On-chain credit: who captures the margin, tranche by tranche

For every dollar of interest paid by an on-chain borrower, the portion retained by the protocol ranges from 0% for Morpho Blue to 23.3% for Venus; the rest is distributed among the depositors providing the capital and, in some designs, third parties managing the vaults. This journey — who pays, who earns for lending their capital, and who retains the margin at each stage — is what this analysis reconstructs across thirteen protocols, using complete daily series from DeFiLlama and a common window from July 9 to August 7. In terms of total fees, Aave V3 leads with 28.56 million dollars over thirty days and a 13.1% retention rate. Morpho Blue follows with 20.94 million and zero retained, as its fee switch is turned off and the margin is captured by its curators. The third row in the category, World Liberty Financial, reports a take rate (the percentage of flow kept by the house) of 100%, a signal that its money originates from stablecoin reserves rather than borrowers. Tracking the margin stage by stage explains why protocol size and the money it generates for its owners are two different rankings, and what needs to be read in each one's adapter before comparing two rows.

Where does the money go in a lending market?

Think of a traditional savings bank. You deposit 1,000 euros and are paid 1% per year. The bank lends that money at 5%. The borrower pays 50 euros, the depositor receives 10, and the remaining 40 are the bank's gross margin, which covers salaries, the office, and, if any remains, the dividend. Three different amounts, three different recipients, and only one of them is "what the bank earns."

An on-chain lending market operates with the same anatomy but with one difference that changes everything: the three amounts do not go to the same entity. The 50 euros from the borrower are the fees: all the money entering the system. The 10 for the depositor go entirely to whoever provided the capital. And the margin is what the protocol retains—the revenue—which may end up in a treasury, in a token buyback, or, as we shall see, nowhere at all.

The ratio between the margin and the total flow is the take rate: what percentage the house keeps. The bank in the example has an implicit take rate of 80%; the on-chain markets we measured range between 0% and 23.3%. The limit of the analogy lies in who decides this distribution. In a savings bank, it is set by the board; in a protocol, it is set by a contract parameter that may be turned off, often for reasons unrelated to the business. The practical consequence is that the size of a protocol and the money it produces for its owners are two different rankings, and confusing them leads to valuing a token based on a flow it will never touch.

The Ranking as of August 10, 2026

Twelve lending markets with relevant activity, ordered by fees, plus Aave V4 at the end as a reference for its deployment. The thirty-day column is calculated over the common window from July 9 to August 7; the revenue column collects what the protocol retains according to its own adapter; TVL is the capital deposited as of August 10.

Each adapter cuts the money circuit at a different point, and each row inherits that cut. The Morpho adapter adds interest plus liquidation bonuses collected by third parties; the Venus one keeps 60% of the interest plus the treasury and risk fund portions; the Compound V3 one carries the entire fee series —547 days, all with value— but the revenue series only from May 26, 2026: 78 consecutive days until August 11 and zero in the four hundred and sixty-nine previous ones. Those zeros are not a measurement: until that day, the adapter carried a handwritten "dailyRevenue: 0" and a note stating that Compound did not bring interest to the treasury, and a commit on May 26 replaced it with the calculation of the actual margin. Its one-year take rate, at 0.9%, therefore divides seventy-eight days of margin by three hundred and sixty-five days of interest, and measures how far the adapter reaches. In those seventy-eight days, Compound retains 6.1% of what borrowers pay as reserves, and 7.2% in the common window of the table; both figures are upper bounds, because the code counts as zero, and not as negative, the markets that pay interest charged to those same reserves. And the World Liberty Financial one measures something else entirely.

Why the dashboard's thirty-day columns cannot be compared

This section was born from our own error. On August 9, we calculated that Morpho was running at 75.4% of Aave and noted that the ratio had been falling from 88.7% on the 6th. Upon repeating the extraction on the 10th, the number had risen to 77.6%. A ratio that rises when both numerators are flat is a sign that the problem lies in the denominator.

And it did. Every protocol on DeFiLlama has its own adapter, and adapters do not end the day at the same time. Since the thirty-day window is counted backward from today, laggards are credited with empty days: Aave V3 loses three days of nearly one million dollars each and appears 6.5% below what it actually collected.

The last column mixes two effects, which is why Morpho appears with a small deviation despite being up to date: in addition to the lag, the two windows start and end on different days. In updated protocols, this difference stays below 2%; in laggards, it skyrockets.

The case of Maple deserves its own warning, because its deviation does not stem from a lag. Maple lends at a fixed term, so its fees arrive in bursts: in the measured window, it has six days at zero and eight days above half a million dollars, with a peak of 1,293 million in a single day. Any thirty-day window that shifts by a couple of days can exclude a large collection and change its total by 16%. With burst-style series, the total for a monthly window depends on the maturity schedule almost as much as on the volume lent.

The correction is simple and is the one applied in the table above: recalculate everyone's fees over the same window, ending August 7 because it is the last day with data for all thirteen. Anyone comparing two protocols using a dashboard's thirty-day column without first checking when each adapter updated may be comparing 30 days against 27.

What portion of Morpho's fees comes from liquidations?

Morpho Blue has collected between $0.48 and $0.66 million per day throughout almost all of July and August. Three days fall outside this band: July 7 with $2.647 million, the 8th with $2.706 million, and the 12th with $3.620 million. Combined, they are worth $8.973 million compared to the $1.63 million they would have been worth at a normal pace, thus contributing $7.34 million in excess—and one of the three still falls within the window of this analysis.

The breakdown by chain identifies the phenomenon without ambiguity. Base and Hyperliquid remain flat during those three days. All the movement is on Ethereum, which goes from 0.232 million on July 6 to 2.392, 2.414, and 3.290 on the three indicated days, returning to 0.318 on the 13th. The adapter methodology states that Morpho fees are "borrowing interest plus bonuses earned by liquidators," and of those two components, only liquidations can multiply the series tenfold on a Tuesday and return it to its baseline on Wednesday: the interest on a 3,500 million dollar portfolio moves slowly.

That is as far as what we can prove goes, with one stated limitation: we have not been able to establish which specific position was liquidated using any public source. The rest of the data does, however, rule out a general market episode. Aave V3 spent those same three days between 0.879 and 0.906 million daily, without a single spike, and neither ETH nor BTC moved enough to drag down entire portfolios, with drops of 1.64% and 1.44% on July 8 and fluctuations below 1.1% on the 12th. An event confined to one protocol and one chain, without public forensic coverage, is all that the figures support.

With this, we can finally answer how close Morpho is to Aave, and the answer depends on what is being measured, so here are the four readings. The dashboard aggregate on August 10 gave 77,6%. Correcting for the adapter lag over the common window, 73,3%. Further excluding the liquidation day that falls within it, 62,6%. And over the last three weeks, from July 20 to August 7, where the Morpho series is flat and there are no anomalies, 56,1%: 0,544 million daily compared to 0,969. Between the first reading and the last, there are twenty-one percentage points, and neither is false: each cuts the series at a different point.

For those who wish to check for themselves, there is a useful date. On August 12, the last of the three anomalous days exits any thirty-day window. From then on, a comparison made on aligned daily series should place Morpho between 55% and 62% of Aave as long as both maintain their current pace. If a dashboard continues to show more than 70%, the difference will be in the adapter lag.

Where does the money for World Liberty Financial come from?

World Liberty Financial appears in DeFiLlama's Lending category with $10.91 million in the measured window, which would place it as the third-largest lending market in the world, ahead of Maple, SparkLend, Kamino, Fluid, Euler, and Compound. Its methodological sheet, accessible from the same page serving the figure, describes what is being counted: "all investment yields from the backing assets of USD1 (Fidelity money market funds)." Its declared chain is "Off Chain." And its take rate comes out to 100% because fees and revenue are literally the same number.

This money originates outside the borrower-depositor circuit: it consists of yields from the reserves backing a stablecoin, the same metric used to measure Tether or Circle's business, which is why its declared take rate is 100% —with no depositors to remunerate, the entire flow is margin—. It is a real figure and a legitimate activity, but it follows a different circuit. Added to this table, it contributes 117,55 million dollars per year to the category that no borrower ever paid, pushing protocols that actually charge interest three spots down. The mechanics of that reserve business, and how fragile it becomes when rates or assets move, are broken down in our analysis of Tether's second quarter. The regulatory context, which is where it is most often covered, can be found in our piece on its license application before the OCC. In this article, we only use it to size the flow that actually originates from borrowers.

Counting only the interest paid by borrowers, the three largest on-chain credit markets by fees are Aave V3, Morpho Blue, and Maple.

Who captures the margin in Morpho if the protocol retains zero?

Morpho Blue moves $222.47 million per year in fees and retains zero. Not in thirty days, not in a year, not in its entire history. We also checked the protocol's parent entity on the same dashboard, in case revenue was registered separately, and it also marks zero.

The explanation has two layers, both verifiable. First: the contract includes a fee switch, capped at 25% of borrower interest, which governance could activate but has not. A February 2025 proposal documented a reason unrelated to the business: the legal and tax work surrounding activation was not finished. Second: in Morpho, the margin that other protocols keep for themselves is taken by curators—third parties who configure and manage each vault and charge management and performance fees set by themselves. That money exists, but it does not pass through the protocol's revenue line because it does not belong to the protocol. We already covered the disabled switch in our analysis of Morpho following its unicorn valuation, and here we simply confirm that it remains the same four months later.

A detail that illustrates the extent of the matter: when the project's first specific commercial income arrived—a low seven-figure license fee paid by Berachain for the right to use Morpho Blue's code—governance voted to send it to the Morpho Association, a French non-profit entity, instead of the DAO treasury. The stated reason was again that the legal work for the DAO to receive fees directly was not finished.

It is worth reconciling this figure with the two we have already published, as they measure different pockets and are easily misread together. In April, we reported $151.9 million annually distributed between lenders and curators: the total flow, the first pocket, at its size then. In the May real revenue ranking, we reported about $30 million annually described as curator fees: the third pocket. And today's $222.47 million is again the first, twelve months later and with a larger protocol. Something similar happens with Aave: the approximately $160 million in May was annualized revenue before the Kelp DAO exploit, and the $112 million-plus from the last twelve months already includes the subsequent drop.

Which protocol generates the most flow for every dollar deposited?

Ordering by absolute fees rewards those with the most capital. The interesting question for someone evaluating a protocol is different: how much flow does every $1,000 deposited generate? There, the ranking flips.

Euler V2 monetizes its capital more than three times better than Aave V3, and Fluid more than double, with TVLs of $352 million and $670 million compared to Aave's $14.248 billion. The warning comes in the same breath as the data: a small, concentrated portfolio monetizes better precisely because it is small, as it can cherry-pick high-rate markets without placing billions in boring collateral. Efficiency measures portfolio quality, but also a lack of scale. Removing the liquidation day, Morpho's efficiency drops from 2.63 to 2.25, bringing it closer to Aave's rather than exceeding it by a third: it is the second cell moved by that single day.

At the opposite extreme is the most striking case in the table: JustLend V1, on Tron, has $3.360 billion deposited—the fourth-highest TVL on the list, above Maple and on par with SparkLend—and produces $1.82 million in thirty days, twelve times less per dollar than Euler. Massive capital, minimal collection. It is a reminder that TVL, the metric by which almost everyone compares lending markets, can be capital that is not working.

How much cash flow do Compound V3 and Aave V4 generate?

Compound V3 collects $1.77 million in thirty days, trailing Kamino, Fluid, Euler, and Jupiter Lend—a market that launched this year and whose entire history, $29.68 million, fits comfortably into a single month of Aave. Compound still appears on any list of "lending giants" because of what it once was, and its $1.155 billion TVL sustains that image, but its current flow is that of a mid-sized protocol. A ranking that orders by historical cumulative data describes the past.

Aave V4 deserves a different reading, with its phase noted alongside it for fairness. It has been on Ethereum since March 30, 2026, and on Avalanche since July 15, with a deliberately reduced set of initial markets and conservative caps that the DAO is gradually expanding. As of August 10, it has gathered $224 million, 1.6% of V3's TVL, and $0.24 million in fees. It is already retaining margin: $0.04 million in the window, a take rate of 15.1% that exceeds V3's. The plan we analyzed in our Aave V4 architecture breakdown finds its four-month confirmation here: it is going where it said it would, and it is going slowly on purpose.

What to watch to see if this order changes

Four signals, all with a specific source and none dependent on opinion.

• The Morpho-Aave ratio starting August 12. This must be done on aligned daily series, not on the dashboard aggregate. If it does not fall within the 55% to 62% range, the baseline has changed and warrants a review.
• Morpho's fee switch. The cap is 25% of borrower interest, which currently goes entirely to the supply side. On an annual flow of $222.47 million, activating it to the maximum would divert up to about $55 million to the protocol, turning the zero into the second-largest revenue line in the sector, behind only Aave. The stated blocker is legal and fiscal, so the signal is a formal governance proposal, not a product announcement.
• The Aave V4 rollout. Two chains and conservative caps, but with margin already retained from the first month. Every cap expansion and every new network moves the only row with a programmed scale change ahead.
• The sustainability of flow for the efficient protocols. Maintaining 6.31 per $1,000 with ten times more capital would be the data point confirming that Euler and Fluid's efficiency is due to their model rather than their size.

The operational conclusion can be summarized in two steps. Before comparing two lending protocols, open the methodological sheet for each and check at which point in the circuit its adapter cuts the flow; then, look at when it was last updated. In this thirteen-row table, the former separates borrower interest from reserve yield and turns Morpho's zero into what it truly is—a pending governance decision; the latter recovers a 6.5% flow for Aave that the three-day lag had left outside the window. For the distribution seen from the other side—the depositor's—there is our comparison of DeFi yield strategies, which measures the APY received while this piece measures where it originates. And to understand how each one works internally, see the comparison between Aave, Compound, and Morpho.

Sources and links: DeFiLlama — Public Fees API, Lending category (query from Aug-10-2026) · DeFiLlama — Morpho Blue sheet and methodology · DeFiLlama — Aave V3 sheet and methodology · DeFiLlama — World Liberty Financial sheet and methodology · DeFiLlama — Maple daily series · DeFiLlama — SparkLend vs Spark aggregate · DeFiLlama — JustLend V1 sheet · Morpho Docs — Vault curator fees · Morpho Docs — Governance organization · The Token Dispatch — The fee switch and Berachain license (Apr-30-2026) · Aave — V4 live on Ethereum · The Block — Aave V4 launch with hub-and-spoke architecture · CryptoRank — Aave V4 deployment on Avalanche (Jul-15-2026) · DeFiLlama Coins — ETH and BTC daily price series · DefiLlama dimension-adapters — the May 26, 2026 commit that began counting Compound V3's margin