Where does the yield come from?
Published 2026-08-19 · data as of 2026-08-17
DeFi has cemented its place as a cornerstone of the Ethereum economy, and one of its foundational building blocks is yield. We are far from the astronomic APYs of DeFi summer, as the ecosystem has matured to the point where even Vitalik is endorsing low risk DeFi. But the ecosystem is still evolving, and it is worth understanding how the sausage is made on Ethereum mainnet. There is roughly $144.0B of capital earning an implied $3.38B/yr, so ~2.4% blended. Where does the yield come from?
In 2026 we are lucky to have multiple data sources: this uses data from DeFiLlama, vaults.fyi and onchain reads, taken on 17th August 2026.
This analysis is experimental, any feedback or corrections are very welcome, please get in touch: adam@azfuller.com
The flow
The most obvious form of yield available on Ethereum is staking ETH itself, and indeed this forms a large part of the yield today, either directly, or via liquid staking. The next largest category is onchain borrower interest, via protocols like Aave and Morpho. Next comes real world interest (mostly tokenized t-bills), followed by token incentives and Sky's savings rate, and then a long tail. "Source" here means who pays, not whether the yield is new value: issuance is dilution, interest is a transfer, T-bills are external cash flow, incentives are subsidy. You can see the estimated breakdown below:
You can visualise how that yield flows through to its ultimate recipients in the following Sankey diagram:
$M per year at current rates, colored by ultimate payer. Fainter flows are estimates. Hover or tap a flow for detail.
We can further break down ETH staking and onchain lending to understand them in more detail.
ETH Staking
There are 42.2M ETH staked, ~35% of all ETH, across ~899K validators. 41% of that is tokenised in Liquid Staking Tokens, where yield is split between tokenholders and operators (Lido 10%, split between node operators and the DAO; Coinbase's cbETH 25%).
This is generating ~$1.9B per year at current prices, paid by consensus issuance plus priority fees & MEV. That is ~2.4% gross (stETH holders net 2.17% after fees), historically low: stake is at an all-time high.
The following shows a Sankey diagram of self stake versus liquid staking:
$M per year, colored by provider. Provider fees are skimmed from staking rewards before they reach holders: Lido 10% (split between node operators and the DAO), Binance 10%, ether.fi ~10%, Coinbase's cbETH 25%, Rocket Pool ~14% node-operator commission; the long tail is drawn at an assumed 10%. Fee flows are estimates (fainter). The direct branch is the untokenized estimate from the map above, drawn as one node.
Onchain borrowers
Lending has emerged as a pillar of DeFi, principally in the form of overcollateralised loans. Stablecoins generate 81% of the interest - $10.3B borrowed at 4.2% avg ($435M/yr). ETH is most of the remainder ($4.5B at ~2.1%, $92M/yr), but this market has compressed significantly, to the extent that supplying ETH to Aave only pays ~1.4%. Almost nobody borrows BTC, which mainly features as collateral (a significant part of Morpho's book, see the Morpho Vaults memo for more details). Meanwhile there is a small amount of unsecured onchain credit, from protocols such as 3Jane and Wildcat.
Protocols take a slice of the interest borrowers pay (e.g. Aave's reserve factor), mostly into DAO treasuries and reserves. Morpho's protocol fee switch is set to zero; its curators charge their own vault fees instead.
The flows from borrowers, via protocols, to depositors and fees are illustrated below:
Interest dollars per year, colored by borrowed coin on the left and by protocol on the right. Hover or tap a left ribbon for the borrowed balance and average rate. "Other stable" bundles PYUSD, USDe, RLUSD, USD1 and GHO. Sky's stability fees go to its surplus rather than to depositors; Morpho's take goes to external curators. The faint Fluid inflow is the DEX layer: its fToken lenders earn slightly more than its mainnet vault borrowers pay.
The rate
This memo focuses on yield generated on Ethereum mainnet, but of course dollar yield is arranged around the T-bill rate. ETH yield anchors on staking instead - which is why ETH lends below the T-bill. 3M T-bill rate is 3.87%, fed funds 3.50-3.75%, and tokenized treasury funds like BUIDL and USDY give you that rate minus a 20-50bp fee. It is worth comparing dollar yield sources against this reference point.
Dot area ∝ pool size. Dashed line: 3-month Treasury yield (FRED, 2026-08-17).
Lending markets below the line are where borrow demand is currently soft (sUSDS is below it for a different reason: a governance-set rate funded mostly by Sky's RWA revenue). Above the line you are being paid by something else, whether that is offering private credit (Maple), perp funding (Ethena) or issuer incentives (RLUSD). In all cases, if the T-bill rate moves, we can expect correlated change, and the spread above the line is payment for risk, not free return.
Notably token rewards are no longer such a major part of the picture, a far cry from 2021 where emissions were a key driver.
Three years of compression
As mentioned, we have come a long way from the extraordinary yields which kicked off DeFi summer, and even in the past few years we have seen declining yields, from the 2024 bull, to an unwind in 2025 capped by the 10/10 flash crash. In 2026 leverage is scarce, and the Treasury rate is predominant, while stETH ground down from 3.5% to ~2.2% as staked share of ETH climbed.
Points above 40% APY are dropped from the plot: sUSDe spent early 2024 above 50%, off this scale. Series start when DeFiLlama began tracking each pool.
What comes next?
This is just a snapshot of a dynamic situation. ETH issuance is the largest single source of yield, and far from being fixed, it has been a contentious topic over the past couple of weeks, with much discussion around a proposed rate-cut EIP-8363. And even as this memo was being written the price of ETH jumped 20% in a single day - lending markets are slower to react of course, but any prolonged change in sentiment and risk appetite will have an impact on rates and flows. Meanwhile Real World Assets and institutions are coming onto Ethereum in record numbers. So this might be where the yield comes from today, but the only thing that is certain is that things will change. Giddy up!
Method
Methodology & reference numbers
Source & scope. The data is a snapshot of DeFiLlama's free yields API, taken 2026-08-17 (yields.llama.fi/pools, /lendBorrow, /chart, plus api.llama.fi/protocols for categories). It covers 4,910 pools on Ethereum mainnet. vaults.fyi was evaluated as an alternative primary source. Its vault data is good and standardized, but its API needs a key. The memo therefore builds from keyless endpoints and uses vaults.fyi as an independent cross-check (see below).
Implied annual yield. Implied annual yield is pool size times current APY. It is a spot rate, not realized revenue. The totals here will not match DeFiLlama's headline "Ethereum TVL", for three reasons. Lending pools are sized by gross supply (totalSupplyUsd) rather than net liquidity, because supplier interest accrues on the full supplied balance. Only yield-bearing positions are counted. And the untokenized stake is added on top, which no TVL dashboard counts. QA checks the total against 30-day-mean APYs; they agree within 1% at the snapshot. For Morpho and Fluid, supplier interest also counts their wrapper pools (Morpho vaults, Fluid fTokens). The wrappers carry the supply-side APY that the market rows do not. Without them, the ~$14M/yr earned by Fluid's fToken holders would look like protocol take. Untokenized staking is the one line item added from outside the pools dataset: total staked ETH minus the LST/LRT share, valued at the LST-implied gross rate. It is an estimate and is marked as one wherever it appears.
Outliers. DeFiLlama's outlier flag removes ~520 pools. They hold only $0.3B of TVL, but their annualized spot APYs would add ~$167M of phantom yield. One near-empty Curve pool "yields" 177,000%. Most of these are DEX pools, so DEX fee yield is the least reliable number here. Treat $46M/yr as a floor, and note that DEX LP returns are quoted before impermanent loss. One lendBorrow row is dropped separately as internally inconsistent: fira's UZR pool claims $480M supplied and $435M borrowed at a zero borrow rate, against a $9M pool. That would be supplier interest with no payer. The pool is counted at its $9M TVL and excluded from the lending rollup.
Source taxonomy. Each pool category maps to an ultimate payer. Liquid staking and restaking map to ETH issuance. Lending and CDP supply map to onchain borrower interest. The source chart shows this line borrow-side: the $537M that borrowers pay, minus Sky's $50M of stability fees, which the chart counts under Sky. That matches the "Lending markets" node in the flow diagram. Suppliers receive $424M: $330M on the market rows, $94M via the Morpho vault and Fluid fToken wrappers. The lending-category bucket is $465M; the extra $41M is tokens with no borrow market (stUSDS $11M, sGHO $7M, 3Jane $5M, sDAI $3M). RWA maps to offchain interest. Maple is kept separate as private credit. Sky's sUSDS and Spark savings form a mixed bucket: Sky's onchain stability fees of ~$50M/yr cannot fund $188M/yr of sUSDS payouts, and the balance is RWA and agent-deployment revenue per Sky's June 2026 financial update. Repackaging layers (Pendle, Convex, yield aggregators, capital allocators: $3.2B, ~$83M/yr) are excluded from the source chart to limit double counting. Rewards are classified by reward-token address: known stablecoins count as issuer incentives, everything else as governance emissions. Rewards on pools whose deposit asset is itself a governance token (SKY, cvxCRV, sdCRV) are attributed to governance stakers, not depositors. The incentive count covers only pool-level reward APYs. Emissions distributed outside pools (EIGEN claim seasons, points programs, retroactive airdrops) are not counted. At 2026 token prices that is a small undercount, perhaps $10-20M/yr with EIGEN at ~$0.19. In 2024-25, when EIGEN traded ~25x higher, the same omission would have been material. Borrowed assets are attributed using DeFiLlama's mintedCoin field. Morpho and CDP markets are keyed by their collateral symbol, and the loan asset lives in mintedCoin: a cbBTC market with mintedCoin USDC is dollars borrowed against bitcoin, not bitcoin borrowed. Without this correction, ~$2.6B of dollar borrowing would be misattributed to BTC and staked ETH.
Onchain TVL corrections. Cross-checking against vaults.fyi surfaced four pools where DeFiLlama's TVL measures the wrong balance. Each was confirmed by reading the contracts directly: totalSupply or totalAssets times the exchange rate, via a public RPC. Rocket Pool's rETH is a $0.71B token float, not the $2.58B DeFiLlama reports. That larger figure is the whole protocol's stake, and most of it belongs to node operators, who count as direct stakers rather than LST holders. Coinbase's cbETH is $0.85B, not $0.36B. Maple's two syrup vaults hold $1.05B (USDC) and $0.42B (USDT), roughly 40% of DeFiLlama's figures. The pipeline overrides these four pools with the onchain values (fetch_onchain.py, keyless) and verifies that WBETH and weETH agree onchain to under 1%. The corrections cut the private-credit bucket roughly in half and move ~$1.4B of staking capital from the tokenized column to the direct column. Maple's smaller syrupUSDG ($0.26B per DeFiLlama) has no independent source and is left as reported.
Known double counting. Capital overlaps across buckets: wstETH supplied to Aave appears in both the staking and lending sizes. Yield dollars mostly do not overlap, because lending pools' apyBase is interest only and staking yield accrues inside the LST. Recursive backing (Sky deploys into Spark, which lends on Morpho; Ethena parks reserves in lending markets) leaves genuine ambiguity in the ultimate-payer split, worth a couple hundred million dollars either way. Two overlaps do reach the headline. The totals keep the repackaging layer's $83M/yr and $3.2B of size, which the source chart excludes. And Morpho vault deposits appear in both the vault and market sizes. Read the universe size as an upper bound.
Cross-checks. Borrowers pay $537M and suppliers receive $424M, a 21% spread, consistent with protocol reserve factors. Aave's scale matches its own protocol API (~$20.5B supplied on mainnet). stETH's APR matches Lido's official API (2.17% 7-day average). Benchmark rates and TVLs were verified against the vaults.fyi API (969 mainnet vaults). Rates agree within ~0.1pp on sUSDS (3.52% vs 3.52%), syrupUSDC (4.82% vs 4.81%), stETH (2.22% vs 2.20%) and the Aave and Compound stables. The largest gap is Aave v3 USDC at 3.28% vs 3.56%, which is spot versus 7-day-average methodology. After the onchain corrections, TVLs match to ~1% across the benchmark set; sUSDS and sUSDe match to the third digit. vaults.fyi's reward attribution on the Sentora RLUSD vault (+3.57pp paid in RLUSD) matches this memo's issuer-incentive split. Their Aave aggregate runs ~10% higher ($23.3B, including non-lending entries); ours matches Aave's API. The staking split uses ultrasound.money's effective-balance sum and DeFiLlama's ETH price, cross-checked against beaconcha.in's epoch totals to under 0.01%. Macro anchors come from FRED (DGS3MO 3.87%, fed funds 3.50-3.75%). Sky, Maple and Ethena figures were reconciled against their own published reporting. An 81-check QA suite gates the snapshot. The pipeline is in pipelines/yield-sources/.
The numbers
| Total universe / implied yield | $144.0B → $3.38B/yr (~2.4% blended) |
| …of which onchain yield-bearing positions | $96.5B → $2.20B/yr (2.28%) · plus untokenized stake $47.5B → est. $1.18B/yr |
| Stablecoin vs other (onchain universe) | $32.1B at 3.66% vs $64.4B at 1.59% |
| ETH staking | 42.2M ETH staked (~35% of supply, ~899K validators), ~$1.91B/yr total · tokenized: 17.3M ETH / $33.1B / ~$737M/yr (stETH 2.17% net) · untokenized: 24.9M ETH earning est. ~$1.18B/yr at ~2.5% gross |
| Lending flow | $35.5B supplied, $15.2B borrowed (43% util) · borrowers pay $537M/yr, suppliers get $424M, spread $113M |
| What's borrowed | stables $10.3B @ 4.2% ($435M/yr, 81% of interest) · ETH $4.5B @ 2.1% · BTC ~$0.1B (BTC is collateral, not borrowed) |
| Real-world interest | RWA funds $6.7B, ~$286M/yr · plus Sky savings $188M/yr (mixed) · Maple private credit $82M/yr (onchain-corrected) |
| Perp funding (Ethena et al.) | $1.74B staked, ~$72M/yr · sUSDe 4.45%, down from >50% in early 2024 |
| Token incentives | $227M/yr (~6.7%): $185M governance emissions + $42M stablecoin-issuer incentives (PYUSD ~$21M, RLUSD ~$16M) · ~$72M of the total goes to governance-token stakers, not depositors |
| BTC on Ethereum | $0.58B of Babylon-staked BTC earns ~0.3% (~$2M/yr); $4B+ of wrapped BTC earns ~0% as collateral |
| Macro anchor | 3M T-bill 3.87% · fed funds 3.50–3.75% (FRED, 2026-08-17) |