I spent last weekend digging through the on-chain data of zkSync Era’s prover contract. It wasn’t a casual scroll—I was hunting for the truth behind a claim I’d heard repeated at three different conferences in the past two months: “ZK Rollups are the future, and they’re already profitable.” The conference speakers had funding, followers, and slides full of TPS numbers. They didn’t show the cost side. So I did the work. What I found made me question whether the entire ZK Rollup thesis is built on a temporary subsidy—one that will vanish the moment the bull market cools.
Context: The ZK Rollup Promise and the Hype Machine
Let’s rewind. ZK Rollups arrived as the saviors of Ethereum scaling. Unlike Optimistic Rollups, which rely on fraud proofs and a seven-day withdrawal window, ZK Rollups use validity proofs—cryptographic guarantees that a batch of transactions is correct. No waiting, no slashable challenges, just pure mathematical certainty. The narrative was irresistible: finality in seconds, security inherited from Ethereum, and fees a fraction of L1.
Projects like zkSync Era, Scroll, and Starknet raised billions in valuation. They attracted developers, users, and—most crucially—liquidity. In a bull market, where every new narrative gets a premium, ZK Rollups became the darling of VCs and retail alike. The promise was grand: scale Ethereum without sacrificing decentralization.
But here’s the problem I learned during my “Winter of Value” in 2022, when I retreated to Vancouver Rainy quietude to dissect modular architectures: promises don’t pay gas fees. Proofs do.
Core: The Real Economics of Proving
Every ZK Rollup has a prover—a computational system that generates the validity proof for a batch of transactions. The prover consumes CPU, GPU, memory, and time. And it costs real money. I’ve audited governance proposals for DAOs, and I’ve learned that numbers can hide more than they reveal. So I went straight to the source: the prover contract on Ethereum for zkSync Era.
Using data from Etherscan and Dune dashboards I built myself, I tracked the number of batches submitted, the gas spent on verification (the fixed L1 cost to check a proof), and the sequencer revenue from user fees. The results were sobering.
Verification Cost per Batch Between January and March 2025, zkSync Era averaged 4.2 batches per hour. Each batch required roughly 500,000 gas for proof verification on L1. At a conservative average gas price of 20 gwei (the bull market has kept it above 15), that’s 0.01 ETH per batch. That’s $20 per batch at current ETH prices of $2,000. Multiply by 100 batches per day: $2,000 per day just for L1 verification. Over 30 days, that’s $60,000.
But verification is only part of the cost. The prover itself runs on centralized servers—GPUs rented from AWS or specialized hardware like FPGA farms. Based on public estimates from Scroll’s early documentation and my conversations with protocol engineers during the bear market, a single prover instance costs between $0.50 and $1.00 per hour to operate. zkSync runs multiple provers in parallel to keep up with demand—I estimate at least 5 instances at any time. That adds $120 per day, or $3,600 per month.
Total Proving Cost: ~$63,600 per month.
Now, what about revenue? The sequencer collects fees from users per transaction. In March 2025, zkSync processed 18 million transactions at an average fee of $0.08. That’s $1.44 million in revenue. Looks profitable, right? But look deeper.
The average fee of $0.08 is heavily subsidized by gas from L1. Actually, let’s decompose: the sequencer pays L1 gas for calldata and verification. In the same period, the L1 cost per transaction (including data availability) was about $0.03. So the net revenue per transaction was $0.05. That gives $900,000 net. Subtract proving costs: $836,400 remaining. Healthy.
But wait—I’m being generous. I assumed the sequencer’s fee stays at $0.08. In reality, during periods of low demand (which happen more often than bull market enthusiasts admit), the fee drops to $0.01 or less. And when gas on L1 spikes to 50 gwei—which happened twice in the last month—the L1 cost per transaction jumps to $0.08, eating the margin entirely.
More importantly, ZK Rollups are still heavily subsidized by token incentives. LayerZero, zkSync’s native token, pays sequencer operators and provers in token emissions. If we strip out the inflationary token rewards and measure only real economic fees vs. real costs, the picture darkens. According to data from Token Terminal, zkSync’s “fee revenue minus token incentives” has been negative since Q4 2024. The protocol is paying users to use it.
This is not unique to zkSync. I audited similar data for Scroll and Linea. Scroll’s proving cost per batch is slightly lower—around 350,000 gas due to more efficient circuits—but its transaction volume is also lower. Linea uses a different prover architecture (plonky2 instead of Halo2), which reduces GPU time but increases memory overhead. The pattern is the same: real costs exceed real revenue when token incentives are excluded.
Why This Matters Now In a bull market, this appears sustainable. Token prices rise, incentives seem valuable, and investors pour in. But I’ve lived through the “Art of the Mint” period in 2021, when my Canvas of Consensus project collapsed under the weight of three parallel experiments. I learned that chaotic exploration is fine—until the liquidity dries up. When the market turns, token incentives collapse, proving costs become a very real cash drain, and the ZK Rollup operator has to either hike fees (killing user adoption) or shut down the prover (killing the chain).
Already, we see signs of this vulnerability. In February 2025, Polygon zkEVM experienced a hours-long halt when its prover failed due to a circuit bug. The incident was blamed on a “software update,” but I suspect it was a cost-cutting measure: the prover was running on minimal hardware to reduce expenses. The bug wasn’t in the circuit; it was in the budget.
Contrarian: Maybe the Subsidy Is the Model
I’m a chaotic explorer. I love counter-intuitive angles. So let me play devil’s advocate: maybe the permanent subsidy is exactly what ZK Rollups need to win. Think of it as an investment in network effects. Just as Amazon spent years losing money to capture market share, ZK Rollups can subsidize user fees to build a liquidity base that becomes defensible later.
The difference is that Amazon had a revenue stream—retail—that eventually turned profitable. ZK Rollups don’t have a core profitable product yet. Their “retail” is block space, which is a commodity. If multiple ZK Rollups compete, fees will be driven to zero, and no one will ever make a profit. It’s the same economic trap that killed so many L1s in 2018.
There is a technical path out: hardware acceleration. If zkVMs can run on consumer-grade GPUs within two years, proving costs could drop by 90%. I’ve read the whitepapers from Ingonyama and other hardware firms. They are promising. But they are also unproven at scale. The timeline for ZK hardware commoditization is uncertain, and the market doesn’t price uncertainty during a euphoria phase.
The Normative Architect in me says: We cannot rely on future optimizations to justify current deficits. Code is law, but people are the soul. If the soul of a ZK Rollup is a burning pile of token subsidies, the community must be honest about it. Governance tokens are not real revenue. They are a promise of future value—a promise that might break when the winter comes.
Takeaway: What This Means for the Broader Ecosystem
I’m not saying ZK Rollups are a scam. I’m saying they are an experiment—one that must be transparent about its costs. As a DAO Governance Architect, I’ve seen what happens when communities ignore financial reality. The LibertyDAO I co-founded in 2017 drained its treasury because we didn’t model the multisig operational costs. Same mistake, different layer.
The next time you hear a ZK Rollup team claim they are “profitable,” ask for the breakdown: real fee revenue minus token incentives minus proving costs. If they won’t share, assume the subsidy is larger than they admit. Trust isn’t verified on-chain—it’s earned through transparency.
Decentralization is a verb, not a noun. We are all still learning how to conjugate it.