A fee switch is a governance-controlled parameter that takes a slice of the fees a protocol already charges its users and routes it to the protocol itself rather than to the people supplying the service. It is the mirror image of liquidity mining: where liquidity mining pays tokens out to rent capital, a fee switch brings revenue in, and the governance question shifts from how much to emit to who receives the income and in what form. Because the fee comes out of someone's existing revenue – usually liquidity providers – turning one on is a redistribution inside the protocol's own economy, not free money, which is why the largest DAOs argued about it for years before flipping the switch.
Most DeFi protocols charge a fee at the point of use: a swap fee on an automated market maker, an interest spread on a lending market, a redemption fee on a stablecoin. In the standard design that fee is paid entirely to the counterparties who take the risk – liquidity providers, depositors, stakers. A fee switch inserts a second recipient: the contract keeps a configurable fraction and sends it somewhere the DAO controls.
The switch itself is usually trivial in code – a factory owner setting a fee numerator – and hard in governance, for three reasons that recur in every fee-switch debate:
Uniswap's governance record is the clearest primary-source account of how long the question stayed open. The Snapshot space shows a fee-switch vote as early as 25 July 2021 (“Turning on the fee switch for V2”, 354 voters), a July 2022 temperature check for ETH/USDC and USDC/USDT pools, a consensus check for a three-pair pilot that closed 9 August 2022 with 2,373 voters, a three-part “Making Protocol Fees Operational” vote on fee options, deployment chain and treasury asset that closed 1 June 2023, and a “V3 Fees: Factory Owner Amendment” vote that closed 9 March 2024 with 50.2M UNI of weight behind it.
None of them resulted in a fee being charged in production. The pattern – repeated large-turnout votes in favour of the principle, followed by no on-chain activation – is itself the lesson: a fee switch is not blocked by the vote count but by the legal, competitive and operational questions that sit underneath it. Uniswap only began charging protocol fees on Ethereum mainnet in late December 2025.
There was also a case for never flipping it, and it is worth naming because the four-year delay is exactly what it predicted. Jacob Horne's hyperstructures essay argued that an unexercised fee switch is the more valuable asset – ownership as “the threat of the fee” – on the reasoning that actually charging it “would immediately lead to an incentivized fork.” UNIfication is the largest test that argument has had, and what the record below shows is holders choosing realised revenue over the threat.
The change arrived as a single bundled proposal, UNIfication, which passed with 125,342,017 UNI for and 742 against and executed on 28 December 2025. It is worth reading as a template, because it shows how much rides alongside the switch:
The mechanism Uniswap shipped is more interesting than “fees buy tokens.” Fees accumulate on each chain in a TokenJar, and searchers claim those accumulated fees by paying UNI, which is bridged back to Ethereum mainnet and sent to the burn address – on Arbitrum Orbit chains through a releaser contract named ArbitrumOrbitResourceFirepit. The protocol never has to sell the fee assets or hold an inventory: it auctions them to searchers and denominates the price in the token it wants destroyed, which is why the burn works identically on any chain the DEX deploys to.
By the July 2026 Robinhood Chain expansion vote, protocol fees had reached ten chains beyond mainnet – Arbitrum, Base, OP Mainnet, Worldchain, X Layer, Soneium, Zora, Celo, BNB Chain and Polygon – and the same proposal records a single-day record of 186,000 UNI burned in June 2026. Uniswap deployed to Robinhood Chain at that chain's 1 July 2026 mainnet debut and crossed $1bn of cumulative swap volume there within ten days.
UNIfication also changed the governance of the parameter, not just its value: fee-parameter proposals now skip the RFC stage and go straight to a five-day Snapshot followed by an on-chain vote. A fee switch that is expected to move often needs a fast path, and building one is part of the design. The v4 protocol fee activation ran through it in July 2026 and drew the first substantial dissent of the rollout – 19.3M UNI for against 1.0M explicitly opposed, from 132 voters.
Once a protocol collects revenue, four destinations dominate practice, and each one distributes benefit differently:
The choice is not primarily financial. Burning and distributing hand value to holders automatically; retaining hands it to whoever controls the proposal process. A DAO with weak participation that retains revenue is choosing to concentrate power, whatever the spreadsheet says. Aragon's Ownership Token Framework, launched in February 2026, scores tokens on this question by reading from their contracts who controls revenue routing (token valuation).
Aave took the buyback route and its governance forum documents, unusually clearly, what happens when revenue falls. The Aavenomics implementation gave a finance committee a mandate to buy AAVE on the secondary market at $1M per week. In March 2026 the DAO proposed cutting the programme from roughly $50M to $30M a year – about $577K a week, down from $962K – citing borrow fee revenue roughly 25% off its peak, with January 2026 at $7.95M against $13.5M in January 2025. The cut passed a Snapshot vote that closed on 15 March 2026, about 363,000 $AAVE for and 3,200 against.
Then the buybacks stopped altogether, and a security incident stopped them rather than revenue. On 18 April 2026 an exploit of Kelp's LayerZero bridge route for rsETH, Kelp's restaked-ETH token, put unbacked rsETH into Aave V3 markets on several chains. TokenLogic, the service provider running the programme, executed no buyback after 19 April, and its request for final comment formalising the pause passed on Snapshot on 1 May 2026, about 617,000 $AAVE for and 4,900 against. It keeps the pause in place until the rsETH position is clearer and names no date or condition for resuming.
A forum thread opened 18 May 2026 asked governance to restart them, with contributors noting that the pause landed while the token traded far below its highs – exactly when a confidence-signalling buyback is most wanted and least affordable. Its last reply, on 25 June, relayed Stani Kulechov's announcement that Aavenomics 3.0 would bring “immutable and automated buybacks of AAVE”, and the thread closed automatically a month later with no decision recorded in it. The direction Aave's founder set was to remove the discretion, replacing committee-executed purchases with an automated, non-discretionary engine. As of 15 September 2026 no proposal to build that engine had been posted to Aave's governance forum, and the August/September 2026 funding update, posted 1 September, provides for GHO runway, operations and a private-credit trial without mentioning buybacks.
That arc is the general lesson of discretionary value accrual: a buyback funded from revenue and executed by a committee is procyclical twice over – the revenue falls when the market falls, and the committee's appetite falls with it. Automation is not a yield improvement; it is a governance decision to remove a lever that gets pulled at the worst time.
Jupiter routes 50% of protocol fees into buying JUP for its Litterbox Trust, with the other half funding development, operations, incentives and ecosystem growth. Because that split is a governance parameter rather than a constant, it becomes a standing subject of debate: a proposal posted 30 May 2026 asks to raise the buyback allocation from 50% to 70% and to burn the purchased tokens outright instead of holding them in trust, which would cut the operating budget to 30%. As of the thread's last reply, on 25 July 2026, it had not been put to a formal vote.
This is the fee switch's second-order effect and the one DAO treasurers should plan for. Turning it on does not settle anything – it converts an operating budget into a percentage that holders can vote to reduce, and every subsequent bull market makes “burn more, spend less” an easy proposal to write and a hard one to argue against.
A caper has no fee switch to vote on, because its treasury takes a share of every trade from the first one. Every buy and every sell against the bonding curve pays the skim, at a rate that falls as the curve fills, and 49.5% of the skim is deposited into that caper's own treasury in XRD, in the same call that moves the tokens. The rest goes to the founder and to the $CAPER treasury.
The shares are platform-wide rather than per-caper, and they are not parameters anyone sets: they are constants compiled into the shared logic component, so moving them takes a fresh logic component and a move of the registry's current_main – the upgrade path, open to the protocol admin badge behind the registry's timelock or to a settled $CAPER UPGRADE proposal. No individual caper spends a governance cycle arguing about whether to charge one, because no individual caper can answer the question.
Nothing is burned and nothing is bought back. The skim share accumulates in the treasury alongside the proposal and vote fees, and holders reach it two ways: a proposal that spends it, or exit, which pays out a share of the treasury computed from the holder's canonical vote weight – a function of both their governance tokens and their vote tokens – and burns their vote tokens on the way out. That is a deliberately narrower design than the ones above: the accrual question is settled in code, and what remains for governance is the spending question.