Tracing the signal through the noise floor.
Over the past 90 days, the average proving cost per transaction on ZK-rollup networks has hovered between $0.12 and $0.18. At current Ethereum gas prices—averaging 8 gwei—the L1 data posting cost per transaction is roughly $0.03. The arithmetic is brutal: operators are spending 4x to 6x more on proof generation than on the actual settlement layer. This is not a sustainable equilibrium.
Context: The Bull Market Assumption
When ZK-rollups first emerged as a scaling solution, the narrative was simple: move execution off-chain, generate a validity proof, post it on L1, and inherit Ethereum security. During the 2021 bull run, with gas fees at 150 gwei, the economics worked. Proving costs were a fraction of the gas savings, and operators could subsidize losses with token incentives. The market was willing to pay for speed and low fees.
Fast forward to 2025. The bear market has reset expectations. Gas is cheap, token prices are down 70% from peaks, and subsidy programs are being slashed. The fundamental equation has flipped. Proving is no longer a rounding error—it is the dominant cost center. Based on my experience auditing Layer-2 protocols in 2023, I have seen internal dashboards where operators are burning through treasury reserves just to keep the sequencer running. The code does not lie, but it is incomplete: the financial models underpinning ZK-rollups were built on bull market assumptions that no longer hold.
Core: The Proving Cost Breakdown
Let me walk through the numbers. A typical ZK-rollup batch might contain 500 to 1,000 transactions. The proof generation time varies from 30 seconds to 5 minutes depending on hardware. For a single prover running a GPU cluster, the cost per batch is approximately $0.05 to $0.15 in cloud compute. That is $0.0001 per transaction—negligible. But real-world proving is not that simple.
Decentralization requirements force the use of multiple provers, often with redundancy. The leading ZK-rollups—zkSync Era, Scroll, Starknet—all run on custom hardware with high upfront capital expenditure. The real cost is amortized over years, not seconds. A single prover node with top-tier NVIDIA GPUs costs around $50,000 to $100,000. Assuming a 3-year depreciation, that is $45 to $90 per day. Add electricity, cooling, network bandwidth, and maintenance—another $20 per day. Total fixed cost: $65 to $110 per day.
Now, the variable cost: each proof requires a computational resource like a polynomial commitment. The larger the batch, the more memory and time needed. For a proof of 1,000 transactions, the memory footprint can exceed 200 GB. Cloud instances with that capacity cost $1.50 to $2.00 per hour. Assuming 10 batches per hour, that is $0.15 per batch. Multiply by 24 hours: $3.60 per day in variable cost.
Total daily cost: $68.60 to $113.60. Daily transaction volume: on a typical ZK-rollup, that ranges from 50,000 to 200,000 transactions. Proving cost per transaction: $0.00034 to $0.00227. That seems low, but wait—the real cost is not the proof generation itself. It is the opportunity cost of capital tied up in hardware, the risk of obsolescence, and the inability to scale down during low activity.
Yields are just narratives with interest rates. The narrative that ZK-rollups are the ultimate scaling solution has ignored the capital intensity of the prover layer. In a bull market, operators could afford to run at a loss because token prices were rising. In a bear market, every dollar spent on proving is a dollar taken from the treasury. The math does not work unless gas prices return to 50+ gwei or volume increases 10x.
Filtering the noise to find the art. The open secret is that most ZK-rollup operators are currently subsidizing their proving costs through token inflation. They are selling future value to pay for present costs. This is not sustainable once inflation slows. The data from on-chain fees shows that the average revenue per transaction on these networks is less than $0.01. When proving costs are $0.002, the margin is positive—but only if volume is high. During low-traffic periods, the fixed costs dominate.
Let me use a concrete example. Scroll's mainnet has averaged 150,000 transactions per day over the past month. At $0.002 per transaction in proving cost, that's $300 per day. But their daily revenue from L1 data posting and sequencer fees is about $450. That leaves $150 for other costs (development, marketing, infrastructure). That is razor-thin. Compare to Arbitrum, an optimistic rollup with no proving cost, which generates $1,200 per day in revenue. The ZK-rollup's cost advantage in gas is offset by the proving overhead.
Contrarian: The Proof Aggregation Fallacy
The conventional wisdom is that proof aggregation—combining multiple ZK proofs into one—will solve the cost problem. Projects like Polyhedra and Nexus are building aggregation layers that promise to reduce proving costs by 90%. The contrarian angle: aggregation adds latency and complexity. It introduces a new attack surface: if the aggregator fails, the entire batch is stuck. Moreover, the aggregated proof still requires the same cryptographic work, just done once. The cost savings are linear, not exponential.
Arbitrage is the market’s way of correcting itself. If ZK-rollup proving costs are too high, the market will shift to optimistic rollups or sidechains. We are already seeing that. Base and Arbitrum have captured more TVL and transaction volume than all ZK-rollups combined. The market is voting with its feet. The ZK narrative is strong, but the economics are weak. The only way to fix this is to either dramatically increase transaction volume (which requires a bull market) or reduce proving costs through hardware innovation.
The second contrarian point: the proving cost is not just money—it is time. Latency matters. For a DEX user, a 5-minute confirmation is unacceptable. ZK-rollups have to wait for the proof to be generated before finality. That wait time is a product feature, not a bug. But in a bear market, users are less tolerant of delays. They want speed and low fees. Optimistic rollups offer faster confirmation (within seconds) and lower operational costs. The ZK-rollup's value proposition—provably correct execution—only matters when the market cares about security theater. In a bear market, survival matters more than guarantees.
Takeaway: The Next Narrative Shift
Storytelling is the new consensus mechanism. The ZK-rollup narrative is running out of steam. The next phase will be about proving infrastructure as a service—decoupling the prover from the rollup. Imagine a marketplace where anyone can rent proving power, with competitive pricing. That is the only way to reduce costs below the current floor. Projects like Espresso are building shared sequencing, but they have not solved the proving cost problem.
Efficiency is the enemy of the outlier. The outlier in this bear market will be the protocol that finds a way to make proving costs zero. That might be through recursive proofs, or through off-chain computation that never needs to be verified on-chain. But that is a different architecture entirely.
The code does not lie, but it is incomplete. The financial models are incomplete. Until the market recovers, ZK-rollup operators will continue to bleed. The ones with the deepest treasuries will survive. The others will pivot to becoming optimistic rollups or shut down. The signal is clear: proving costs are the bottleneck. The noise is the hype around ZK. Filter the noise, find the art.
Final thought: The next bull run will not be about ZK-rollups versus optimistic rollups. It will be about which scaling solution can offer the lowest cost per transaction at scale. Right now, the answer is not a rollup at all—it is a sidechain like Solana or a monolithic L1. The ZK-rollup has a proving cost problem. And until that problem is solved, it remains a niche product for institutional use cases, not mass adoption.