Unpacking Aave: Quantifying the Leverage in DeFi's Largest Looping Market

This report was originally sent directly to clients of Galaxy Trading and Galaxy Asset Management on May 1, 2026. Trade or invest with Galaxy to receive the most timely research directly in your inbox.

Introduction

A full view of Aave, the largest lending market in decentralized finance (DeFi), is hard to come by. While onchain data makes individual loans transparent and accessible, aggregating that data into a coherent, high-level view has been largely underexplored. This is partly due to the complexity and scale of some of these applications, but also because the industry has never had a pressing reason to ask the question. Until now.

The recent $290 million exploit of KelpDAO’s rsETH and LayerZero’s bridging infrastructure sent shockwaves throughout the DeFi ecosystem. The attacker exploited a vulnerability to unlock 116,500 rsETH from the Ethereum mainnet escrow. The stolen tokens were immediately deposited as collateral on Aave, Compound, and Euler, against which the attacker borrowed an estimated $236 million in WETH and wstETH. As a result, Aave suffered the equivalent of a bank run, with depositors withdrawing assets, pushing utilization rates on stablecoins and ETH pools near 100%, and bad debt built up to the tune of $123 million. In response, the industry coalesced around the DeFi United initiative to raise more than $300 million to backfill the hole created by the exploit.

For the first time, there is a strong impetus to understand what the loans underpinning the largest onchain lending market look like. The following analysis draws on Galaxy Research's proprietary view of Aave markets to examine the composition of those loans, the amount of leverage carried by borrowers, and the sensitivity of these markets to events like the KelpDAO exploit.

Unpacking Aave Loans

There are two types of loans on Aave. Vanilla (non-e-mode) borrows use uncorrelated assets on the collateral and borrow ends and have baseline loan parameters. By contrast, “e-mode” borrows use correlated collateral and borrow assets and receive preferential loan parameters (primarily increased loan-to-value ratios, or LTVs, and higher liquidation thresholds) to reflect the intended steady relationship between the values of the asset and the liability. The vanilla category includes loans where the borrower pledges wrapped bitcoin (WBTC) to borrow the stablecoin USDT. The e-mode category includes loans that “loop” staked ether (stETH) to borrow ETH in order to gain leveraged exposure to the Ethereum staking rate. Because the assets in e-mode loans are correlated, they primarily face liquidation in one of two scenarios: 1) when borrow costs exceed the yield achievable by the collateral asset (the liability side of the loan grows faster than the asset side), which pushes LTVs toward liquidation thresholds, or 2) when the collateral asset depegs, effectively breaking correlation with the borrow asset, and drags the asset side of the loan down. Looped loans account for a meaningful share of Aave’s borrow total value locked (TVL), representing a unique risk to the protocol. This was evidenced [1][2] in the aftermath of the KelpDAO exploit as tightly levered looped loans face various degrees of liquidation risk that threaten the health of Aave’s markets. The following section examines the e-mode loan landscape of Aave V3 Core, the largest onchain e-mode market (and the largest onchain lending market overall).

Assumptions and Filters

To add some clarity on the data used and how it was filtered: This analysis is built from an onchain snapshot at Ethereum block height 24932111 (April 22,2026, 01:14:23 AM UTC). Position totals and line notional amounts are taken from the indexer (oracle USD/base currency units). We do not reprice assets after the snapshot block, replay pool state forward, blend in live market feeds or offchain overlays, or incorporate any information outside of how Aave manages its markets.

Structural filters

We filtered loans so dust (small loans of insignificant value) or extreme outliers would not dominate the headline cohort statistics:

• Minimum debt ($100): We excluded positions below this threshold from the main aggregates. That removes dust but skews summaries toward larger loans.

• Health factor (HF) reporting cap (HF <= 50): We also excluded positions with snapshot health factors above 50 from the headline cohorts. Among loans that do fall below the cap, debt-weighted HF averages and HF percentile summaries count only loans with snapshot health factors greater than or equal to 1 and less than or equal to the cap of 50. (Loans with health factors below 1 are omitted from these HF readouts; any loan where the health factor equals 1 is included.) The higher the health factor, the safer the loan; a health factor below 1 means the position is eligible for liquidation.

• Debt-to-equity (D/E): We computed this ratio only for loans where collateral exceeds debt (positive equity). We omitted positions without positive equity from D/E distributions and from the debt-weighted average D/E. That rule is separate from the minimum-debt filter: a loan can clear the $100 threshold and still be excluded from D/E-only aggregates when equity is not positive.

• E-mode collateral analysis: the e-mode collateral analysis only observes assets that are most likely to be looped. It excludes volatile assets that get e-mode designation for certain loan types but are less likely to be used as collateral in looped positions.

E-mode granular depeg

These sections hold borrow notional amounts flat in base units and haircut only the collateral lines in the shocked asset, using the same weighted numerator/debt HF framework as above.

Per-collateral-symbol depeg: The numeric grids apply a price shock only to collateral lines whose symbol matches the target ticker (all other collateral and borrow positions remain unchanged). Stressed HF is calculated as the sum of each collateral position's shocked value multiplied by its liquidation threshold, divided by total debt, where each line's liquidation threshold reflects its effective e-mode or position-level setting.

Who enters the simulator: Positions must be in e-mode, exceed the minimum debt threshold, have a snapshot HF ≥ 1, and carry positive enabled collateral. The HF ≤ HF_CAP (health factor at or below the cap) filter is not re-applied inside the simulator, but because the pipeline passes through an already-filtered book, the results remain consistent with the capped cohort used throughout.  

Debt‑weighted snapshot table under each symbol: built on the same capped book with minimum debt, HF ≤ HF_CAP, HF >= 1, and ≥ $1 enabled in that ticker; the debt‑weighted HF in that table uses HF >= 1 and ≤ HF_CAP (same band as headline HF summaries).

E-mode implied loops

The “Implied # of loops” row is generated only for symbols in the liquid staking ETH, liquid restaking ETH, yield-bearing stable, or Pendle principal token (PT- prefix) e-mode depeg buckets. For that row only, debt-weighted average LTV and (D/E) are computed on loans where the named symbol is ≥ 99% of enabled collateral (exact ticker match for spot tickers; PT lines match the full PT-… symbol), plus ≥ $1 in that symbol, debt ≥ $100, snapshot HF ≤ 50, HF >= 1. If no loan passes the 99% filter or the formula has no finite solution, the loops row is omitted; the rest of the per-symbol snapshot table is unchanged and still uses the broader ≥$1 cohort.

Implied number of loops is calculated using natural logs N = ln((1 − (D/E)(1 − L)) / L) / ln(L) where L = debt-weighted average current LTV as a decimal (LTV% ÷ 100) and (D/E) = debt-weighted average debt/equity where equity > 0, both from the ≥99% concentration subcohort described earlier. Weights are each loan’s debt (Di); LTV% and (D/E) are debt-weighted averages Σ(Di x mi)/ΣDi over the ≥99% subcohort, where mi is that loan’s current LTV% or (D/E)i.

A heuristic note on the calculation: the implied loop count is only meaningful when the ≥99% slice approximates the idealized case of one collateral asset looped at a constant LTV; it ignores path-dependent LTV, multi-asset books, and execution frictions. Use it as a sanity-scale for nested leverage inside individual collateral assets, not a realized total loop count for all loans backed by a specific asset.

What this analysis is not

This is not a simulation of liquidation ordering, bonus, partial liquidations, bad debt accrual after the snapshot, or any related measure. It is a transparent, repeatable read on the e-mode book at the snapshot time to show how much risk sits in the system across various asset shocks from a high level.

Market-Wide View

After applying the aforementioned filters, we find that Aave houses $10.7 billion in outstanding loans against $17.37 billion in collateral at a market-wide debt/collateral ratio of 61.65%. E-mode loans make up the greater portion of Aave’s outstanding debt, capturing 58.84% of the book. In the e-mode cohort is $6.3 billion of debt against $7.05 billion of collateral, for an 89.4% debt/collateral ratio. Non-e-mode loans are less than half as levered, at an aggregate debt/collateral ratio of 42.7%.

This table below summarizes debt-weighted risk metrics for the filtered book, split into all positions, e-mode, and vanilla. E-mode borrowers carry much higher leverage on average. Their debt-weighted LTV is about 90% with a debt-weighted health factor near 1.05 and debt-to-equity around 10.4, meaning a small collateral shock can move many of these loans toward stress. Non–e-mode loans look far more cushioned, with a debt-weighted LTV near 47%, HF near 1.88, and D/E near 1.

Debt-weighted average is calculated as D/E = Σi (Di × (Di ÷ (Ci − Di))) ÷ Σi Di over loans with Ci > Di, where Di is debt and Ci is collateral for individual position i.

The next table ranks all enabled collateral lines across Aave as a cumulative market by how much USD value sits in each symbol across the cohort. ETH-linked collateral dominates: WETH, wrapped Etherfi restaked ETH (weETH), and wrapped Lido stETH (wstETH) each account for a large share of the total (roughly 25%, 21%, and 14%, respectively, for a combined share of 58.7% of enabled collateral), with WBTC also material (~13%). So, a handful of tickers carry most of the book’s posted collateral. Below that tail there are smaller but still meaningful slices in stablecoins and other yield tokens.

The following table ranks borrowed assets by symbol in USD and as a share of the cohort’s total borrowing. WETH dominates liabilities at a little over half of borrowed notional (~51%), which is typical when we consider the meaningful presence of leveraged looping strategies where ETH-correlated assets serve as collateral. Stablecoin borrows are also large: USDT and USDC together make up close to 40% of the total (~21% and ~18%, respectively), and everything below that is comparatively small.

E-mode High-Level View

Within e-mode loans, enabled collateral is heavily skewed toward ETH staking/restaking wrappers: weETH (not to be confused with WETH) alone is nearly half of the category (~48%), and combined with rsETH and wstETH, the top three lines cover roughly 80% of e-mode posted collateral. As a result, e-mode risk is less “diversified collateral” and more a concentrated bet on Ethereum staking basis.

Yield‑bearing stables and related legs still matter at single‑digit percentages each, showing some secondary exposure to stable yield/peg mechanics, not just pure ETH. The tail includes Pendle principal tokens and small WETH/LBTC (Lombard BTC) lines, which are individually tiny as shares but are nonetheless useful markers. Note: these values are based on a point-in-time snapshot. The cyclicality of onchain stablecoin yields, with yields at multi-year lows, reduces the opportunity for stablecoin looping trades and allows ETH-based collateral to dominate on a relative basis.

For e-mode borrows, the book is overwhelmingly WETH‑denominated. WETH alone is ~85% of e‑mode debt, which is exactly what you would expect when users are looping ETH-correlated collateral against ETH debt. Stablecoins still show up meaningfully. USDT, USDe, and USDC together are on the order of low‑teens percentages of e‑mode borrows.

Again, these values are based on a point-in-time snapshot. The cyclicality of onchain stablecoin yields, with yields at multi-year lows, reduces the opportunity for stablecoin looping trades and allows ETH-based borrows to dominate on a relative basis.

How Levered are E-Mode Loops?

This slice ranks e-mode positions that touch each collateral ticker (with debt-weighted risk and, where applicable, implied loop counts on the ≥99% single-asset collateral subcohort). As a result, it reads like a “which asset is being looping hardest?” scoreboard, unlike the neutral market-cap tables above. Liquid restaking/staking ETH wrappers cluster at the top with high debt-weighted LTVs, D/E often in the high single digits to low teens, and health factors not far above 1. This is consistent with tight, repeated ETH-beta loops.

Yield-dollar legs (sUSDe, SyrupUSDT, USDe) sit in the same broad band with 80%+ LTV and material implied loops, which is what you’d expect when borrowers use e-mode to run stable yield/carry strategies. Unexpired Pendle PTs show up among the most levered names as well, though they only total about 3.6% of collateral supplied among the cohort of e-mode loan collateral that is likely to be looped.

Quantifying the Risk

The data reveals a heavily concentrated leverage risk profile on Aave V3 Core, with a small number of liquid staking and synthetic dollar assets accounting for most of the stressed exposure across depeg scenarios. weETH (again, not to be confused with WETH) is by far the dominant risk concentration. As the chart below shows, if weETH were to depeg from ETH by 10%, Aave would be left with ~$2.47B in debt against ~$2.42B in post-shock collateral. In other words, the collateral basket would already be underwater relative to the debt at that stress level, with 205 accounts breaching a health factor below 1. What's even more striking is how sharply weETH exposure jumps between the 3% and 5% depeg bands (~$1.6B to ~$2.1B in debt), suggesting a cluster of highly leveraged positions that would get triggered in that range.

rsETH is the second largest concern at ~$1.16B debt and ~$1.14B post-shock collateral at 10%, again with collateral dipping below debt in that scenario — but its account count is only 22, pointing to a small number of very large, highly concentrated positions. A similar story holds for wstETH (~$312M debt, 48 accounts) and OSETH (~$277M, 155 accounts), both of which show collateral falling below debt at the 10% level, though at far smaller scale.

Stablecoin-pegged assets (USDe, sUSDe, PT variants) show moderate sensitivity — their debt figures are in the $150m–$300m range and collateral holds up better at lower depeg levels, but by 10% the gap narrows meaningfully. The PT (principal token) instruments are interesting because they have a hard maturity, so their depeg risk is more bounded in practice, though the data still shows meaningful stress at 5%–10%.

The smaller assets (ezETH, cbETH, rETH, LBTC, syrupUSDT) are relatively immaterial in absolute terms — all under $10m in debt at 10% depeg — though LBTC jumps from near-zero to ~$2.9m between 7% and 10%, hinting at threshold-sensitive liquidation clustering. The overall picture is that Aave leverage risk is top-heavy in weETH.

Conclusion

This note is a point‑in‑time, snapshot‑style read of Aave V3 e‑mode leverage. It highlights how borrowing is concentrated across a small set of ETH‑linked collateral and WETH‑heavy liabilities; how aggressive typical LTV, health factor, and D/E profiles are inside the filtered cohort; and how much notional stress shows up when simple collateral haircuts are applied.

The primary takeaway is concentration and sensitivity. A large share of the economics sits in a handful of LST and LRT assets where small depeg assumptions can move a material amount of debt toward liquidation.