2026 with a zero base rate, and the debt did not leave the protocol but rather switched currencies within it.

Why is USDe deposited in Aave not lent out while USDT is?

Utilization measures what fraction of the deposited amount is lent out, and in Aave, it is the variable that sets the price: the higher it is, the more expensive it is to borrow. It is also what indicates, without the need for interpretation, what part of a currency is circulating and what part is stored.

The three major stablecoins of Aave v3 Ethereum Core reached 1-oct-2026 with incomparable readings. USDT had 2,849.74 million dollars deposited and 2,667.49 borrowed: 93.6% utilization and 182.25 million available for withdrawal. USDe had 557.46 million deposited and 24.21 borrowed: 4.3% utilization and 533.25 million idle. This represents an 89.3-point difference between two assets to which the same protocol assigns, on that same day, the same price.

The USDC row requires methodological precision, as it is the one that fluctuates within the day itself. On Sep-30-2026, USDC utilization in Aave v3 Core remained within the 94.5 to 95.1% band for almost the entire day, with two minute-long spikes that pushed it to 99.9%: deposit inflows and outflows of 120 to 200 million that are immediately unwound. A reading taken during one of those spikes describes the spike and not the day. The figures in the table are from two readings on Oct-1-2026 taken fifteen minutes apart, neither during a spike, which show the same debt to the cent and utilization rates of 93.8 and 94.0% for USDC.

With that correction, USDC and USDT remain where one would expect for two assets with identical parameters: their utilization rates for 1-oct-2026 are separated by less than half a point: 94.0% for USDC and 93.6% for USDT. The market asymmetry does not lie between those two.

It remains to be explained why 557.46 million USDe are sitting there if almost no one is borrowing them. Outside of e-mode, USDe has an LTV of 0 and cannot be used as collateral for anything; within e-mode 11, it is valued at 90% LTV, with a liquidation threshold of 93%, and allows for the withdrawal of exactly two assets, USDC and USDT. The USDe deposit in Aave is, above all, a collateral box. Our analysis of the Ethena model described in May 2026 the "Aavethena loop" —depositing sUSDe as collateral in Aave, borrowing USDC, buying more USDe, and depositing again— as the engine that inflated and then deflated Ethena's size; it is the most plausible explanation for what sustains that balance, although attributing it requires account-level data that we have not reviewed.

Why does Aave value USDe at the same price as USDT?

Because it was decided by vote. On 2-ene-2025 Chaos Labs, co-authored by LlamaRisk, proposed to "hardcode the price of USDe to match the price of USDT in Aave feeds." The argument was based on liquidation risk: the previous oracle used the market price of USDe, and Chaos Labs calculated that a 5 % deviation would leave about 300 million dollars in USDe-backed positions at risk of liquidation, while market liquidity only allowed for the execution of about 6 million within a 4 % price impact —the margin charged by the liquidator—. The rest of those liquidations would not be executed because they would not be profitable, and the loss would remain in the protocol as bad debt.

The decision was approved by vote on 6-feb-2025 with 645.6 thousand votes, the code change was merged on 3-mar-2025, and the Aave oracle began returning the new source for USDe on 8-mar-2025. It remains in effect: the reading from 1-oct-2026 gives the same value for USDe and USDT —0.99964 dollars—, while USDC follows a different path and trades at 0.99999.

A technical clarification is necessary to make the statement more accurate without weakening it. USDe and USDT do not literally share the same feed contract: each has its own adapter, and both declare the same description, "Capped USDT/USD," and read the same Chainlink aggregator for the USDT/USD pair, returning the same response. The effect is that of an identical face value imposed by decree: within Aave, and solely for calculating collateral and liquidations, one USDe is one USDT.

That decree should be read alongside who signed it. In the 2025 thread itself, a participant pointed out that Chaos Labs had worked for Ethena designing its risk framework prior to launch and that LlamaRisk is part of the Ethena risk committee, which is remunerated. This is a public observation from the thread and not our own conclusion: the two teams recommending the USDe parameters on Aave have a declared prior relationship with the USDe issuer. The details of how this synthetic dollar is constructed can be found in the analysis linked above and in the change in its backing composition.

The rest of the table goes in the opposite direction of the decree. USDe pays a 25% reserve factor compared to 10% for USDC and USDT, has a lower reserve liquidation threshold, a higher liquidation bonus, and a borrow cap that is 7.1% of its supply cap, whereas that ratio is 90.0% for USDC and 88.8% for USDT. In the oracle, both are valued equally; in the risk parameters, USDe pays 15 points more in reserve factor and lacks the 75 LTV points that USDT does have.

What did the Aave risk committee submit between 28-Aug and 29-Sep-2026?

The base rate of the USDe curve. On an Aave interest rate curve, the base rate is what is paid at zero utilization: a floor that does not depend on demand. USDC and USDT have it at 0 on 1-oct-2026, so with an empty pool, borrowing them would be free. USDe also started at 0 on 1-ago-2026 and ended at 6.60%.

The dates in the table are those of execution in the contract, localized by bisection of the rate strategy, and not those of publication for each recommendation. The two series are similar but do not coincide, and mixing them shifts the steps by up to a full day.

The combination matters. A high floor with an almost flat slope means that the price of borrowing USDe on Aave is no longer set by demand, but by the parameter. With 4.3% utilization on 1-oct-2026, the rate was 6.61%; with the pool at 90%, it would be 6.85%. Twenty-four basis points of range between being nearly empty and being at the kink, which is the utilization point after which the rate begins to rise sharply.

The anchor is declared and explicit. LlamaRisk wrote on 2-sep-2026 that the steps continued "the move toward the 5.25% base rate established in the stablecoin rate changes proposed by TokenLogic, which prices USDe borrowing in line with Ethena's native staking rate." TokenLogic had formulated it bluntly on 24-ago-2026: aligning USDe borrowing rates with its staking rates "so that users cannot borrow these assets at rates lower than what their native protocols allow." On 28-sep-2026 LlamaRisk provided the full arithmetic: the seven-day average of sUSDe yield ending 23-sep-2026 was 5.00% and the thirty-day average was 4.83%; "grossed up" by the 25% reserve factor, those two figures equal 6.66% and 6.44%. The 6.60% falls between the two. We have not verified the sUSDe yield on our own: the 5.00% figure is what LlamaRisk declares, and it is the relevant one here because it is the anchor that the committee itself claims to be using.

What that anchor does is close a specific trade. Before the ladder, on 27-ago-2026, the USDe curve had a base rate of 0 and a slope 1 of 4.00% with an optimal level at 90%; at the 60.6% utilization that day, the borrowing rate came out to 2.70%. Borrowing USDe at 2.70% and staking to earn around 5% leaves a margin of just over two points. On 1-oct-2026, the USDe borrowing rate was 6.61% compared to the 5.00% that LlamaRisk attributes to sUSDe: the same loop subtracts 1.61 points. By raising the floor 660 basis points, the committee set the price of lending USDe 1.61 points above what USDe itself yields according to LlamaRisk, which is the level at which the trade is no longer profitable.

Did the 50 million cap stall USDe lending on Aave?

It is the most immediate alternative explanation: if the protocol places a 50 million cap on USDe debt, utilization decreases due to arithmetic rather than price. The reserve configuration read from historical blocks rules this out.

The USDe borrow cap on Aave v3 Core was 700 million dollars on 1-ago, 27-ago, and 12-sep-2026, with a supply cap of 902 million. Throughout the entire ladder, therefore, the ceiling remained at 700 million: 1.6 times the debt peak of 442.20 million on 27-ago-2026. It did not tighten at any point during the period. The reduction to a 50 million borrow cap and 700 million supply cap was recommended by LlamaRisk on 28-sep-2026 and was in effect by 29-sep-2026, when debt had already fallen to 37.23 million: it arrived after the drop and did not cause it.

The distance to the current ceiling points in the same direction. A market constrained by its cap stays pinned to it, yet USDe debt was 74.5 % of the ceiling on 29-sep-2026, 63.3 % on 30-sep-2026, and 48.4 % on 1-oct-2026. It moved away from the ceiling instead of pressing against it.

The cap does set a maximum, even if it has not been reached. With 557.46 million deposited on 1-oct-2026, the 50 million limit restricts USDe utilization to 9.0 %, regardless of demand. The observed result —4.3 %— is below that limit, and the step-by-step trajectory aligns with the rate ladder rather than the date of the cut. The ceiling exists and constrains; what emptied the pool was the price.

The contrast with the neighboring asset in the same month goes in the opposite direction: on 7-sep-2026, Core's USDC was at 95.7% of its 2,250 billion supply cap and LlamaRisk recommended increasing it; by 1-oct-2026, that cap was at 2,700 billion. In the same market and during the same weeks, the committee expanded the circulating capacity of USDC and cut that of USDe.

Is USDe also being hoarded where there is no risk committee?

If USDe were to remain stagnant by its very nature —because everyone who holds it wants the yield and no one wants to owe it— the Aave committee would be a mere spectator. There are two ways to verify this, and both yield the same partial answer: the asset tends to be hoarded, and yet the specific draining of the Aave pool is not explained by that.

The first is to look at the same pool before the ladder. On 1-ago-2026, with the USDe base rate at 0, Aave v3 Core had 783.60 million USDe deposited and 579.13 borrowed: a 73.9% utilization. That day the market was indeed taking USDe, and it was taking more than the current cap would even allow one to consider. Hoarding within Aave is not a property of the asset: it appeared when the 6.60% floor appeared.

The second way is to look for a market where no one sets the rate. Morpho has no risk committee: anyone can create a market with the parameters and curve they want. Across the 500 largest Morpho markets on all its chains, on 29-sep-2026, USDe as a borrowed asset totaled 0.4 million dollars. As collateral, USDe and sUSDe totaled 858.9 million, distributed across markets lending USDC on Base with 391.0 million; USDG on Robinhood with 334.3; PYUSD on Ethereum with 102.5; and RLUSD with 31.1 — all of them with utilization rates from 90.0 to 93.3 %. Without a committee, the way the market uses USDe is the same: it is posted as collateral to borrow another currency.

Outside of any protocol, the pattern repeats, albeit with a low-resolution measurement that should be taken as an order of magnitude. In a sampled window of 300 Ethereum blocks on Sep-29-2026, excluding issuance and burns, USDC moved 1.816% of its supply on the network, USDT 0.975%, and USDe 0.027%: USDe rotated about 66 times slower than USDC. This figure represents a single hour of a single day rather than an annual velocity.

USDe is an asset to be held, and that remains true with or without a committee. What changed between 28-Aug and 29-Sep-2026 in Aave v3 Core was the degree: a pool that went from 73.9% utilization on 1-Aug-2026 to 4.3% on 1-Oct-2026, with the only relevant variable adjusted manually.

Where did the debt that left USDe in Aave end up?

Between 27-Aug and 29-Sep-2026, on Aave v3 Ethereum Core, USDe debt decreased by 405.0 million dollars, USDC debt increased by 197.3 million, and USDT debt decreased by 6.5. The net balance of the two "good" currencies according to the committee is 190.8 million, equivalent to 47.1% of what USDe lost. This is a reserve figure: it indicates what happened to market aggregates, not that the same accounts were necessarily releasing one currency and taking the other.

At the account level, the measurement is provided by LlamaRisk and has not been replicated by us. There are two published windows that yield different results, so the answer depends on which one is being viewed.

In the short window, which covers 107.0 of the 405.0 million drop, accounts that completely exited Aave v3 Core account for approximately 4% of the reduction: of the 50.2 million not replaced, LlamaRisk breaks down 38.3 —33.6 million from accounts that still owe in other reserves; 0.1 from accounts that maintain collateral without debt; and 4.6 from accounts without any position in the market—. The remaining 11.9 million are not broken down by the thread. In the long window, which reaches 330.8 million, the replaced portion drops to 37.0% and there are 139.6 million in accounts that did not take any other stablecoin debt in that market. For the period after 11-sep-2026, LlamaRisk did not publish the breakdown by account.

The honest reading of the two windows is that currency transfer explains a large and time-decreasing portion of the outflow, and that protocol abandonment explains very little of the first phase. The money mostly remained in the same market and changed denomination, which is what Gresham's law describes: circulation does not stop, it moves to the cheaper medium. With USDe at 6.61% and USDT at 4.38% on 1-oct-2026, the cheap media are USDT and USDC. LlamaRisk adds the correct warning in the same text: aggregate changes at the reserve level do not prove that the same accounts moved both sides, because each reserve also captures external activity.

There is a detail that completes the symmetry. USDT is "good" for the committee under the same parameters as USDC, and its debt did not grow during the period: it decreased by 6.5 million. Its utilization rose nonetheless, from 91.7% on Aug-27 to 93.6% on Oct-1-2026, because it was the deposit that decreased. The two assets the committee rates as the best ended Oct-1-2026 at 94.0% and 93.6% utilization, with identical parameters: LTV 75%, threshold 78%, reserve factor 10%, base 0, slope 1 at 4.40%, and optimal 94%. The asymmetry lies in USDe compared to both.

Why is Gresham's law appearing inverted in the Aave market?

Thomas Gresham was the financial agent of the English Crown in Antwerp. In a letter to Elizabeth I on the occasion of her accession to the throne in 1558, he explained why the English shilling was in the state it was: her father, Henry VIII, had replaced 40% of the silver in the currency with base metals to increase state revenue without raising taxes. Since the law mandated that both old and new coins be accepted at the same face value, merchants hoarded the good ones and paid with the bad. Henry Dunning Macleod named this principle Gresham's Law nearly three centuries later.

The condition that makes the mechanism work is not that one currency is better, but that an authority forces both to be accepted at the same value. Without that decree, the seller raises the price of the bad currency and the arbitrage ends. Aave reproduces that condition with precision: the oracle forces one USDe to be worth the same as one USDT within the protocol since 8-mar-2025, and the reading from 1-oct-2026 confirms that it remains so.

The result, however, comes with the order reversed. In Gresham's England, the coin that the minting itself recognized as superior was hoarded, while the debased one circulated. In Aave, on 1-oct-2026, the asset that the risk committee classifies as worse is hoarded —USDe, with 0 LTV in the reserve, a 25 % reserve factor, and 95.7 % of its deposit sitting idle— while the one classified as better circulates: USDT, with 75 % LTV and 93.6 % borrowed. The ranking governing behavior is not the risk rating set by the committee, but rather yield; and the committee knows it, because they set the price for lending USDe exactly at its native yield.

The analogy has a limit that is worth mentioning. In 16th-century England, the same person chose which currency to pay with and which to keep. In Aave, the depositor and the borrower are different roles, and the distribution of USDe is the aggregate result of two independent decisions: it enters as e-mode collateral and does not leave because leaving costs 6.61%. It is the same distribution produced through a different path.

How much did the Aave DAO miss out on by setting USDe at 6.60%?

The reserve factor is the portion of the interest paid by the borrower that does not go to the depositor, but to the protocol. For USDe, it is 25%, the highest among the three large stablecoins on Core. Using the rate and debt read from the contract, the contribution of that reserve can be annualized.

On 27-Aug-2026, with 442.20 million in USDe debt and a borrowing rate of 2.70%, the protocol fee on that reserve was running at a rate of 2.98 million dollars per year. On 1-Oct-2026, with 24.21 million in debt and a rate of 6.61%, it was running at 0.40 million. The difference is 2.58 million dollars less annually, calculated by us based on those two readings and under the assumption that balances remain constant, which is precisely what they did not do during the month.

The figure has an uncomfortable mirror within the program itself. On Aug-24-2026, TokenLogic estimated that the 50 basis point increase in slope 1 across 22 stablecoin reserves would add approximately 2.55 million dollars in annual revenue to the DAO, provided demand remained unchanged. The two numbers do not cover the same perimeter: the 2.55 million comes from 22 reserves spread across various deployments, while the 2.58 million comes from the Core USDe reserve alone, so they are not subtracted. However, the order of magnitude is the same, and a reply in that day's thread pointed it out: the proposal quantified the revenue added by the general adjustment but failed to quantify the revenue lost by the USDe adjustment.

The anchor formula itself explains why revenue could not be the target: the sUSDe yield divided by one minus the reserve factor turns that 25% into part of the price. The higher the reserve factor, the higher the base rate must be for the loop to stop being profitable, and the faster the pool generating the fee is depleted. Anyone wishing to see how that margin is distributed between protocols and curators can find it in our ranking of who keeps the margin in DeFi lending and in the analysis of Aave, Morpho, and Compound take rates.

What is still unknown about USDe on Aave and what to watch starting Oct 1, 2026?

There is one piece of data that the table at the beginning does not capture, which downwardly corrects a tempting reading. Those who deposit USDe in Aave earned 0.22% annually from the protocol on Oct-1-2026, well below the 5.00% that LlamaRisk attributes to sUSDe staking, but that 0.22% is not all they earn: according to TokenLogic, on Aug-31-2026, of the approximately 4,100 million dollars in USDe that were in some channel with Ethena rewards, 1,300 million were in sUSDe and 1,200 million were deposited in Aave. The USDe deposit in Aave also received rewards from the issuer. At what rate is more uncertain: a forum participant estimated on Sep-1-2026 that the Merkl campaign on Ethereum was around 1.7%, up from 1.3%, and compared to promised rates approximately double that. We have not verified this. The yield sacrifice exists, but it is less than the direct subtraction between 0.22% and 5.00%.

There are also three things that this reading cannot claim, and it is worth noting them.

• What strategy supports the 557.46 million deposited. The parameters for 1-Oct-2026 enable the collateral loop via e-mode, and the "Aavethena loop" described in May 2026 aligns with what is observed; attributing the deposit to a specific operation requires account-level data that we have not reviewed.
• The breakdown of the migration after 11-Sep-2026. The two published windows cover 330.8 of the 405.0 million drop in debt; for the remainder, there is no published account-level measurement, nor is there a breakdown of the 23.7% of the short window that LlamaRisk leaves unclassified.
• The real yield of sUSDe. All figures in this article regarding Ethena staking payouts are those reported by LlamaRisk in each thread —a seven-day average of 4.75% on 11-Sep-2026 and 5.00% on 28-Sep-2026— without independent verification. These are relevant because they are the declared anchor for the base rate, yet they remain figures from an interested party.

There also remains a disagreement regarding scope that is not anyone's error. LlamaRisk aggregates the five Aave deployments that carry USDe —Core, Plasma, Monad, Mantle, and Avalanche— while this analysis measures only Ethereum Core. Across the aggregate of all five, LlamaRisk reported 199.7 million in USDe debt on 11-sep-2026 and 83.3 million on 28-sep-2026, against approximately 1,190 million deposited: a 7.0% combined utilization, above the 4.3% for Core. Both series reflect the same data at different scales, and the numbers from one should not be subtracted from the other.

For anyone wishing to verify for themselves whether the distribution holds, there are three metrics visible on the Aave interface itself that require no interpretation. The USDe debt on Aave v3 Core against its 50 million cap: if it approaches the ceiling, demand is pushing against the price and the cap becomes the constraint. The utilization of USDe against that of USDT: on 1-oct-2026 they were separated by 89.3 points, and a convergence below ten would mean the distribution has unraveled. And the USDe base rate, which is a parameter changed via a Risk Stewards thread, not by demand: it drops the day the committee decides that the sUSDe yield has decreased. The case of Aave V4 on Arc shows the other side of the same lever, where the constraint is the credit quota rather than the rate. And if you are looking for where a stablecoin yields today without entering loops, it is in our guide to stablecoin yields.

Sources and links: Chaos Labs and LlamaRisk ARFC on USDe and sUSDe price feed (2-ene-2025) · TokenLogic, August 2026 stablecoin interest rate adjustments · Risk Stewards, 2-sep-2026 (debt migration by account, short window) · Risk Stewards, 4-sep-2026 · Risk Stewards, 7-sep-2026 (USDC caps on Core) · Risk Stewards, 9-sep-2026 · Risk Stewards, 11-sep-2026 (long migration window) · Risk Stewards, 28-sep-2026 (6.60 % and cap reduction) · Aave v3 Ethereum Pool contract on Etherscan · USDe pool series on Aave v3 (DefiLlama) · Morpho markets · Ethena · Gresham's law (historical context)