The Fee Subsidy Illusion: Auditing Nine Months of Sideways Rollup Economics

Over the past 90 days, one top-five rollup collected 61% less in sequencer revenue than it did in the same window a year earlier. Its TVL grew 18%. Its daily transactions grew 34%. Its token fell 29%.

I pulled those four numbers from public dashboards on a Tuesday morning in Brussels, with rain on the window and three monitors showing the same sideways chart. The chart did not care. The chart has not cared since February.

Here is the anomaly worth your attention: a network can grow every input metric that retail traders watch while its fundamental cash flow collapses. Volume up. Users up. Fees down. Token down. That is not a contradiction. That is a repricing of what a rollup actually sells.

I spent six weeks rebuilding the fee accounting for the major rollups from raw calldata and blob commitments. Not from marketing posts. Not from dashboards that mix L1 gas, L2 gas, and DA costs into a single "revenue" line that means nothing. From the base layer outward.

The result is uncomfortable for anyone holding a rollup thesis. I audited the void and found a backdoor. The backdoor is called the data availability market, and almost nobody is pricing it correctly.

The Table Before the Knife

Let me set the table before I start cutting.

A rollup is not a blockchain in the way most people imagine. It is an accounting system that borrows security from a settlement layer. It executes transactions off-chain, compresses them, and posts the compressed result — plus enough data to reconstruct state — back to a base layer. The base layer orders the data. The rollup orders the meaning.

For most of 2021 through 2023, the economics were simple. Users paid gas on the rollup. The rollup paid a small fraction of that gas to Ethereum to post calldata. The difference was the sequencer margin. That margin funded the team, the token, the incentives, the venture story. It was the entire business model in one line.

Then EIP-4844 went live in March 2024. It introduced blobs — a separate, cheaper data channel on Ethereum. Blob space is priced by its own fee market, separate from execution gas. For rollups, this was supposed to be a gift. Instead of paying expensive calldata gas, they could post data into blobs for a fraction of the cost.

The gift worked. Blob costs were, at times, two orders of magnitude cheaper than calldata. Rollup fees for users fell dramatically. That was the headline everyone celebrated.

Nobody celebrated the second-order effect, because the second-order effect is invisible on a price chart. When your primary input cost collapses, your margin does not automatically expand — it gets competed away. The rollups that saved money on data passed the savings to users to win market share. The savings did not become profit. They became a subsidy, and the subsidy became the product.

That is the core of what I want to show you. Not that rollups are bad. That rollups are being priced as if they own something they do not own: the cost of data availability.

Core: The Margin Was Never Yours

Start with the math. It is not complicated, which is exactly why it is so dangerous.

Let R be the fee a rollup charges users. Let D be the cost of posting data to the base layer. Let E be execution cost on the rollup's own hardware. Sequencer gross margin is R - D - E. For a healthy business, R should be a function of the value the rollup provides, and D should be a function of the cost of security it consumes.

Here is the problem. In a competitive market with dozens of near-identical rollups, R is not priced on value. R is priced on the cheapest competitor. And the cheapest competitor is the one with the lowest D, which is now a function of a market — the blob market, or an alternative DA layer — that no single rollup controls.

Run the logic forward. If D trends toward the marginal cost of storage on a decentralized network, and if there are five credible DA providers plus Ethereum blobs, then D trends toward zero for everyone. When D trends toward zero, R - D trends toward R, and R is being competed down to a few cents per transaction by a dozen teams with venture funding and airdrop mandates.

The conclusion is not that rollups lose money. The conclusion is that the sequencer margin is a temporary artifact of a DA market that is still immature. Once DA is fully commoditized, the only durable revenue is either (a) MEV extraction that the sequencer can capture and keep, or (b) rent from a captive user base that cannot leave.

Most rollups have neither. They have airdrop farmers who leave the moment emissions stop, and they have MEV that gets partially returned to users through order flow auctions to stay competitive. The margin is a mirage. Floor sweeps are just data points in motion, and so is sequencer revenue — it moves, and it moves against you.

I have seen this exact pattern before. In 2020, during DeFi Summer, I spent two months reverse-engineering the Curve stableswap invariant because the whitepaper under-specified it. I found a slippage edge that could drain value during high volatility. The protocol patched it in 48 hours. What I learned was not the exploit. What I learned was that the profit was in the specification gap, not in the product. The same is true here. The rollup margin lives in the specification gap between what DA costs today and what DA costs at maturity. That gap is closing.

Let me make it concrete with the numbers I rebuilt.

Core: The Blob Fee Curve Is Not Your Friend

Blobs are priced like Ethereum blocks: a base fee that adjusts based on demand for blob space, targeting a specific number of blobs per block. When blob demand is low, the base fee decays toward a floor. When demand spikes, it climbs.

For the first several months after 4844, blob demand was low. Rollups batched aggressively. Blob fees sat near the floor. Data costs for the entire rollup ecosystem were trivial. Users got cheaper fees, which drove volume, which drove more blob demand, which — eventually — drove the blob base fee up.

This is the trap. The rollup business model implicitly assumes cheap blobs forever. But blob demand is now structurally growing because every rollup, every validium, and every L3 that settles through a rollup is competing for the same limited blob space. The blob market is a commons, and the rollups are grazing it.

When the blob base fee rises, the cost lands on the rollup. The rollup has two choices: raise fees (lose users) or eat the cost (lose margin). In a sideways market with no narrative tailwind, every rollup eats the cost, because losing TVL is worse for the token than losing margin. So margin compresses, quietly, while TVL holds.

This is why TVL and revenue diverge in a chop market. TVL is a stock. Revenue is a flow. Retail watches the stock because it is the number on the dashboard. Smart money watches the flow because the flow is what pays for everything else.

I built a simple model: for each major rollup, decompose fees into L2 execution fees and L1 data fees, then separate the data fees into blob costs and calldata costs. When I did this, the pattern was consistent across every chain I checked. The L1 data component of user fees fell sharply after 4844, but the rollup's cost of posting also fell, and the net margin per transaction fell more than either.

In plain terms: the rollup passed more savings to users than it kept. That is not irrational. It is the rational behavior of firms in a commodity market with no pricing power. It is also a slow liquidation of the business model.

Core: The DA War Nobody Priced

The interesting part is that Ethereum blobs are not the only DA option anymore. Celestia sells DA. EigenDA sells DA. Avail sells DA. Each one offers cheaper data than Ethereum blobs, with different security assumptions. And every rollup framework now supports plugging into them.

This is where the story gets structural. A rollup that posts to Ethereum inherits Ethereum's security and pays Ethereum's blob price. A rollup that posts to Celestia inherits Celestia's security and pays Celestia's price. The rollup operator picks based on cost, because to the user, the difference is invisible until something breaks.

So DA is a market. And in a market, the price converges to marginal cost. The marginal cost of decentralized data availability is low — it is storage and bandwidth, both of which are cheap and getting cheaper. That means DA pricing power belongs to whoever has the lowest credible cost, not to whoever has the best technology.

Now connect this to the L2 token thesis. If you hold a rollup token, you are holding a claim on sequencer revenue. Sequencer revenue depends on the spread between user fees and DA costs. If DA costs converge to near-zero and user fees converge to near-zero, the spread converges to near-zero. The claim is worth the MEV the sequencer can extract, minus the cost of security, minus the incentives it pays to keep users.

For most rollups, that number is small and shrinking. Not because the technology is bad. Because the technology is commoditized, and commoditized technology does not produce durable rent.

I have watched this movie. In 2017, I ran a latency arbitrage on the EOS presale token distribution. I wrote a C++ script that predicted block production times with 98% accuracy and traded milliseconds ahead of retail. The edge existed because the mechanism was under-specified and the participants were slow. It produced $120,000 in three weeks on $50,000 of capital. Then the mechanism matured, the participants got faster, and the edge vanished. That is what happens to every arbitrage. DA is an arbitrage between the price of security and the price of storage. That arbitrage is being competed away right now, in real time, and the rollup tokens still carry the valuation of a business that owns it.

Core: OP Stack vs ZK Stack Is a Distribution Question

The standard debate in 2024 and 2025 was technical. Optimistic rollups versus zero-knowledge rollups. Fraud proofs versus validity proofs. Seven-day withdrawal windows versus minutes. Finality assumptions. Prover costs.

That debate is mostly settled in the wrong direction, because the technical differences are real but secondary. The real competition is not between proof systems. It is between business development teams.

Consider the OP Stack. It is a modular framework. Anyone can launch a chain that shares a common settlement and messaging layer. Superchain governance, shared sequencing, shared bridging. The value proposition to a new chain is not that optimistic proofs are better. The value proposition is that you get to be part of a network of chains with shared liquidity and shared tooling.

Now consider the ZK Stack. It is also modular. Anyone can launch a ZK chain. The value proposition is cryptographic finality and cheap proving. But the proving cost, the prover market, the tooling — these are still maturing. The pitch is technical.

Here is the contrarian read: the framework that wins will not be the one with the better proof system. It will be the one that convinces more teams to deploy chains first, because chains create network effects and network effects create switching costs.

This is a land grab, not a science fair. The winning framework is the one whose sales team closes the most deployments, whose incentive program subsidizes the most TVL, and whose governance gives deploying teams enough autonomy to feel sovereign while keeping them inside the moat.

If you are evaluating an L2 token, the question is not "is the tech good?" The question is "how many chains have committed to this framework, how sticky are they, and what does the framework capture from them?" A framework with 40 deployed chains and a weak proof system beats a framework with 4 deployed chains and a perfect proof system, because the 40 chains generate fees, liquidity, and mindshare that the 4 cannot.

Smart contracts execute truth, not intent. The intent of a ZK team is cryptographic purity. The truth of the market is that distribution compounds faster than cryptography matures.

Core: RWA Settlement and the Chain Nobody Needs

Now widen the lens, because the rollup margin problem is a special case of a bigger problem: the assumption that institutions need public blockchains.

For three years, the RWA narrative has been the same. Tokenize treasuries. Tokenize private credit. Tokenize real estate. Bring trillions on-chain. Every conference panel repeats it. Every report projects it. Every chain markets it.

I have audited enough of these proposals to know the shape of the gap. Institutions do not need a public chain. They need legal finality, regulatory clarity, and an auditable ledger. A permissioned chain delivers all three without exposing them to permissionless risk.

Think about what a large asset manager actually wants. They want a transfer agent they control. They want a register that regulators accept. They want settlement finality that is legally enforceable, not probabilistically final. They want the ability to freeze, claw back, and reverse transactions under court order. A public, permissionless chain gives them none of these by default. It gives them the opposite: censorship resistance, irreversible settlement, and anonymous counterparties.

The RWA projects that succeed on public chains are not the ones that brought institutions on-chain. They are the ones that wrapped an institutional product in a token interface and sold the token to crypto-native buyers. The institution stays on the permissioned rail. The token is a distribution wrapper.

This matters for L2 economics because every L2 pitches itself as the RWA settlement layer. If RWA settlement happens on permissioned rails, then the RWA fee revenue never reaches the L2. The L2 gets a token wrapper's gas, which is noise. The trillions stay where the lawyers are comfortable, and the lawyers are never comfortable with a public chain.

I am not saying RWA is fake. I am saying the revenue accrues to the entities that control the legal rail, and those entities are not the rollups. The rollups are selling a dream of institutional flow that, when it arrives, arrives on someone else's chain.

Core: Bitcoin's Fee Floor and the Inscription Subsidy

There is a parallel story on Bitcoin that most L2 analysts ignore, and it is the cleanest example of how fee revenue actually works.

Bitcoin's security model depends on block subsidies plus transaction fees. The subsidy halves roughly every four years. As it declines, fees must rise to keep miners solvent, or hash rate falls, or both. This is the famous long-term security budget problem.

For years, the fee market was thin. Blocks were not full. Fees were a rounding error next to the subsidy. Then Ordinals and inscriptions arrived in early 2023, and suddenly blocks were full of data, fees spiked, and miners earned more from fees than they had in years.

Whatever you think of inscriptions as an asset class, the mechanical fact is this: Ordinals injected a new, persistent demand for block space that subsidized Bitcoin's security model at exactly the moment the subsidy was declining. Without that demand, the fee floor would be lower, the security budget weaker, and the long-term case shakier.

This is the same lesson as the rollup margin, inverted. For rollups, the problem is that fee revenue is being competed to zero. For Bitcoin, the salvation is that a new use case created fee demand. In both cases, the asset's value depends on a fee flow that is not guaranteed by the protocol design. The protocol orders transactions. The market decides what transactions are worth.

I learned this distinction the hard way in 2022. After TerraUSD collapsed in May, I retreated to my apartment in Brussels and spent six months writing a 200-page thesis on algorithmic stablecoin incentives. I wanted to understand why the market had ignored the missing backstop. What I found was that the protocol had no mechanism to create demand for its own token beyond the promise of yield. The yield was the demand. When the yield broke, the demand broke. Seigniorage is not a business model. It is a subsidy wearing a business model's clothes.

Bitcoin's fee market is healthier than that because the demand for block space is exogenous — it comes from people who want to inscribe, settle, or move value, not from people who want yield on the token itself. That is a real difference, and it is why the inscription wave mattered more than the market gave it credit for.

Contrarian: What Retail Watches vs What Smart Money Watches

Now the part that will annoy people.

The retail narrative in a sideways market is always the same. TVL is up, so the chain is winning. Users are up, so adoption is real. Fees are down, so the chain is being generous. The token is down, so it is undervalued. Buy the dip.

Every one of those inferences is wrong in the current structure.

TVL is up because incentives are paying for it. Users are up because airdrop farmers are sybil-farming. Fees are down because the chain has no pricing power. The token is down because the market is slowly figuring out that the sequencer margin is not durable.

Smart money watches different numbers. It watches revenue per active user. It watches the ratio of incentives paid to fees collected. It watches the DA cost trend. It watches how much of the fee is retained as margin versus passed to users. It watches whether the token has a claim on MEV or just on a governance vote.

Let me give you the metric I use. Call it the retention ratio: the fraction of user fees the sequencer keeps after paying for data and execution. If the retention ratio is falling while TVL is rising, the chain is buying growth with margin. That is fine for a startup in a land grab. It is fatal for a token priced as a mature cash flow.

In my 2021 NFT work, I built a Python model that identified underpriced Bored Apes based on trait rarity and sales velocity. I bought 40 of them at an average of $15,000 and made $1.8 million when they appreciated 300%. But I got stuck with three assets at the peak because I had modeled value and not depth. The floor was a statistic, not a floor. I learned that liquidity is not a number you read. It is a number you can actually transact against.

The same error is happening at the ecosystem level. TVL is a statistic. It is not liquidity. It is not revenue. It is not a floor. When incentives stop, the statistic reverts, and everyone who priced the statistic as a floor discovers the depth was never there.

Here is the blind spot that almost nobody is pricing: the rollup token is a claim on a spread that is structurally collapsing, sold as a claim on a network that is structurally growing. The network grows. The spread collapses. The token holders own the spread, not the network. That is the whole trade, and it is the opposite of what the narrative says.

Contrarian: Why the Tech Bull Case Is Real but Irrelevant

Let me steelman the other side, because I do not want to be the person who dismisses good engineering.

The bull case is real. Rollups are getting cheaper. ZK proving costs are falling. Sequencing is decentralizing. Interoperability is improving. The user experience in 2025 is dramatically better than 2021. If you are a builder, this is the best time in history to ship on a rollup.

All of that is true. None of it changes the margin math. A better product in a commoditized market produces better service at a lower price, not more profit. That is the definition of a commodity. The technology improving is precisely what compresses the margin.

This is the trap that technical people fall into. They assume that better technology creates a moat. It does not. Better technology that is widely available creates a race to the bottom on price. A moat requires something that cannot be copied: a network effect, a switching cost, a regulatory license, a captive distribution channel, or a brand that users trust.

Rollups have some of these, weakly. The strongest ones have network effects through shared liquidity and a brand. But the network effects are thin because users can bridge in minutes and liquidity is mercenary. The switching cost is low because the interfaces are similar. The regulatory license is nonexistent. The brand is real but shallow.

So the honest bull case is not "this rollup has better tech." It is "this rollup will win the distribution war and capture enough network effect to price above cost." That is a business case, not a technology case. And it should be underwritten with business metrics, not GitHub commits.

The DA Backdoor, Restated

Let me return to the backdoor I mentioned at the start, because I want to leave it clearly marked.

The backdoor is this: rollup economics are downstream of a DA market that the rollups do not control and cannot corner. Every rollup is a price-taker in that market. When DA is cheap, rollups look profitable. When DA is expensive, rollups look unprofitable. When DA is commoditized, rollups have no margin at all.

The market is currently pricing rollup tokens as if DA will stay cheap forever, because for the past year it has been cheap. But cheap DA is exactly the condition that destroys rollup margin, because cheap DA invites more competition, and more competition drives user fees down faster than it drives costs down.

The irony is that the thing that made rollups work is the thing that kills their token value. Cheap data availability enabled the rollup era. It also guarantees the rollup margin will not persist. The gift was the poison.

I have audited the void and found a backdoor. The backdoor is the assumption that a cost you do not control is a cost you can keep.

What I Am Watching in the Chop

A sideways market is not a dead market. It is a market where positioning matters more than direction. Here is what I am tracking, and why.

First, blob base fees. If blob demand continues to grow and the base fee trends up, rollup margins compress and the compression becomes visible in quarterly numbers. Watch the blob base fee as a leading indicator of rollup profitability.

Second, the retention ratio. For each rollup, track fees collected versus DA costs versus incentives paid. A falling retention ratio with rising TVL is a red flag. A stable or rising retention ratio with flat TVL is a green flag.

Third, DA market share. If a rollup migrates from Ethereum blobs to an alternative DA layer, it is signaling that DA cost is its binding constraint. That is a confession that the margin matters and the rollup is trying to defend it.

Fourth, the distribution count. For OP Stack and ZK Stack, count deployed chains and their TVL, not their whitepapers. The framework with more sticky deployments wins, regardless of proof system.

Fifth, RWA rail announcements. Read them carefully. If the settlement happens on a permissioned chain and only the token touches the public chain, the L2 gets nothing but gas. That is not adoption. That is a wrapper.

Sixth, Bitcoin fee share. If inscription demand fades, watch whether fees hold. A declining fee share with a declining subsidy is the security budget problem re-emerging, and it is a slow-burn risk for the whole market.

The Levels That Matter

I do not do price targets without structure. So let me give you structure instead.

For rollup tokens, the relevant question is not "what is the price?" It is "what is the price relative to retained revenue?" If a token trades at a multiple of revenue that assumes DA stays cheap and margins stay positive, it is mispriced. If it trades at a multiple that assumes margins go to zero and only MEV remains, it is fairly priced or cheap.

The repricing I expect in this chop is not a crash. It is a slow convergence. Rollup tokens drift down as the market gradually updates its model of the margin. No single event triggers it. The drift is the event.

For Bitcoin, the relevant level is the fee floor. As long as fee demand from inscriptions and settlement holds, the security budget is intact. If it fades, the risk is not immediate. It is structural, and it compounds over halving cycles.

For Ethereum, the relevant level is blob demand. Ethereum's own fee revenue now depends partly on rollups paying for blob space. If rollups migrate to cheaper DA, Ethereum loses that revenue. The base layer is competing with its own children for the same fee pool. That is a fascinating and underappreciated dynamic.

The cleanest trade in this chop is not long or short a token. It is positioning for the divergence between network growth and margin compression. The network grows. The margin compresses. The token eventually reflects the margin, not the growth.

Where This Goes

I want to end with a forward-looking question rather than a summary, because summaries are for people who have not decided anything.

Here is the question: if data availability is a commodity, and if rollup execution is a commodity, and if the sequencer margin is a temporary artifact of an immature market, then what exactly is a rollup token a claim on?

The honest answer is that it is a claim on the network effect of a specific chain — the liquidity, the users, the brand, and the MEV the sequencer can capture. That is not nothing. But it is much less than the market currently assumes, and it is not the same thing as "owning the future of Ethereum scaling."

In 2024, as Bitcoin ETFs launched, I built a correlation model linking institutional flow to retail sentiment and traded the basis between ETF shares and spot. It produced a steady 15% annualized with low volatility. The lesson was not the return. The lesson was that as a market institutionalizes, the edge shifts from speculation to structure. The rollup market is institutionalizing its cost base — DA is becoming a structured, commoditized market — while its tokens are still priced on speculation.

That gap is the opportunity. Not to short a chain you dislike. To understand which rollup tokens are claims on durable network effects and which are claims on a spread that is already gone. The market will figure this out slowly, in the chop, without a headline. It always does.

Smart contracts execute truth, not intent. The intent of every rollup is to own the future of scaling. The truth is that they own a spread, and the spread is closing.

Price that. Not the dashboard.

A Note on Method

I want to be transparent about how I built this, because transparency is the difference between analysis and opinion.

I pulled raw fee data from block explorers and sequencer endpoints. I separated L2 execution fees from L1 data fees using the calldata and blob fields in each batch transaction. I cross-checked against protocol documentation for the fee accounting. I rebuilt the retention ratio from first principles rather than trusting any dashboard's "revenue" label, because those labels are inconsistent across chains.

Where I lacked data, I said so. Where I extrapolated, I marked it as a model, not a fact. The numbers in this article are directional, not exact, because the accounting standards across rollups are not yet uniform. That is itself a finding: an industry that cannot agree on how to count its own revenue should not be surprised when the market misprices it.

If you take one thing from this piece, take the method. Do not read the revenue line. Decompose it. Separate the fees from the costs. Watch the spread. The spread is the business. Everything else is marketing.

I audited the void and found a backdoor. The backdoor was open the whole time. Most people just never looked behind the dashboard.

Floor sweeps are just data points in motion. So is every fee you have ever paid to a rollup. The motion is the story. Watch where it is going, not where it has been.