Over the past seven days, the average gas price on Ethereum has hovered around 8 gwei. That’s sluggish. Dead. Yet ZK rollup operators are still spending millions on proving costs.
Here’s the number that should make you pause: $1.2 million per month for a single ZK-rollup sequencer to generate proofs at current gas levels. That’s not a typo. That’s the real cost of running a zero-knowledge proof system when the network isn’t congested.
I’ve been auditing smart contracts since 2016. I traced the DAO reentrancy exploit before the hard fork. So when I see a protocol burning capital on a fixed cost that doesn’t scale with revenue, I smell a structural flaw. This isn’t a narrative problem. This is a math problem.
— Root: Auditing the DAO and Ethereum
—
Let’s rewind. ZK rollups work by bundling transactions off-chain, generating a cryptographic proof (a zk-SNARK or zk-STARK), and posting that proof to Ethereum. The proof is small—about 200-400 bytes. But the computation to generate it is enormous. A single proof can require thousands of CPU hours on specialized hardware.
The cost structure is inverted. Traditional rollups (like Optimism) pay a fixed L1 data cost per batch. ZK rollups pay a fixed proving cost per batch, plus L1 data cost. The proving cost doesn’t drop when the network is quiet. It stays constant.
When gas was 100 gwei in 2021, the L1 data cost dominated. Proving was a rounding error. Now? Gas is low. L1 data costs are trivial. But proving costs remain stubbornly high because the hardware and electricity are priced in dollars, not gas.
I’ve built automated yield farming bots. I’ve run copy trading communities. I know how to spot a hidden liability. The proving cost was the liability everyone ignored during the bull run.
—
Here’s the core technical breakdown. Let’s use a real case: a major ZK-rollup (I’ll keep the name out for now, but the data is public).
In Q1 2024, this rollup processed 1.5 million transactions per day. Each batch of 10,000 transactions required one proof generation. The hardware cost per proof: approximately $150 (using AWS p3.16xlarge instances, 8 V100 GPUs, running for 20 minutes). That’s $0.015 per transaction in proving cost alone. Add L1 data cost: at 8 gwei, each batch costs about $0.50 in L1 gas. That’s $0.00005 per transaction.
Total cost per transaction: $0.01505. The rollup’s revenue per transaction? Fees—usually less than $0.01 for most users. But the rollup doesn’t keep all fees. It pays L1 gas, and the proving cost is a loss leader subsidized by the team.
Wait—it gets worse. The proving cost is amortized over the batch. If the batch size drops (e.g., during a quiet period like a weekend), the cost per transaction increases. On a low-volume Saturday, batch size can drop to 2,000 transactions. Proving cost per transaction jumps to $0.075. The rollup is now losing $0.065 per transaction.
— Root: Auditing the DAO and Ethereum
This is the hidden scalability crisis. ZK rollups are supposed to be the endgame. But the proving cost creates a fixed overhead that doesn’t scale with volume. It scales with proof generation frequency, which is roughly constant. Double the volume, double the number of batches, double the proving cost. There’s no economy of scale in proof generation when you’re constrained by latency requirements.
—
Now, the contrarian angle. The market narrative says ZK rollups are the future because they are trustless, final, and secure. And they are. But the economics are broken unless gas prices return to 50+ gwei or transaction volumes explode 10x.
Retail investors look at daily active users and TVL. They see the shiny experiment. Smart money looks at the burn rate. They see a team subsidizing every transaction by $0.05, hoping that volume will eventually cover the subsidy. But volume growth is linear. Proving costs are linear. There’s no magic lever.
Some rollups are trying to cheat. They compress proofs, aggregate multiple batches into one, or use recursive proofs to reduce cost. But these are optimizations, not fundamental fixes. The base cost of generating a proof—the hardware, the electricity, the time—is bounded by the security assumption. You can’t compress a proof into five seconds without sacrificing the very properties that make it trustless.
I’ve seen this before. In 2022, I shorted Luna because I verified the lack of cryptographic reserves. The same pattern: a flawed economic model dressed up as a technological breakthrough. The Terra collapse was a code failure, but the trigger was a capital structure failure. ZK rollups have a capital structure failure: proving costs are a fixed liability that grows with adoption, not revenue.
—
So what’s the takeaway? Actionable price levels. Track the gas price on Ethereum. If it stays below 10 gwei for three more months, expect ZK-rollup teams to start consolidating, merging, or cutting corners. Lower proving costs will force layoffs or protocol changes. If gas rises above 30 gwei, the math flips. L1 data costs dominate again, and proving costs become a smaller fraction of total cost. The sweet spot for ZK rollup viability is 20-30 gwei average gas over a rolling quarter.
Watch the batch frequency. If a rollup reduces its batch frequency (e.g., from once per minute to once per 10 minutes), it’s a sign they’re trying to amortize proving costs. That increases latency. Heavy users will notice. That’s a signal to exit.
—
I’m not saying ZK rollups are dead. I’m saying the current economic model is unsustainable at current gas prices. The market is pricing in a bullish future—proving costs will drop due to hardware improvements, or gas will rise. But hardware improvements are incremental. ASICs for proof generation are still years away. Gas is a function of Ethereum activity, which is mostly L2 activity anyway. It’s circular.
We farmed the yields until the protocol farmed us.
— Root: Auditing the DAO and Ethereum
So here’s my open question: How long will VCs keep subsidizing proving costs before the music stops? If you’re running a portfolio, don’t just look at TVL. Look at the proving cost per transaction. Divide the operating expenses by the transaction count. If that number is higher than the net fee per transaction, you’re holding a liability.
Chop markets are for positioning. Position yourself short ZK-rollup tokens that are overvalued relative to their burn rate. Position yourself long the infrastructure that actually scales—like simple data availability layers that don’t require proving. The math doesn’t lie. Code doesn’t care about your narrative.
— Root: Auditing the DAO and Ethereum
