Hook
Ninety days. That's how long it took for three of the top ten Ethereum rollups to spend more on proof generation and data settlement than they collected in sequencer fees. Gross — before token incentives, before treasury grants, before the ecosystem fund quietly backstopping the whole thing.
Numbers are mine. I modeled them; I didn't lift them from a deck. Assumptions are stated below, so you can argue with the arithmetic instead of the vibe.
Proving a batch on a general-purpose ZK circuit runs somewhere between $0.0008 and $0.011 per ETH-transfer-equivalent. That spread is wide because it depends on circuit complexity, batch utilization, and whether you rent GPU capacity on spot or reserve it on contract. A user pays $0.02 on the same rollup. Contribution margin lands in the low single-digit thousandths of a dollar per transaction. Then subtract L1 blob settlement, prover redundancy, RPC serving, indexers, a security council's legal retainer, and payroll.
What's left is a token narrative wearing a balance sheet.
— Root: Auditing the DAO and Ethereum
Context
Ethereum's rollup economy has three cost centers. Only two of them are cheap right now.
Data availability is the cheap one. Blob space after EIP-4844 collapsed the cost of posting L2 state to L1 by orders of magnitude, and in a low-demand regime the blob base fee sits near its floor. When blob demand is quiet, an L2 settles a full batch for less than the tip on a single L1 swap.
Settlement and verification are the second cost. For optimistic rollups that's a bond and a challenge window; for ZK rollups it's a verifier contract call — gas-cheap and boring.
Proof generation is the third, and it's the one nobody prices. A ZK rollup doesn't just post state; it must produce a cryptographic argument that the state transition was valid. That argument costs real compute, real electricity, real GPU time. It is the only line item in the stack that has not gotten structurally cheaper since 2021. Unit costs have fallen through better circuits and better hardware — but volume has grown faster than unit costs have dropped. That's the trap.
Meanwhile, sequencer revenue is narrow. Base fee plus priority fee, plus whatever ordering value the sequencer captures. On most L2s, ordering value — MEV, in plain language — is the actual profit center, not the fee line. The fee line is marketing. It's what makes a token look like it has product-market fit.
The rollup-centric roadmap assumed fee revenue would eventually cover all three cost centers. Two of three are solved. The third is the sword.
— Root: Auditing the DAO and Ethereum
Core
Let's do the math properly. The model is simple: fixed monthly opex F, contribution margin per transaction c, break-even volume V* = F / c.
Fixed opex for a mid-tier ZK rollup, conservatively:
- Prover fleet amortization and reserved capacity: $400K–$900K per month
- Prover redundancy across two regions, because a stuck prover is a stuck chain: $150K–$300K
- RPC, indexers, archival nodes: $80K–$200K
- Security council compensation, legal, audits, bug bounties: $120K–$350K
- Protocol engineering, research, and growth headcount: $600K–$1.5M
Call it $2.4M per month at the midpoint. That's $28.8M a year to keep a chain alive that charges two cents a transfer.
Now the margin. If a user pays $0.015 and L1 settlement plus proving consumes $0.011, then c = $0.004. If settlement is subsidized by blob floors at $0.002 and proving is well-optimized at $0.006, c = $0.007.
Be conservative. Use c = $0.002.
V* = $2.4M / $0.002 = 1.2 billion transactions per month. Forty million per day.
The busiest rollups in this cycle clear somewhere between three and ten million transactions per day. Not forty.
So a mid-tier chain running honest unit economics needs four to thirteen times its current throughput just to break even on operations — before it pays the L1 a dollar for security, before it funds a grants program, before it does anything a token holder would call growth.
Here's where the audit gets interesting. Those chains are not shutting down. They are running. Something is closing the gap, and it is not the fee line.
Three candidates.
First, token emissions routed through liquidity mining and ecosystem incentives. This is a transfer from token holders to users, laundered through a growth narrative. It works until the token price stops working. We farmed the yields until the protocol farmed us.
Second, sequencer MEV. On an L2, the sequencer sees every transaction before it lands, and most chains have no credible commitment device preventing reordering. The fee line is public and embarrassing. The ordering value is private and profitable. If you want to know how an L2 actually funds itself, look at who controls the mempool, not who publishes the fee schedule.
Third, venture runway spent to buy market share ahead of an inevitable consolidation. That's not a business. That's a countdown.
The tell is disclosure. Ask an L2 team for proof cost per transaction and sequencer margin per block. Ask the way you'd ask a bank for a capital ratio. What you get instead is total value locked, transaction count, and a blog post about modularity. TVL measures deposits, not solvency. Transaction count measures subsidy efficiency, not demand.
I've seen this exact silence before. In 2022 I spent three weeks reading Terra's mint-and-burn mechanism against its stated reserves and found the same shape: a system whose public numbers looked fine and whose private cost structure was terminal. I restructured 60% of my portfolio into stables and BTC before the peg broke. Not because I'm clever — because I read the mechanism instead of reading the community.
The same discipline applies here. ZK proving cost is the reserve. If fee revenue can't cover it, something else is covering it, and that something else has a clock on it.
One more thing, since everyone is repeating it: liquidity fragmentation is not the problem. Fragmented liquidity is what every market looks like before it matures. It is the standard sales pitch for the next bridge, the next intent layer, the next solver network — sold by people who profit from the aggregation layer whether or not the fragmentation was ever real. The actual problem is that rollups carry a fixed cost structure that scales with proofs, not with users. No amount of cross-chain messaging fixes a unit-economics hole.
— Root: Auditing the DAO and Ethereum
Contrarian
Retail's mental model: L2 tokens are discounted infrastructure equity. Buy the chain, own the growth.
Audit's mental model: the L2 token is a subsidy instrument whose primary economic function is to cover the proving bill and buy throughput until proof costs fall — or until the chain consolidates.
Those are not the same asset.
The governance layer makes it worse. Read the last ten grant proposals on any major L2 forum. The ones that pass are the ones with delegate support, and delegate support is purchased with delegation from a handful of funds. Turnout on these votes routinely lands under 5% of circulating supply — sometimes under 2% of eligible holders. "Community decision-making" describes a forum, not a polity. When the treasury is the only thing keeping the chain solvent, the people deciding how to spend the treasury are functionally deciding the chain's runway.
If you hold the token, you are not voting on growth strategy. You are voting on how fast the subsidy burns.
Takeaway
Watch three numbers, not ten: proof cost per transaction, sequencer margin per block, and the blob fee regime. When blob demand returns and L1 data stops being free, the second and third rollups by market share will find out what their unit economics actually are.
The price of ETH is not the signal. The cost of a proof is.
