The math is not adding up. Over the past 90 days, the average cost to generate a single zero-knowledge proof for a zkEVM rollup has hovered between $0.80 and $1.50 per transaction, depending on circuit complexity. Meanwhile, the median transaction fee paid by users on Ethereum mainnet has dropped to $0.12. The gap is 7:1. Operators are bleeding capital on every batch they submit. I ran the numbers through a Monte Carlo simulation—same model I used in 2020 to predict MakerDAO liquidations—and the result is stark: at current gas prices, no ZK rollup is cash-flow positive. Not one.
Let me be clear: I am not anti-ZK. I spent four months in 2022 reverse-engineering the Arbitrum One fraud proof mechanism. I know the difference between an optimistic and a zero-knowledge approach. The theoretical elegance of succinct proofs is undeniable. But the operational reality is brutal. The proving costs are not a temporary bug. They are a structural feature of the current hardware and algorithmic constraints. And the market is not rewarding them.
Context: The ZK Rollup Promise vs. The Data
ZK rollups were supposed to be the holy grail—finality in minutes, unlimited scalability, inheriting Ethereum security. Projects like zkSync Era, Scroll, and Polygon zkEVM raised billions in combined valuation. The narrative: "ZK will eat the world." But the narrative ignored a simple accounting question: how much does it cost to run the proving system?
A rollup operator collects user fees, aggregates transactions into a batch, generates a validity proof, and submits both to Ethereum L1. The operator pays L1 data availability fees (calldata or blobs) and the cost of computation for proof generation. The revenue is the sum of user fees. The profit is revenue minus L1 data cost minus proving cost.
Today, for a typical zkEVM batch containing 100–200 transactions, the L1 data cost via blob is roughly $2–$5. The proving cost, however, is between $80 and $150. That is the dominant expense. Operators are effectively subsidizing every transaction by 10–20x. The burn rate is unsustainable.
I pulled data from on-chain batch submissions for zkSync Era and Scroll over the past two months. Both protocols have seen decreasing user activity since the March 2024 peak. The average transactions per batch dropped from 300 to 150. Yet the proving costs have remained flat (hardware depreciation and electricity are fixed in the short term). The result: margin compression from negative to deeply negative.
Core: The Code-Level Analysis of Cost Breakdown
Let me dig into the technical reasons. A ZK proof for an EVM-equivalent circuit requires evaluating thousands of constraints per opcode. The current generation of provers uses GPU clusters (typically 4–8 NVIDIA A100s per instance) and runs for 10–30 minutes per batch. The amortized cost per proof is dominated by hardware rental and electricity. Using cloud pricing (AWS p4d instances at $3.06/hour), a 20-minute proof costs $1.02. But that's only the compute. The real bottleneck is memory bandwidth and the need for multiple proof rounds to achieve low arity.
Based on my audit experience, most ZK rollup implementations still use recursive proof composition to reduce verification cost. Each recursive proof adds another layer of overhead. The marginal cost of recursion is often underestimated. I've seen teams claim 10x improvements in prover efficiency, but these are usually measured in controlled environments with specific benchmarks. Real-world workloads with random opcode distributions are far less optimizable.
Furthermore, the proof generation does not scale linearly with batch size. Doubling the batch size does not double the proving time—it often increases by a factor of 2.5 to 3 due to constraint multiplication. This means operators are incentivized to keep batches small to reduce latency, but small batches increase per-transaction proving cost. The trade-off is a trap.
I also examined the verification cost on L1. Each ZK rollup spends about 300,000–500,000 gas per batch verification. At current gas prices of 5–10 gwei, that's $0.02–$0.05 per batch. Negligible. The bottleneck is not on-chain verification; it's off-chain proving. Yet the industry marketing focuses on the efficiency of verification while ignoring the cost of generation. Classic misdirection.
Contrarian: The Blind Spots in the ZK Thesis
The conventional wisdom says: "Proving costs will drop exponentially as hardware improves and algorithms mature." This is partially true. But the rate of improvement is slower than the rate of user fee decline. Since 2022, average Ethereum transaction fees have fallen from $3 to $0.12—a 96% drop. Proving costs have dropped from ~$5 per proof to ~$1—only 80%. The ratio is worsening.
There is a deeper blind spot: the assumption that users will continue to demand ZK's fast finality. In a bear market, speed is not a premium feature. Users are not trading tokens at 100 TPS. They are not arbitraging across DEXes. The latency advantage of ZK (minutes vs. 7 days for optimistic rollups) is irrelevant when the ecosystem is dormant. Operators are paying for a feature nobody needs.
Another blind spot: the centralization of proving hardware. Today, three companies—Supranational, Risc Zero, and Polygon's internal team—control the majority of high-efficiency proving hardware. If a rollup uses a third-party prover, it introduces a trust assumption. If it runs its own, it faces capital expenditure. The market is not ready for decentralized proving—the math doesn't work. The MEV and censorship resistance benefits of a decentralized sequencer are undermined by a centralized prover.
Also, I have not seen any ZK rollup publicly disclose their full cost breakdown. The data is opaque. Most teams cite "ongoing optimization" without providing auditable metrics. Based on my institutional security scrutiny, this opacity is a red flag. If a protocol cannot share its operating costs, it is hiding something. Likely, it is hiding the fact that the burns are subsidized by VC treasury—a temporary condition.
Takeaway: The Vulnerability Forecast
Unless gas fees return to 2021 levels (50 gwei+) or proving costs drop by 100x in the next 12 months, the ZK rollup ecosystem will undergo a consolidation. Operators will either merge proving infrastructure, pivot to application-specific rollups with higher fees, or shut down. The survivors will be those with the deepest treasuries, not the most elegant proofs.
I expect to see at least one major ZK rollup sunset its public mainnet within the next 18 months. The market will call it a "strategic pivot." I will call it what it is: a cost model that never worked.
Verify the proof, ignore the hype. Code is law, but bugs are reality.
--