Perp DEX Fees: Auditing the $125 Million Number Nobody Can Source

Somewhere in the last cycle, a number started circulating: Perp DEXs generate $125 million a month in fees. It appeared in a research note. It got quoted. It got screenshotted. It became a premise. By the time it reached me it had the texture of settled fact, which is exactly the texture I have learned to distrust.

Here is what the number does not carry. No protocol name. No data provider. No time window. No definition of gross versus net. No treatment of maker rebates. No statement of whether token incentives are netted out. Four missing fields and a currency symbol — that is not a data point. That is a rumor with a decimal.

I have spent the better part of a decade auditing claims that people badly wanted to be true. In 2018 I found an integer overflow in a staking contract weeks before a mainnet launch; the team's first reply was not "thank you," it was "can we ship anyway." The lesson calcified then and has never softened: a number without provenance is a narrative wearing a lab coat. This piece is the audit that should have accompanied the $125 million claim and could not, because the claim arrived naked.

Tracing the fault lines where code meets capital — that is where the sector's real story lives. Not in the headline number. In the plumbing beneath it.

The Sector That Learned to Stop Asking Permission

Perpetual futures are the most successful product DeFi ever shipped, and the least discussed in proportion to their importance. Spot DEXs get the ideological oxygen. Perps get the volume. A perpetual contract has no expiry, so it never has to roll, so it never has to explain itself. It is a leverage instrument that runs forever, settles on-chain, and asks the trader to fund the position through a periodic rate rather than a dated premium. It is elegant, and it is dangerous, and it is the reason decentralized venues now handle a meaningful fraction of global crypto derivatives flow.

The technical lineage splits three ways, and the split is not cosmetic — it determines who gets paid before a single dollar of fee moves.

The order-book camp runs an on-chain central limit order book, increasingly on a purpose-built chain with its own sequencer. Matching is fast, latency approaches centralized exchange territory, and the economics resemble a real venue: makers are rebated for quoting, takers pay, and the venue sells order flow priority to whoever provides depth.

The vault camp removes the market maker entirely. Traders take the other side of a liquidity pool. The pool is the counterparty. This is a fundamentally different animal: the trader is not paying a venue, the trader is buying insurance from depositors, and the depositors eat the tail risk when the market gaps.

The vAMM and aggregator camp virtualizes the curve and routes flow outward, taking a spread for the service of finding someone else's liquidity.

Each architecture routes fee revenue to a different beneficiary, and none of them route it to the token holder by default. That is the axis this entire analysis turns on. Everything else — throughput, TPS claims, sub-second matching — is engineering theater until you answer the only question that matters: when the fee lands, whose pocket is it in?

I have watched this sector mature through three narrative cycles. The 2021 perp wars were about who could list the most exotic collateral. The 2022 collapse was about who survived the liquidation cascade when the exotic collateral turned out to be correlated. The 2024–2025 points era was about who could manufacture volume with the cheapest emissions. Each cycle the technology improved and each cycle the fee distribution stayed stubbornly concentrated in the same three hands. The headline number changed. The plumbing did not.

Cross-Validating the Number Before Believing It

Start with the only hard figure in the source material and stress it until it either holds or breaks.

Perp DEX blended taker fees typically run between 2.5 and 5.5 basis points, with makers on negative fees — rebated — for providing quotes. Take a blended realized rate of 3 to 5 bps and reverse-engineer the volume that $125 million a month implies.

At 3 bps, you need roughly $417 billion in monthly volume, or about $13.9 billion a day. At 4 bps, $312 billion monthly, $10.4 billion daily. At 5 bps, $250 billion monthly, $8.3 billion daily.

So the number implies a sector turning over somewhere between $80 and $140 billion a day. That is the right order of magnitude for the perp DEX sector in 2025. The figure is not absurd. It survives the first pass.

But surviving the first pass is not the same as being trustworthy, and the reverse-engineering exposes something the original claim never said out loud.

If the whole sector clears $80 to $140 billion a day, and one dominant venue has historically absorbed 60 to 80 percent of that flow, then "Perp DEXs generate $125 million a month" is not a statement about a sector. It is, semantically, a statement about one protocol. The plural is doing load-bearing work it cannot support. A sector total and a single-venue total have converged so tightly that the distinction has stopped being meaningful — and that convergence is the most important structural fact in the entire story.

This is what I call the trench-coat problem. A category with one real occupant and a long tail of subsidized pretenders reads, in aggregate data, like a thriving market. In reality it is a monopoly plus a subsidy program. The tail exists to make the leader look like an ecosystem and to keep the leader's fee take from looking like what it is: rent.

And rent is exactly the right word, because the moat is not the code. Anyone can fork an order book. The moat is liquidity network effects — the depth that makes a venue worth trading on, which attracts more depth. That flywheel is real and it is sticky for the leader. It is not sticky at all for the tail, and it is not sticky for the market makers, who are the actual swing voters in this economy.

The Fee Flow Map Nobody Draws

The original claim's deepest flaw is a category error. It equates fees with revenue, and revenue with value accruing to token holders. Both equations are wrong, and the second one is wrong by an order of magnitude.

Follow a single dollar of taker fee through the machine.

First, a chunk goes back out the door as maker rebate — negative fees paid to professional market makers and high-frequency firms for the obligation to quote. This is not generosity. It is the cost of buying a quote book, and it is non-negotiable, because market makers are nomadic. They migrate between venues by changing an API endpoint. There is no switching cost, no lock-in, no loyalty. Which means the rebate is a permanent structural tax on gross fees, not a growth-stage subsidy that fades with scale. Scale does not fix this. Scale makes it bigger.

Second, a referral and builder-code layer skims 10 to 30 percent of fees. This goes to front-ends, wallets, aggregators, trading bots, and the KOLs who funnel retail flow. This is customer acquisition cost wearing a different hat — a channel tax. It is also where the intent-based routing conversation becomes relevant, and I want to be precise about that. Intent architectures do not remove extractive intermediaries; they relocate them. The auction moves off-chain into a solver network, and the solver's privileged position over order flow becomes the new rentier. The MEV did not disappear. It changed employers. Anyone who tells you a solver network democratizes execution has not looked at who runs the solvers.

Third, the insurance fund and the vault pool absorb their cut. In vault-style venues this is the largest single destination, because the depositor is the counterparty and is being paid for tail risk. Note what that means: in a bad week, the depositor's fee income is dwarfed by their directional loss. Fee revenue and LP profit are not the same line item. A venue can post record fees and record LP drawdowns in the same month.

Fourth, validators and sequencers get theirs. On a general-purpose chain this leaves the protocol as gas and blob fees. On a self-built appchain it is internalized as validator income — which is precisely why every serious perp venue eventually builds its own chain. It is not about performance vanity. It is about recapturing the upstream cost.

Fifth, the treasury. And only from the treasury, and only if the protocol has explicitly wired the mechanism, does anything reach the token holder — as buyback-and-burn or as a fee dividend to stakers.

Sixth, and in the opposite direction, the incentive program. Points, trading rewards, market-maker incentive tokens. This is money leaving the building. It is not a fee destination at all; it is a subsidy that runs against fee revenue and, in most venues, at a comparable magnitude.

Now rank the beneficiaries by realized capture.

Market makers and channel partners first. Volume farmers and airdrop hunters second. Liquidity depositors third. Token holders fourth — and only when a buyback or dividend exists. Ordinary traders last, as the net payer, the one whose blood funds the entire circuit.

That ranking is the answer to the question the original article asked but never answered. It is also a structural judgment that holds without knowing which protocol we are talking about, which is the only kind of judgment I trust when the source material refuses to name names.

The Fee-Emission Loop

Here is where the analysis gets uncomfortable, and where I want to be careful, because the mechanism I am about to describe is not fraud. It is worse in a way — it is a legitimate accounting structure that produces illegitimate conclusions.

Many perp venues run a trading incentive program alongside their fee engine. The protocol issues tokens to reward volume. Traders chase the rewards, generating volume, generating fees. The fees are then reported as revenue.

Stop and look at the direction of the arrows. The protocol is paying out tokens on the left hand and collecting fees on the right. If the volume that generates the fees is itself manufactured by the token payout, then the "revenue" is a reflux of the protocol's own emissions. It is not an external value input. It is a round trip.

I call this the fee-emission loop, and it has three diagnostic signatures. One: volume collapses when incentives stop. Two: a large share of fee-paying volume is self-trading or wash-adjacent, generated by the same wallets that farm the rewards. Three: the protocol is paying "yield" in newly minted tokens whose source is not external demand.

When signatures one and two both hold, you are not looking at a business. You are looking at a closed circuit with a marketing department.

I built my first serious bear framework on exactly this logic. In early 2022 I watched Anchor Protocol advertise a stable yield that could only be sustained by a reserve being drained in real time, and I shorted the structure weeks before the crowd found the exit. The tell was never the yield number. The tell was the direction of the subsidy — the protocol was paying out more than its external demand could justify, and calling the difference "growth." Every perp venue running a points program is asking you to make the same judgment about the same arithmetic.

Which brings me to value capture, and the three-tier model I use to grade any fee-generating protocol.

Strong: one hundred percent of fees route to market buyback or an equivalent distribution. This is the only structure that hard-binds the token to the cash flow.

Medium: a portion of fees is distributed to stakers. Better than nothing, but staking dilution usually eats the dividend.

Weak: fees accrue to the LP pool and the treasury's operating budget, and the token confers governance only.

Most of this sector sits in the weak tier and advertises itself as strong. Governance rights over a fee stream you do not receive are not ownership. They are a suggestion box with a market cap.

And governance is not a neutral arbiter here — it is the institutional cause of the misallocation. Who decides where fees go? The people who hold the governance token. Who holds the governance token? Teams and venture funds, with top-ten concentration routinely above fifty percent and voter participation routinely below five. Fee allocation proposals are almost never brought to a vote, because they would touch market-maker and channel economics, and those are the constituencies the governance coalition actually depends on. So the allocation does not get voted on. It gets inherited.

The most dangerous version of this: a governance token concentrated in venture hands, a protocol that announces a fee-funded buyback, and a vesting cliff approaching. The buyback is not value sharing. It is a bid for the unlock. I have seen this movie and I know how it ends, and the ending is not the token holder's.

The Regulatory Floor Is Derivatives, Not Securities

Every DeFi analyst runs the Howey test and stops. Money in, common enterprise, expectation of profit, efforts of others — yes, mostly, medium-high risk, move on. That analysis is correct and incomplete, because it misses the exposure that actually distinguishes perps from the rest of DeFi.

A perpetual future is a derivative. In the United States that puts it under CFTC jurisdiction. Offering leveraged derivatives to retail requires registration as a designated contract market or a futures commission merchant. Most on-chain venues do not have it. The workaround is a familiar triad: geographic blocking, front-end restrictions, and an offshore foundation structure with a Cayman or BVI entity wrapped in decentralization narrative.

Here is the problem with the workaround. Geographic blocking is a front-end control, not a protocol control. Anyone with a VPN and a direct contract call is through it in ninety seconds. On-chain, the block is advisory. Which means the decentralization defense, which has some purchase in the securities context, is materially weaker in the derivatives context, because the thing being regulated — leveraged retail access — is not something the front-end can actually prevent.

Now stack the tension on top. The mechanism that improves value capture — fee-funded buybacks — is the same mechanism that strengthens the "expectation of profit from the efforts of others" prong. A buyback reads, to a regulator, like a share repurchase. Share repurchases are shareholder returns. Shareholder returns imply a security. The stronger the value capture, the stronger the securities case. There is no configuration that maximizes both. Every protocol optimizing its tokenomics is quietly worsening its legal position, and almost none of them have priced that in.

There is a broader precedent I keep returning to, and it sits underneath all of this. The Tornado Cash sanctions established that publishing immutable code can be treated as providing a service. I said at the time that this was the most dangerous line in the sand anyone had drawn, because once writing is a crime, every auditor, every contributor, every maintainer in this industry is a potential co-defendant. Perp venues are watching the derivatives question. They should be watching the code-liability question, because that one does not stop at the front-end. It goes all the way down to the people who wrote the matching engine.

And the regulatory shock, when it comes, will not land first on the protocol. It will land on the front-ends and the KOLs and the aggregators — the channel layer. Geo-fencing them is trivial. Their compliance cost spikes, their take rate rises to cover it, and the channel tax that was already permanent becomes permanent and expensive. Net revenue compresses from the bottom of the stack. Survival is the first metric; profit is the second, and the channel layer is where survival gets tested first.

The Picks and Shovels Nobody Prices

There is one more cost the gross-fee number hides, and it is the one I would flag in any technical viability check.

A perp venue that clears eight figures a day in fees is, by construction, one of the largest payers on whatever chain it settles to. Gas, blob space, sequencer fees — this is a top-tier customer for the underlying L1 or L2. Which means a meaningful slice of that $125 million is not staying in the protocol at all. It is flowing upstream as settlement cost. The self-built appchain exists to internalize that flow, which is rational, and which imports a new set of problems: a centralized sequencer, an admin key that can pause trading or move margin parameters, and a security budget that now has to be funded out of the same fee stream it was supposed to protect.

I have been skeptical of the dedicated data-availability thesis for years, and this is where the skepticism cashes out. I have yet to see a rollup whose blob throughput would strain Ethereum's calldata budget, let alone justify a separate DA committee with its own trust assumptions. The DA layer is sold as a scaling necessity and bought as a narrative. The perp venue building an appchain does not need a DA layer. It needs a sequencer it controls and a settlement guarantee it trusts. Those are different purchases, and conflating them is how teams end up paying for infrastructure they will never saturate.

The other unpriced cost is liquidation bad debt. When the market gaps, the liquidation engine either keeps up or it does not. When it does not, the shortfall is covered by the insurance fund or socialized across the pool. That cost never appears in a gross-fee headline. A venue can report record fees in the same quarter it quietly refilled its insurance fund from treasury. The fee line and the solvency line are two different stories, and only one of them gets published.

Every bug is a bug in the human expectation. The expectation here is that a fee is a profit. The bug is that a fee is a gross inflow with six exits and no disclosure of which exits are open.

The Contrarian Read: The Cash Flow Is Real, the Claim On It Is Not

Let me now argue against myself, because the bear case that only sees fraud is as lazy as the bull case that only sees volume.

Here is the strongest version of the bull argument. The fees are real. This is not a token with a whitepaper and a dream; it is a venue where leveraged traders pay actual money for actual execution, and that money is denominated in stablecoins, not in the protocol's own inflated unit. In a market where most "revenue" is circular, perp DEX fee flow is one of the few genuinely exogenous cash streams in DeFi. The demand is real because leverage is a permanent human appetite, not a narrative that can be turned off. On that reading, the $125 million is not a vanity metric at all. It is the most honest number in the sector.

I find that argument persuasive, and it is precisely why the bear case is so much worse than it looks.

The problem is not that the cash flow is fake. The problem is that the cash flow is real and structurally pledged to people who do not hold the token. The market makers are owed their rebate. The channels are owed their cut. The depositors are owed their risk premium. The validators are owed their security budget. These are contractual or quasi-contractual claims that sit senior to the token holder, and they are all denominated in the same fee stream. The token holder is the residual claimant on a stream that has already been fully allocated before it arrives.

So the contrarian insight is not "the fees are fake." It is sharper than that. The most bullish fact about perp DEXs — that the fees are real — is also the most bearish fact, because real fees attract real senior claims, and the token holder is last in line. A fake fee stream can be repriced upward when the narrative improves. A real fee stream that is fully pledged to market makers and channels cannot. You cannot narrative your way out of a seniority structure.

The second contrarian turn is about who the real winner is. It is not the protocol. It is not the token holder. It is the venue that sells the leverage and the layer that settles it and the market maker who quotes on all of them simultaneously. The protocol is the chokepoint that captures the least of its own throughput. Building empires on the volatility of belief — the empires belong to the picks and shovels, and the believers get the equity tranche.

And the third, which nobody wants to hear: the transparency gap is not a flaw in the analysis. It is the analysis. A sector that cannot produce a sourced, timestamped, gross-versus-net fee figure — while publishing precise volume dashboards to three decimal places — is telling you something about its disclosure culture. The missing data is data. When the number that anchors the entire bull case has no provenance, the absence of provenance is the finding.

The Metric That Replaces the Number

I am not going to give you a price target. I am going to give you the three fields that should replace the $125 million headline, because in a bear market the only question worth asking is which of these venues is actually solvent once the subsidies stop.

First, net protocol revenue, defined as gross fees minus maker rebates minus channel splits minus validator and sequencer cost minus insurance fund top-ups minus incentive spend marked to market. Most venues have never published this number, because for most venues it is close to zero or negative.

Second, post-incentive retention — the thirty-day user retention measured after the points program ends. This is the golden metric, and almost nobody reports it, because the answer is embarrassing. Volume that survives the subsidy is the only volume that was ever real.

Third, buyback coverage — the ratio of actual market buybacks to token emissions. If emissions exceed buybacks, the token is a subsidy instrument, not an equity claim, and the fee headline is decoration.

Shorting the hype to fund the truth is not a rhetorical flourish. It is a discipline. The number will keep circulating, because numbers with no source are the easiest things in the world to repeat. Your job is not to repeat it. Your job is to ask the four questions it cannot answer — who, when, gross or net, and paid to whom — and to notice that the silence after the fourth question is where the entire sector's valuation problem lives.

The next narrative will not be about how much perp DEXs earn. It will be about who is left holding the claim when the music stops. That is a different story, told by different numbers, and the venues that survive it will be the ones that started publishing the net figure before a regulator or a redemptions queue forced them to. The rest will keep quoting $125 million a month and hoping nobody asks for the source.