Why it matters
A protocol holding billions of dollars of user deposits that can change its own fee schedule, collateral parameters and treasury allocation by token vote is governed by whoever holds the tokens. That is not a criticism; it is the design. The question worth asking is arithmetic: how many addresses does it take to pass a proposal, and how many to stop one.
The answer determines what a governance token is. If a proposal needs the assent of thousands of independent holders, the token is a political instrument and the protocol has a constituency. If a handful of delegates can reach quorum between them, the token is a call option on control, and the published process ratifies decisions taken elsewhere. Depositors and integrators taking a protocol's parameters as given are exposed either way, and are not counted in the vote.
How it works
The mechanics are near-identical across major Ethereum protocols because most inherited the same contracts. A holder votes directly or delegates voting power to an address. A proposal needs a minimum balance of delegated power to be submitted, then a voting period, a quorum in favour before it can be queued, and a timelock before execution.
Uniswap's process is representative and unusually well documented. A proposal needs 1m UNI of delegated power to be submitted onchain and 40m UNI voting in favour to reach quorum, with an off-chain temperature check at 10m UNI beforehand and a two-day timelock before execution. Against a supply of one billion, quorum is 4% and the proposal threshold 0.1%.
Concentration enters through delegation rather than holdings. Tokens distributed to tens of thousands of addresses are delegated to a much smaller number of delegates, and it is delegated power, not ownership, that votes. A 2025 study of five large protocols by Messias and Ide found Gini coefficients of voting power above 0.94 in every one, with a single delegate holding a majority in one and the top handful sufficient to meet quorum in the others. Most holders never delegate, so the effective electorate is a fraction of the nominal one.
The result is visible in tallies rather than theory. Uniswap's proposal 93, the UNIfication package that ended the fee switch debate by directing fees to burn UNI, executed on 28 December 2025 with 125,342,017 votes in favour and 742 against, not a contested election but the recorded output of a decision already taken.
Economic mechanism
Follow the money. A governance token confers no dividend by default, so its value rests on two things: the expectation that governance will route protocol revenue to holders, and the option value of controlling parameters other people's money depends on. Both make voting power worth buying, and both are exercised by the marginal decisive holder rather than the average one.
Acquiring enough delegated power to be decisive costs a fraction of the value under management, because a token's market capitalisation reflects a discounted stream of uncertain revenue while the parameters it controls govern deposits many times larger. The gap between the cost of control and the value controlled is the structural vulnerability, and exploiting it requires no bug.
The Uniswap and Aave outcomes show the benign version. UNIfication ended the argument about whether fees should accrue to holders by burning 100m UNI from the treasury and routing protocol fees to burn, with 20m UNI a year to Uniswap Labs. Aave's April 2026 vote gave the DAO control of revenue from Aave-branded products, against protocol revenue CoinDesk reported at around $140m for 2025. Concentrated voting power made decisions a dispersed electorate would have taken years to reach; concentration is not only a risk, it is what makes onchain governance capable of acting.
The costs land elsewhere. Delegates are unpaid or paid by grant, and bear no liability for a parameter change that liquidates borrowers. Depositors bear the loss and hold no vote unless they also hold tokens. Risk and treasury managers are paid by the treasury they advise, a conflict every DAO acknowledges and none has resolved. And an author with enough power can put text onchain the community has not agreed: in Aave's December 2025 dispute over routing swap fees, roughly 1.8m AAVE of voting power was expressed on a proposal whose stated author disowned it.
Where it breaks hardest is the treasury. Compound's proposal 289 in July 2024 sought to move 499,000 COMP into a yield-bearing product controlled by the proposers, and passed with 682,190 votes against 633,640 on a 400,000 quorum before being withdrawn under pressure. Nothing in the code broke: a coordinated bloc met the threshold, and the only remedy was social.
Participants
Token holders supply the nominal electorate and mostly abstain. Delegates (venture funds, delegate firms, university blockchain groups and individuals) hold the power that votes. Core development companies hold large allocations and depend on grants voted by the same process. Auditors and risk managers are paid from the treasury they advise. Governance platforms host the interfaces and vote records. Foundations sit between token and developers, sometimes with emergency powers outside the vote.
Examples
Uniswap's UNIfication proposal executed on 28 December 2025 with 125,342,017 UNI in favour and 742 against, burning 100m UNI and directing protocol fees to burn. It is among the largest value decisions taken by an onchain vote, and among the least contested.
Compound's proposal 289 of July 2024 briefly passed a transfer of 499,000 COMP from the treasury to a vehicle controlled by the proposing group, on a margin of about 48,000 votes, and was reversed only after public pressure, not by any protocol mechanism.
Aave's revenue dispute ran from December 2025, when a proposal to route order flow fees to an Aave Labs entity was disowned by its named author, to April 2026, when a vote directed revenue from Aave-branded products to the DAO.
Risks and limitations
The unsettled question is whether concentration is a defect or a stage. Several large delegates and most protocol development companies argue that concentrated delegation is how any shareholder system works, that professional delegates are better informed than a dispersed retail base, and that the alternative is paralysis. Much of the academic work and a number of vocal holders argue that a system marketed as decentralised while a handful of addresses can meet quorum is misdescribed, and that the misdescription has regulatory consequences. Both sides are describing the same tallies.
Measurement is hard, and published concentration figures should be read with care. Gini and Nakamoto coefficients computed over addresses understate concentration when one entity uses several addresses and overstate it when a custodian's single address holds many clients' tokens. Delegated power moves between snapshots, and exchange and bridge addresses distort every distribution. A figure quoted without a snapshot date is not comparable to another.
The regulatory exposure is rarely priced. If a small identifiable group directs a protocol's economics, the argument that no one controls it becomes hard to sustain, and questions about the token's status and the duties of the decisive holders follow. No supervisor has tested this in a case with a well-documented vote record, and the records are public.
Finally, low participation is not straightforwardly bad. Rational abstention by small holders is efficient, and forcing turnout through incentives produces uninformed voting and vote-buying markets. Several protocols have concluded that paying delegates to be informed beats paying holders to vote. That trade-off is usually made implicitly.
Key metrics
No series carried here measures governance concentration, which reflects the state of published data. The measures that would matter are the Nakamoto coefficient of delegated voting power against each protocol's own quorum, the share of circulating supply ever delegated, turnout as a share of delegated power, the count of proposals whose outcome would change if the largest delegate abstained, and the cost of decisive voting power against the value of assets it controls. Each is computable from public data, and each needs a snapshot date and a stated treatment of custodial addresses.
Related research
The Messias and Ide paper is the most useful recent quantitative treatment, because it computes concentration consistently across several protocols rather than one. For mechanics, a protocol's own documentation beats any summary: Uniswap's process page states its thresholds plainly, and vote records on governance portals give tallies without interpretation. The Compound proposal 289 record and the Aave dispute of December 2025 to April 2026 are the case studies worth reading in full, because in both the formal process worked as specified and the outcome was still contested.