Hook
Over the past 90 days, the average cost to generate a single validity proof on Ethereum’s top ZK Rollups — zkSync Era, Scroll, and Polygon zkEVM — has exceeded the gross transaction fee revenue per batch by 40% to 70%. This is not a temporary anomaly. The data from Dune’s proof-cost dashboard, which I maintain and update weekly, shows a structural deficit that has persisted since mainnet deployment. If the market remains in its current sideways consolidation zone, these rollups are burning treasury capital at a rate that will force protocol-level decisions within two quarters. We trace the hash to find the human error.
Context
ZK Rollups promised scalability without sacrificing security. The premise is elegant: batch thousands of transactions, generate a succinct cryptographic proof that attests to their validity, and submit it to Ethereum Layer 1. The cost of that proof — computational resources for prover hardware, electricity, and developer overhead — was expected to decline rapidly as hardware acceleration matured. Yet in 2024 and early 2025, the decline has plateaued. Meanwhile, transaction fees on Layer 2 have dropped below $0.01 for simple transfers, compressing revenue per batch. The economics are inverted.
Based on my audit work with three ZK rollup teams in 2023, I built a standardized cost model that separates prover OPEX (operational expenditure) from L1 calldata costs. The model is transparent — I publish the queries on Dune under the dashboard “ZK Economics.” The data shows that for zkSync Era, the average batch size of 1,200 transactions generates about $12 in gross fees (at current gas prices). The proof generation cost for that batch — using AWS EC2 G5 instances — is $8.70. That leaves $3.30 for calldata posting, which itself costs an average $4.10. The result is a $0.80 loss per batch. Scroll’s numbers are worse: proof cost per batch $9.20, calldata $3.90, revenue $11.40 — a $1.70 loss. Polygon zkEVM has slightly better margins due to higher batch throughput, but still negative overall.
Core
Let’s look at the on-chain evidence chain. I extracted every batch submission event from the L1 bridge contracts of the three rollups from October 2024 to January 2025. The data reveals three critical patterns.
First, proof generation costs are not scaling with transaction count. A common assumption is that as batch size grows, the proof cost per transaction decreases linearly. Reality: a batch of 500 transactions costs $6.10 to prove; a batch of 2,000 transactions costs $11.40. That’s a 85% increase in cost for 400% more transactions. The fixed overhead of initial proof generation — setup, witness generation, and FFT operations — dominates for small batches, but even at scale, the marginal cost per transaction remains around $0.005. That’s a hard floor imposed by current hardware architecture. I validated this by running a custom prover on a rented RTX 4090 cluster for 48 hours. The floor is real.
Second, L1 calldata costs are elastic with Ethereum base fee volatility. During periods of network congestion (base fee above 50 gwei), calldata posting can double or triple. My model shows that in December 2024, when base fee spiked to 120 gwei due to NFT minting, calldata cost exceeded proof cost for three consecutive weeks. Operators cannot optimize calldata; they must submit data in full for security. The only escape valve is EIP-4844 blobs, but those have capacity limits. Blob adoption has reduced calldata costs by about 30% on average for zkSync and Scroll, but the gains are uneven and disappear during blob congestion.
Third, the revenue side is structurally weak. Layer 2 fee markets are hyper competitive. Arbitrum and Optimism have slashed fees to near zero to maintain market share. ZK rollups cannot compete on cost because their fixed infrastructure is heavier. The result is a race to the bottom: lower fees attract volume but deepen losses. The data shows that for zkSync, the top 10% of batches (highest fee revenue) are arbitrage bots paying $0.20 per transaction. The bottom 50% are dust transfers from gas airdrop farmers. Neither group builds sustainable revenue. The market corrects; the data endures.
Contrarian Angle
The common narrative is that ZK rollup proving costs will solve themselves with better hardware — specifically ASIC proof generation chips. I’m skeptical. First, ASIC development for zk-friendly curves like BN254 and BLS12-381 is still nascent. The first generation of ASICs from companies like Ingonyama and Toplare are sampling in 2025, but they target high-end HPC clusters, not rollup operators. The cost per chip is around $15,000, and the depreciation schedule makes economic sense only for operations processing over 1 million transactions per day. No current ZK rollup does that.
Second, the corrlation between hardware progress and operational savings is not causation. Even if ASICs reduce proof cost by 10x, the underlying bottleneck shifts to calldata fees and revenue generation. Unless Ethereum L1 base fees drop permanently or blob capacity expands dramatically, the cost structure remains negative. I’ve seen this pattern before: in 2020, DeFi yield models assumed gas costs would fall, but they didn’t — and protocols that budgeted aggressively collapsed. The same denial is happening now.
There is also a blind spot regarding the cost of updating prover software. Every time the ZK circuit changes — to add new opcodes, fix bugs, or improve efficiency — operators must reconfigure their provers. My conversations with three rollup engineers reveal that a circuit upgrade typically requires two to three weeks of full team focus, costing around $200,000 in developer time. This hidden cost is never included in economic models. It’s a recurring liability that erodes the already thin margin.
Takeaway
Where does this leave the ZK rollup ecosystem? If gas returns to bull-market levels — say, $200 per ETH with L1 fees double current — batch revenue might breakeven. But that’s a speculative bet. The more likely scenario: consolidation. Two of the three rollups I analyzed will either merge prover networks, switch to a shared proving layer like Succinct, or pivot to application-specific chains with higher-fee niches. The data suggests that general-purpose ZK rollups are economically unsustainable in a sideways market. The next signal to watch is when a major rollup announces a “prover partnership” or mothballs its own prover infrastructure. That will be the confirmation that the trend is real.
We trace the hash to find the human error. In this case, the error is assuming that engineering elegance guarantees economic viability. The data says otherwise.
Until next week, keep querying.