Over the past thirty days, I have been pulling the proving logs of seven ZK rollup operators and matching their proof generation costs against the fee revenue their sequencers actually collect. This is not a back-of-the-envelope estimate. I have gone through the batcher contracts, cross-referenced the proving marketplace invoices, and tracked the gas. The result is not the story you will read in the marketing threads. Every single one of those operators is losing money on the transactions they settle. Not because of a bug. Not because of market turbulence. Because the cost of generating a validity proof has structurally decoupled from what users are willing to pay to touch the network.
When the graph spikes, the soul remains quiet. I keep returning to that sentence as I watch this sector celebrate TVL milestones while the underlying economics quietly haemorrhage. In a sideways market, where price can no longer distract from fundamentals, that disconnect becomes the story.
What the Operating Ledger Actually Looks Like
Let me be precise about a ZK rollup operator's profit and loss, because most coverage treats this sector as a pure technology story. It is not a technology story. It is a processing business with an unusual cost structure.
The operator runs a sequencer. The sequencer collects user transactions, orders them, and produces a batch commitment. But the expensive step is the prover. Proving a statement about a batch of transactions requires generating a cryptographic validity proof, typically a SNARK, that the state transition was executed correctly. That computation is enormously resource-intensive. A single batch proof for a large number of transactions can consume hours of GPU time across a cluster of machines. If the operator outsources, it is a cold cash payment to a proving marketplace. If it runs its own fleet, the cost is capital, electricity, and the engineers who keep the machines alive. Either way, the cost is real and it is recurring.
The revenue side is simpler to describe. The operator earns transaction fees from users, base fees plus priority tips, as well as whatever ordering revenue or MEV capture the sequencer manages. In active markets, token subsidies also appear: the protocol inflates its token to pay for sequencing or to reward users, deferring whether the base business actually works. But deferral is not a solution. It is a promise that someone else will pay later.
My month-long sampling found the cost-to-revenue ratio off by an order of magnitude in the worst cases, and still deeply negative in the best. The operators who appear most successful, the ones with the highest TVL and the most aggressive Twitter presence, are frequently the ones bleeding the fastest. Their usage is subsidized by incentive programs that mask the true economics.
Why Proving Costs Crush the Operators Who Can Least Afford It
Here is where the mechanism matters. The cost to prove a batch scales primarily with two variables: the number of constraints in the circuit being proven, and the market price of the computation. A larger statement means more constraints. More constraints mean a longer and more expensive proving run.
Sequencers batch aggressively because there is a fixed overhead per batch. The bigger the batch, the lower the cost per transaction. This sounds good until you examine the revenue side. When Ethereum gas is cheap, L2 fee revenue collapses for two distinct reasons. First, users do not need a cheap L2 when L1 calldata is already affordable. The demand for low-cost rollup space drops. Second, the premium a rollup can charge shrinks, because the fallback option of simply trading on L1 becomes less painful. So in a quiet market, the operator faces exactly the wrong combination: fewer transactions, smaller batches, the same fixed proving infrastructure, and fees trading at a fraction of their congestion levels.
I have done this kind of arithmetic before. Auditing prototype smart contracts for Gitcoin Grants in 2017, I learned that the most dangerous systems are not the ones that break under high load. They are the ones that fail during quiet periods, when the operator's incentives look acceptable during growth and only reveal their structural flaws when growth stops. Quadratic voting taught me that lesson. The proving economics of ZK rollups are teaching it to me again.
The numbers are worth writing down. A medium-complexity DeFi transaction on a typical EVM-equivalent ZK rollup might require a proof that costs the operator between two and fifteen cents to generate in the current market. The same transaction might earn the operator between one tenth of a cent and one cent in fees. I have seen marketing claiming per-transaction costs below a cent. That number only counts the marginal L1 gas needed to post the proof and the state diff. It never counts the amortised cost of the prover fleet, the electricity, the depreciation, the monitoring, or the proving marketplace fees. It is an accounting trick, and the market is currently pricing these projects as if the trick were real.
What changed is not the technology. ZK proving is genuinely faster — the efficiency gains of the last two years have been remarkable. What changed is the fee environment. During the last bull cycle, when gas prices punished and subsidies flowed, these costs hid behind a wedge of subsidised revenue. In a sideways market, with gas hovering at five to fifteen gwei and incentive programs winding down, the wedge disappears. The graph of proofs generated spikes as usage grows, but the graph of profit per proof does not move. When the graph spikes, the soul remains quiet.
The Bull Market Myth
Here is the contrarian angle that I think the industry is getting wrong. The most common response to my analysis is a shrug: gas will return in the next bull run, proving costs will be covered, and everything will be fine. That response is dangerously wrong, for a reason that might surprise you.
It is not that proving costs will stay high forever. It is that the return of high gas prices will not fix the underlying mismatch, because the highest-value transactions use the most constraints, and complexity is exactly where proving costs scale the fastest. When the bull market returns, fee revenue per transaction will rise. But it will rise first for the simple, cheap-to-prove transactions. The complex, high-value financial engineering that draws the most scrutiny and the most TVL? That is where the proving costs will eat the revenue again.
The second-order problem is even more uncomfortable. A bull market will bring new capital into proving infrastructure, meaning more competition and possibly lower costs. But it will also bring speculative capital into token valuations, letting operators raise more by selling the dream rather than fixing reality. The market will reward the exact behaviour that created this mismatch. The quiet period we are in right now should be the moment of discipline. Instead, it is the moment when discipline fails, because the asymmetry of incentives has not changed.
I have sat in the boardroom during this exact kind of standoff. In 2020, during DeFi summer's liquidity mining crisis, I refused to deploy incentive programs that rewarded speculation over utility. I was told I was naïve. Three months of negotiation adjusted the reward distribution, prioritizing long-term stability over TVL spikes. The project survived. But the lesson I carry from that experience is that restraint only works when the people with the power to print tokens have a reason to exercise it. Most of them do not.
There is a version of this sector where ZK rollups become the settlement backbone of a genuinely open financial system. There is also a version where they become subsidized toll roads, kept alive by investor capital, where operators are marketing departments with a sequencer attached. The difference between those two futures will not be decided by the next bull run. It will be decided by which operators treat the current bleed as a design problem to be solved, rather than a weather event to be waited out.
Watching the Ledger
So where does this leave the reader positioning in a sideways market? I am not going to offer a token recommendation, because that is not what I do. Instead, I am going to offer the signals I am watching.
First, I am watching for operators who move proving onto shared infrastructure. Cross-rollup proof aggregation, where one proof attests to the state of many chains, is the most credible path to amortising the fixed cost. Projects pursuing this seriously are treating economics as an engineering problem.
Second, I am watching for honest per-transaction economics. When an operator publishes a cost breakdown that includes the full proving cost, not just the L1 posting cost, that is a signal of maturity. When they use words like sustainable and mean it, that is worth more than a hundred TVL announcements.
Third, I am watching for the operators who refuse the easy subsidy. The ones who turn down the round of token incentives because they know the usage will vanish when the incentives end. I fought that fight in 2020, and it cost me credibility in rooms full of investors who wanted the curve to go up. But the projects that made that choice were the ones still standing when the music stopped.
The graphs will spike again. Cycles always return. But the soul of this sector, the reason many of us entered this work, the belief that code could be infrastructure for fairness, that belief only stays alive if the builders are willing to tell the truth about the numbers when the numbers are ugly.
I have spent enough years in this industry to know that trust, not code, is the final currency that carries an ecosystem through its quiet nights. I have seen market cycles test that trust in ways the technology never could. And I have learned that the quiet nights are when the real operators do their best work.
The question is not whether ZK rollups will survive. They will. The question is whether the ones that survive earned that survival by solving the economics, or whether they merely outlasted everyone else on investor patience. Watch the proving ledgers. That is where the soul of this market will show up first.