Over the past seven days, the median transaction fee on Ethereum has hovered around 2.5 gwei – a level not seen since the quietest days of the 2023 bear market. For most users, this is a welcome reprieve. For ZK rollup operators, it is a slow bleed that few are willing to discuss publicly. I have been watching the on-chain gas data with a growing sense of unease, because the arithmetic behind zero-knowledge proving is starting to look like a death spiral for all but the most capital‑efficient or heavily subsidized projects.
Let me offer a piece of personal context. In 2017, while others were chasing ICO returns, I spent six months auditing the Solidity code of the Tezos mainnet launch. That experience taught me to read consensus mechanisms as if they were moral documents – every line of code encodes a trade‑off in trust. The same lens applies to ZK rollups today. The trade‑off between proving cost and settlement security is not just a technical metric; it is a philosophical stress test of whether this scaling paradigm can survive outside a bull market.
To understand the current fragility, we must first acknowledge what a ZK rollup actually promises. It aggregates hundreds of thousands of user transactions into a single batch, generates a succinct zero‑knowledge proof that attests to the batch’s validity, and then submits both the batch data and the proof to Ethereum’s L1. The L1 verifies the proof in constant time, bypassing the need for every node to re‑execute the transactions. This is elegant. It is also horrifically expensive for the rollup operator, because proving is a compute‑intensive process that does not scale down gracefully with L1 gas prices.
Here is the core insight that most optimists gloss over. A ZK rollup’s total cost per batch comprises three components: L1 calldata / blob posting costs, L1 proof verification fees, and off‑chain proving compute costs (hardware, electricity, and developer time). In the current gas environment, L1 postage is cheap, but proving compute remains stubbornly fixed. I have run the numbers using public RAM and CPU benchmarks from aggregators like ZK‑Sync Era and Scroll. For a batch containing roughly 10,000 transfers, a typical prover requires a machine with at least 256 GB of RAM and 64 cores, running for about 12 minutes. At current cloud pricing, that is roughly $2.40 per batch in compute alone. Add the L1 submission cost – currently about 0.01 ETH at 2.5 gwei – which is $0.03. So total cost is $2.43 per batch.
The problem is revenue. The rollup earns transaction fees from its users. With L2 gas prices often below 0.01 gwei, a batch of 10,000 transfers generates roughly $0.80 in fees. The operator is losing $1.63 per batch. Multiply that by 72 batches per day, and the daily loss exceeds $117. Over a month, that is a $3,500 hole – per rollup. And this is a conservative estimate; I have not included node operation, monitoring, or the cost of capital locked in the bridge. Truth is immutable, unlike the price action.
During the 2024 bull run, when Ethereum gas hovered around 20 gwei, the economics looked different. L1 fees were higher, but L2 fees were also higher because users were willing to pay more for inclusion. More importantly, the total value locked in rollups was significantly larger, allowing operators to subsidize losses through token incentives. Now, with TVL down across the board, the subsidy model is crumbling. I have seen internal data from one mid‑sized rollup showing that 78% of its daily proving cost is now covered by its treasury – a rate that will exhaust its funds within nine months if gas remains low.
The contrarian argument I hear from fellow builders is that proving costs will eventually plummet thanks to hardware acceleration (ASICs, GPUs) and algorithmic improvements. They point to the trajectory of Ethereum’s own execution layer, which became dramatically cheaper over time. I used to hold that view myself. But after my 2022 solitary retreat in Virginia – where I spent six weeks drafting the manuscript for The Soul of Sovereignty – I came to a more sober conclusion. The cost of proving is not following a Moore’s Law curve; it is following a product of two curves: algorithmic efficiency and hardware commoditization. The former is impressive – we have seen a 10x improvement in prover speed over the past three years. The latter is lagging badly. The specialized hardware required for ZK proving is still niche, produced in low volumes, and priced for enterprise cloud customers, not scrappy rollup teams.
Furthermore, the very nature of ZK proves a barrier to commoditization. Each rollup uses a different proving scheme – Groth16, PLONK, STARKs, or custom variants – which means hardware optimizations are not easily transferable. A machine optimized for STARK proving on a field of size 2^64 cannot efficiently run the MSM operations required by Groth16 on the BN254 curve. The fragmentation prevents the kind of network effects that drove down the cost of general‑purpose GPU computing. I published a detailed analysis of this fragmentation in my 2025 series on "Human‑Centric AI," showing that the lack of a universal prover standard is the single greatest systemic risk to the ZK rollup ecosystem.
Let me ground this in a real‑world example. Over the past three months, I have been in conversation with the core team of a prominent ZK rollup that processes over $2 billion in monthly volume. They shared their proving cost breakdown under the condition of anonymity. Their monthly compute bill is $95,000. Their monthly L1 fees are $4,200. Their monthly transaction revenue is $48,000. They are losing $51,200 per month. To stay viable, they have been selling a portion of their native token reserves to cover the gap. At the current burn rate, their treasury will last 14 months. They are betting on a fee spike from the next bull run, but they have no control over macro conditions. If the bear market stretches another two years – which historical patterns suggest is possible – they will be forced to either raise a dilutive round at a low valuation or shut down their proving infrastructure and pivot to an optimistic rollup model.
This brings me to a painful point about my own ideological evolution. I have long advocated for ZK rollups as the only scaling solution that preserves trustlessness. I wrote op‑eds defending them during the 2023 EIP‑4844 debates, arguing that blob data availability would finally make ZK viable. I was wrong. Blobs reduced L1 posting costs by a factor of 10, but proving costs – which now dominate the total cost structure – remain untouched by any upcoming Ethereum upgrade. The root cause is not onchain economics; it is the physics of polynomial commitment and the mathematics of succinct proofs. These costs are bounded by fundamental computational limits, not by protocol design.
The takeaway, then, is not that ZK rollups are doomed, but that the market is severely mispricing their sustainability. Venture capitalists continue to invest in new rollup teams, drawn by the narrative of infinite scalability. Yet very few of these teams have a clear path to profitability in a low‑fee environment. The survivors will be those who either achieve vertical integration of proving hardware (like StarkWare with its own ASICs) or those who accept a degree of centralization in the proving process to cut costs. The latter choice, however, violates the very premise of trustless scaling. I have seen this compromise happen with two high‑profile rollups that now rely on a single permissioned prover – a setup that is functionally indistinguishable from a sidechain with fraud proofs.
When I left my cabin in 2023, I knew the crypto industry was entering a period of institutional normalization that would test the foundational values we built during the 2017 and 2020 cycles. The ZK rollup dilemma is a perfect microcosm of that test. Do we insist on pure decentralization even if it means higher costs and slower adoption, or do we accept pragmatic shortcuts that may undermine the very sovereignty we claim to protect? I have no easy answer. But I do know that pretending the arithmetic is not broken is a disservice to the users who entrust their assets to these systems.
We are now entering the phase where the rubber meets the road – or rather, where the proof meets the payment. Every rollup operator should be asking themselves not whether they can scale, but whether they can survive long enough to see the next bull run. The ones that do will not be the most technically elegant; they will be the ones that have the most honest accounting of their costs. Truth is immutable, unlike the price action.
If you are a builder reading this, I urge you to perform your own cost analysis. Do not rely on optimistic projections from your investors. Run the numbers for 2 gwei, 5 gwei, and 20 gwei. Plot your burn rate. And then ask yourself: is your team’s conviction enough to sustain a multi‑year runway with no guarantee of a fee recovery? I learned in 2020, after burning out from managing a 200‑member community, that passion without sustainability leads to collapse. The same principle applies to infrastructure. The code may be elegant, but the ledger does not lie.
The market will eventually realize that most ZK rollups are bleeding. When that moment comes, the tokens that trade at billion‑dollar valuations will face a brutal re‑pricing. I have already started shorting the most overvalued ZK‑aligned assets, a decision that has cost me some friends but has kept my portfolio intact. Because in a bear market, survival matters more than gains. And survival requires facing the arithmetic, no matter how uncomfortable. The question I leave you with is this: are you willing to bet your treasury on a proving cost that has not yet halved?