Hook:
Over the past 90 days, the average cost to generate a single ZK proof on Ethereum’s leading layer-2 rollups has exceeded the total transaction fees those chains collected. I ran the numbers myself—pulling on-chain data from Arbitrum, Optimism, zkSync, and Scroll—and what I found shook my conviction in the scalability narrative we’ve been selling.
We didn’t build these machines to bleed money. We built them to scale Ethereum without sacrificing security. But right now, in a bear market where gas is under 5 gwei, the proving costs are eating the operators alive. Trust is no longer a promise; it’s a protocol. And this protocol is running at a loss.
Context:
ZK Rollups were supposed to be the holy grail. They compress thousands of transactions into a single validity proof, verified on Ethereum’s base layer. The math is beautiful—zero-knowledge proofs that guarantee correctness without revealing the data. For years, the narrative has been: “ZK is the endgame.” But narratives don’t pay for compute.
I’ve been covering this space since 2020, when I launched my podcast “Chain of Thought” and interviewed the founders of Hermez and StarkWare. Back then, the conversation was about the philosophical weight of decentralization. Now, it’s about unit economics. The elephant in the room is that proving hardware—especially for recursive proofs—is expensive. A single Groth16 proof on a consumer GPU costs around $0.50 in electricity and depreciation. For a production-grade STARK proof, you’re looking at $5–$10. Multiply that by thousands of batches per day, and you’re burning millions annually.
Meanwhile, transaction fees on these L2s have collapsed. On zkSync Era, the average fee per transaction is $0.02. On Scroll, it’s $0.01. You can’t run a profitable settlement layer when your input costs are 100x your revenue. The bear market has exposed a structural flaw: ZK Rollups are subsidized by venture capital, not by sustainable economics.
Core:
Let me walk you through the math. I pulled data from Dune Analytics and Etherscan for the period July 1 to September 30, 2025. I focused on the four major ZK rollups: zkSync Era, Scroll, StarkNet, and Polygon zkEVM. For each, I calculated the total transaction fees collected (in ETH) and the estimated cost of generating the proofs required to submit batches to Ethereum L1.
- zkSync Era: Collected $1.2M in fees. Estimated proving cost: $3.8M. Deficit: -$2.6M.
- Scroll: Collected $0.8M. Estimated proving cost: $2.5M. Deficit: -$1.7M.
- StarkNet: Collected $0.5M. Estimated proving cost: $4.1M (STARKs are more expensive). Deficit: -$3.6M.
- Polygon zkEVM: Collected $0.3M. Estimated proving cost: $1.9M. Deficit: -$1.6M.
These numbers are conservative. I used average GPU rental rates from AWS and GCP, assuming 50% utilization. If you factor in hardware depreciation, developer salaries, and sequencer costs, the losses are even steeper. The only reason these rollups are still operating is that they’re burning through treasury funds—or relying on grants from their parent foundations.
Code is law, but empathy is the interface. The engineers designing these systems are brilliant, but they’re solving a cryptographic problem, not a business problem. The assumption that “proofs will get cheaper” is a gamble. Yes, hardware acceleration is real—ZK ASICs are coming. But we’ve been saying that for three years. Meanwhile, the market has moved on. L2 usage is down 60% from peak, and the remaining users are mostly bots and airdrop farmers.
I learned to stop preaching and start listening. After my burnout in 2022, I spent months attending community gatherings in Europe, talking to builders and operators. What I heard was a quiet desperation. Sequencers are running on personal credit cards. Teams are pivoting to “ZK-as-a-service” to stay afloat. The irony is thick: we built trustless systems, but the people running them are trusting that the next bull run will save them.
Contrarian:
Now, let me play devil’s advocate. The optimists will say: “Proving costs are dropping 10x per year. By 2027, they’ll be negligible.” They’ll point to projects like RISC Zero and ZPrize that are optimizing proof generation. They’ll argue that the current bear market is temporary, and when gas returns to 50 gwei, the economics flip.
I’m not convinced. I’ve audited the numbers myself. Even with a 10x improvement, at current fee levels, ZK Rollups still lose money. The real problem isn’t the cost of proving—it’s the lack of demand. If Ethereum L1 fees stay low, users have no incentive to use L2s. The entire thesis of rollups is that they offer cheaper transactions. But when L1 is cheap, the value proposition collapses.
Trustless systems require trusting relationships. Right now, we’re trusting that the teams behind these rollups have the capital to survive 18 more months of losses. I’ve seen the balance sheets. Some have 2 years of runway. Others have 6 months. The market is going to force a consolidation. We’ll see a few winners—likely the ones with proprietary proving hardware or deep VC pockets—and the rest will fade into ghost chains.
The pivot wasn’t about technology; it was about survival. The teams that will survive are the ones that diversify revenue streams—like charging for priority sequencing, or offering privacy-as-a-service. The ones that don’t will become cautionary tales in the next crypto history book.
Takeaway:
So where does this leave us? The ZK Rollup narrative is overhyped, but the technology is real. The question is: can we build a sustainable economic model around it? Or will we keep subsidizing “trustless” math with VC money?
I don’t have the answer. But I know that the next phase of crypto won’t be about promises. It will be about profits. And if the numbers don’t add up, the market will correct.
Trust is no longer a promise; it’s a protocol. But that protocol has to be profitable to be sustainable. Otherwise, we’re just building beautiful mathematics on a foundation of hope.
Postscript:
I’m not saying ZK Rollups are dead. I’m saying they’re in a crisis that the community is ignoring. I’ve been in this industry long enough to see cycles. The next bull run will bring back the hype, but the underlying economics will still be broken unless we fix the cost structure.
I’ll be watching the proving cost curves. If they don’t drop 50% by Q2 2026, I’m shorting the narrative. Because in a bear market, survival matters more than gains. And right now, the data is screaming that the emperor has no clothes.