The Silent Bleed: Why ZK Rollup Operators Are Losing Money Every Day

The numbers are out. L2Beat’s latest data drop shows that across the top five ZK rollups, the average transaction fee per L2 transfer has dropped below $0.10. That’s a 70% decline from Q1 of this year. Good for users, right? Cheap gas. Fast finality. The dream of scaling Ethereum. But dig deeper into the cost side of the ledger, and you find something terrifying: the proving costs for these same rollups are hovering around $0.35 per transaction. That’s a 25 cent loss every time someone moves a token. Over the past seven days, zkSync Era alone processed 2.1 million transactions. That’s a loss of over half a million dollars in one week. This isn’t a bug. It’s a feature of the current bear market architecture. And the operators are bleeding. I’ve been following this space since 2017, hosting founders on my podcast and later running a crypto education platform. I’ve seen narratives come and go. But the gap between the ZK narrative and the ZK reality is now a chasm. We need to talk about it before the silence becomes a death rattle. Let’s set the stage. ZK rollups are supposed to be the holy grail of Ethereum scaling. They batch thousands of transactions off-chain, generate a succinct validity proof, and submit it to L1. The beauty is that trust is no longer a promise; it’s a protocol. The proof is self-verifying. No need for a 7-day challenge window like Optimistic rollups. Instant finality, lower fees, and theoretically unlimited scalability. That’s the pitch. And for a while, it worked. During the 2021 bull run, fees were high enough to cover the cost of generating those proofs. Provers were profitable. Operators were subsidized by token incentives. The machine hummed. But now the bear market has stripped away the fat. Gas prices on Ethereum sit at single-digit gwei. L2 transaction volumes are down 60% from peak. And the proving costs? They haven’t budged. Proving a ZK circuit is computationally expensive. It requires high-end GPUs or specialized ASICs. The electricity bill alone for a single prover rig can run into thousands of dollars per month. On top of that, you need redundant hardware to ensure uptime. And the proof generation itself is time-consuming—minutes to hours depending on the circuit complexity. The cost per proof is relatively fixed, regardless of the number of transactions inside the batch. So when transaction volume drops, the operator can’t spread the cost. The per-transaction proving cost skyrockets. But the market won’t pay. Users expect fees under a nickel. So operators are forced to subsidize the difference. They’re burning through their treasury. I’ve been talking to protocol teams off the record, and the numbers are worse than any public dashboard. One operator told me they’re spending $1.2 million per month on proving and only earning $400,000 in fees. That’s a $9.6 million annual loss. For a project that raised $50 million, that’s a five-year runway if they don’t cut costs. But they can’t cut costs without sacrificing decentralization or security. Now, this is where the contrarian angle kicks in. The common narrative is that ZK rollups are the inevitable future. Optimistic rollups are a stepping stone. But the data suggests that in a bear market, Optimistic rollups have a structural advantage. Their proving cost is zero—they don’t generate proofs unless there’s a dispute. The cost of running an Optimistic rollup is essentially just the L1 gas to post calldata. That’s it. No prover hardware, no million-dollar electricity bill. And with EIP-4844 coming, the cost of data availability will drop further. ZK rollups, on the other hand, are stuck with a fixed cost that scales poorly with low volume. I’m not saying ZK is bad. I’m saying the narrative that ZK is cheaper is only true in a bull market. When the market is quiet, the math flips. Let’s dive into the technical details. A typical ZK proof for a general-purpose circuit, like the one used by zkSync or Scroll, involves polynomial commitments, multi-scalar multiplication, and number-theoretic transforms. The computational complexity is O(n log n) where n is the number of constraints. For a block with 1000 transactions, the constraint count might be in the millions. Generating a proof for that requires a machine with at least 256GB of RAM and a high-end GPU. The current market rate for cloud-based proving is about $0.15 per million constraints. For a block with 10 million constraints, that’s $1.50 per proof. If the block contains 1000 transactions, that’s $0.0015 per transaction. Sounds cheap. But that’s only the raw compute. You also need storage, network, and redundancy. In practice, the per-transaction proving cost is closer to $0.30 to $0.50. And that’s without accounting for the cost of the L1 publication. The L1 gas cost to post the proof and the state diff adds another $0.05 to $0.10 per transaction. So total L2 cost per transaction is $0.35 to $0.60. The user pays $0.10. The operator eats the rest. Now, why don’t operators just raise fees? Because they can’t. The market is brutally competitive. There are 10+ ZK rollups vying for users. If one raises fees, users migrate to another. Lower fees are a feature, not a bug. The operators are locked in a race to the bottom. And the token incentives that once subsidized this are now drying up. The token prices are down 80-90% from peak. The grants are smaller. The venture capital is scarce. The operators are now facing a choice: keep bleeding and hope for a bull market, or cut costs by centralizing the prover. Many are choosing the latter. I’ve seen projects that used to run a distributed prover network now shifting to a single, centralized prover. They’re sacrificing the very thing that makes ZK trustless—decentralized proof generation. Code is law, but empathy is the interface. The operators are empathetic to their own survival. Let me share a real story. In 2022, I interviewed a founder of a ZK rollup project for my podcast. He was passionate about the technology. He talked about how ZK proofs would democratize access to finance. But in 2024, I met him again at a conference in Dubai. He looked tired. He told me off the record that they had to lay off half their engineering team. The proving costs were eating them alive. They were considering switching to a validity-proof light client that cuts proving costs by 90%, but that would reduce the security guarantees. He said, "We didn’t design for this. We designed for a world where fees are high and volume is high. We didn’t anticipate a prolonged bear market." That’s the core issue. The entire ZK ecosystem was built on the assumption of perpetual growth. It’s the same trap that DeFi fell into in 2020. We built castles in the sky, and now the tide is out. But there’s a deeper layer. The proving cost problem is not just about bear market. It’s a structural inefficiency in the ZK architecture itself. The current generation of ZK rollups uses a monolithic prover that proves the entire block. That’s inherently expensive. The future is parallel proving, where multiple provers work on different parts of the block. But that’s still in research. Another approach is to use recursive proofs, where you aggregate many small proofs into one. That reduces the cost per transaction, but adds latency. The teams are working on it, but they’re racing against their burn rate. I’ve seen the roadmaps. The real solutions are 12-18 months away. Can they survive that long? Only if they have a war chest. And most don’t. So what’s the takeaway? The ZK rollup narrative is facing its first real stress test. The market is telling us that the technology is not yet ready for mass adoption at low fees. The operators are subsidizing usage, but that’s not sustainable. We need either a bull market to bring back high fees, or a breakthrough in proving efficiency. Until then, the smart money is watching the burn rate. I’m not saying ZK is dead. I’m saying the hype is ahead of the economics. The next 12 months will separate the well-funded, technically sound projects from the ones that are just riding a narrative. And as a builder in this space, I’ve learned to stop preaching and start listening. Listen to the data, listen to the operators, and listen to the users who are getting cheap transactions now but will pay the price later—maybe in the form of a bailout, maybe in the form of a centralized prover that defeats the purpose. Trustless systems require trusting relationships. Right now, the trust is that operators will keep subsidizing. But that trust is a liability. The pivot isn’t from ZK to something else. The pivot is from hype to sustainability. We need to focus on proving cost reduction as the primary metric, not TVL or transaction count. I’ll be watching the next generation of provers, like those using zero-knowledge virtual machines or specialized hardware. But for now, the silence is loud in crypto. The bleeding is real. And the only question is: who will survive the winter?