The advertised percentage doesn’t come from nowhere
You open a protocol’s interface and see a large number next to the word “APY” or “yield”. The uncomfortable question isn’t whether that number is high or low, but who pays it. If you don’t know, you don’t know what risk you are taking on either.
In practice there are three main sources: the interest paid by people who borrow, the fees generated by people who use the protocol, and the new tokens the protocol itself issues as an incentive. Each has a different payer and therefore a different risk. On top of these come yields tied to real-world assets off-chain and synthetic trades built on derivatives.
CoinDesk Research, in its report State of the Blockchain 2025 (19 December 2025), proposes a similar classification: real or macro yield, synthetic crypto-native yield and lending-based crypto-native yield. We will go through all of them.
A note on scope: this article explains mechanisms and risks. It does not say where to deposit or what pays the most. If you need the groundwork first, there is a general explanation in our guide to what decentralised finance is and what risks it carries.
First source: interest paid by borrowers
How it works
In a decentralised lending market, some users deposit funds and others borrow them, leaving collateral worth more than the loan (overcollateralisation). Borrowers pay interest, and that interest, minus the share the protocol keeps, reaches the depositors.
Why the depositor earns less than the borrower pays
The Aave v4 documentation (governance forum, version of 31 January 2026) gives the formula:
deposit rate = borrow rate × utilisation × (1 − reserve factor)
The document itself offers an example: 3.67% × 33.3% × 90% = 1.10% APY. There are two reasons for the gap. First, not all deposited capital is lent out: utilisation measures what fraction is, and only that fraction generates interest. Second, the protocol keeps a share. In the example, the remaining 90% implies a reserve factor of 10%.
The two-slope model
Rates are not fixed. According to the same document, at 20% utilisation rates fall to encourage borrowing; at 50% they hold; at 80% they rise to attract deposits, and at 95% there are aggressive increases (for example, 60% APY or more) to protect solvency.
The example of the DAI market on Ethereum makes this concrete. Below 80% utilisation, the rate is the base rate plus (utilisation / 80%) × slope 1. With 40% utilisation and a slope 1 of 4%, the borrow rate is 2% a year. Above 80%, with a slope 2 of 75%, at 90% utilisation the example’s borrow rate reaches 41.5% a year.
The reading that matters: a very high rate in a lending market usually signals scarce liquidity, not generosity. It is the mechanism telling you that few funds remain available.
What is being paid for
What is being paid for is demand for leverage. And that demand follows market sentiment. According to State of DeFi 2025 (DefiLlama Research, 28 March 2026), revenues in the lending sector went from around USD 10 million a month in 2024 to between USD 15 million and USD 25 million a month in 2025, with volumes following market sentiment. Aave captured close to half of all fees in the sector.
Second source: swap fees
How it works
A decentralised exchange (DEX) such as Uniswap does not match buy and sell orders. It uses pools: reserves of tokens supplied by liquidity providers (LPs), against which trades are made. According to the official documentation (updated on 5 August 2026), each swap pays a fee that compensates the LPs and, in some cases, a share goes to the protocol.
Details of the mechanism
- Price range. In v3 and v4, only in-range liquidity (positions whose range includes the current price) earns fees. If the price moves out of range, the position stops accruing them until it returns.
- Who sets the fee. Uniswap v4 removes the fixed tiers: the pool creator can set any fee from 0% to 100%, and pools can use dynamic fees through hooks.
- The protocol’s share. After the “UNIfication” governance proposal was approved in December 2025, protocol fees are active on all v2 pools and on selected v3 pools. At launch they amount to roughly 1/6 of the swap fee. With a v3 tier of 0.30%, that is 0.25% for the LP and 0.05% for the protocol. These values are set by governance and may change.
What is being paid for
The activity of the people swapping. Without volume there is no yield, and a figure advertised today says nothing about tomorrow’s volume. In addition, impermanent loss (what an LP loses compared with simply holding the tokens when the price moves) is a usual cost of this model; the sources consulted do not quantify it, so I give no figures.
Third source: new tokens as an incentive
How it works
The protocol creates governance tokens and distributes them to those who supply liquidity or deposit funds. That is how a good part of the sector started: according to the report Evaluating Risk in DeFi (ExponentialFi, published on DefiLlama Research on 27 March 2026), in the early days yield was driven by the inflationary issuance of governance tokens. Protocols such as Compound, Synthetix and Yearn encouraged participation with newly created tokens that users farmed (farm) and sold. The report gives the “DeFi Summer” of 2020 as its example.
What is being paid for
It is paid for by diluting the other holders and by the market’s interest in that token. The report argues that this model was unsustainable because yield depended on token inflation rather than on real economic activity.
Hence “mercenary capital”: liquidity that hops between protocols to collect incentives, often with no concern for the protocol’s continuity. Put plainly, TVL (total value locked) inflated by incentives measures how long the incentive lasts.
What the data says
CoinDesk Research calculates that, once emissions and incentives are deducted, of the seven protocols with incentive data available on DefiLlama only three are profitable: Lido, Sky and Aave. The report doubts the sustainability of the others’ revenues, which appear to depend largely on subsidies to drive activity. This is a finding of the report on a small sample; it is not an assessment of any particular protocol.
The token narrative has also changed. According to DefiLlama Research (28 March 2026), for much of the previous cycle governance and emissions sustained the narrative, while revenues accumulated in treasuries or with intermediaries. In 2025 a growing share of fees was redirected to token holders. Derivatives, CDPs, DEXs and yield platforms now pass a significant share of their operating surplus to holders; lending, liquid staking and trading interfaces remain closer to the zero-payout model of 2020-2023.
There is an added risk: concentration. The same report points out that the treasuries of Uniswap, Cardano and the Optimism Foundation are almost 100% exposed to their own tokens, and Arbitrum’s is over 97% in ARB.
Other sources: real-world assets and synthetic trades
Not all yield is born on-chain. CoinDesk Research describes real or macro yield with the example of Sky (formerly MakerDAO), which earns interest from real-world assets off-chain, such as Treasury bills, and depends on global interest rates. Along the same lines, DefiLlama Research notes that tokenised private credit almost doubled, from USD 9,830 million to USD 18,710 million, and describes it as real capital formation, not a mere change in price.
Synthetic yield is a different matter. CoinDesk cites Ethena, which uses delta-neutral basis trades and perpetual funding rates. Put simply: you hold one position that is offset by an opposite one in derivatives, so the price of the asset stops mattering, and you collect the funding rate paid by those betting with leverage. The report describes it as highly dependent on market sentiment.
The counterparty changes and so does the risk, but the question does not: here what is being paid is traditional interest rates or appetite for leverage in derivatives.
Risks by source
- Lending: liquidity. With high utilisation, part of the funds is lent out and withdrawing may be difficult. The rate model exists precisely to push borrowers to repay or others to deposit more.
- Lending: default and liquidations. If the collateral falls in value, it is liquidated. The fact that the system depends on correct prices adds an oracle risk: the oracle is the mechanism that tells the contract the price of assets.
- Swap fees: volume and range. Without volume there are no fees, and a position that is out of range stops earning them.
- Swap fees: governance. Whoever sets the pool’s fee, or votes on the protocol’s share, changes what the LP receives.
- Token issuance: dilution and price dependence. The yield is worth whatever the token it is paid in is worth, and mercenary capital can leave as soon as the incentive drops.
- Real-world assets: counterparty and interest rates. You depend on whoever manages the assets off-chain and on global interest rates.
- Synthetics: sentiment and funding. If funding rates or market mood change, the figure changes.
- Cross-cutting: contract and governance. “Audited” is not the same as “secure”: an audit reduces risks, it does not eliminate them. When the risk depends on how the contract is written, code analysis is Dex’s territory.
One risk remains, in how the figures are read. The ExponentialFi report warns that many investors evaluate opportunities solely by the advertised yield and that, without a transparent breakdown of risk, it is common to confuse short-term returns or protocol rewards with sustainable yield.
What to look at before trusting a number
Questions, not recommendations:
- Who pays this yield: borrowers, traders or token issuance?
- How much of it arrives in the protocol’s own token?
- What happens to the figure if volume or utilisation falls?
- Who can change the parameters: fees, reserve factor, rates?
- Does the on-chain data confirm what the interface says?
Beware of definitions of “fees”
On DefiLlama the word does not mean the same thing across protocols. In Aave V3 it includes borrowing interest, flash loan fees, liquidation fees and penalties paid by borrowers; in V4, interest plus the protocol’s share of liquidation bonuses. In Morpho Blue it adds up borrowers’ interest and the liquidation bonuses paid to liquidators. In Moonwell it is only borrowers’ interest. A fees figure is not comparable until you have read its definition.
TVL is not income
TVL measures deposited capital, not profits. On 28 October 2026, the DefiLlama homepage showed Lido’s TVL at USD 21,346 million, Aave’s at USD 15,698 million and Morpho’s at USD 10,706 million. For Aave it also showed negative changes of −9.61%, −12.32% and −17.62%; the source consulted does not say which period each one covers, so I do not interpret them. They serve as a reminder that TVL changes, and that none of these figures says how much a depositor earns.
The data is in plain view for anyone who wants to look at it. If you want to learn how to do so with the tools, Luk covers the practical side; for the groundwork, there is the guide to blockchain and cryptocurrencies for starting from scratch and the one on Web3 in practice (in Spanish).
Before looking at the percentage
Every yield figure has a payer. The job is to identify them before looking at the percentage.
Sources
- Evaluating Risk in DeFi (ExponentialFi report, DefiLlama Research, 27 March 2026)
- Aave v4: Interest Rates – General (Aave governance forum, version of 31 January 2026)
- Fees | Uniswap Developers (official documentation, updated 5 August 2026)
- State of the Blockchain 2025 (CoinDesk Research, 19 December 2025)
- State of DeFi 2025 (DefiLlama Research, 28 March 2026)
- Base Fees Rankings – User Fees by Protocol (DefiLlama)
- DefiLlama – DeFi Dashboard & Crypto Analytics (homepage, 28 October 2026)
Image: PublicDomainPictures on Pixabay
Translated and adapted by Ollie.