Over the past 30 days, data from L2BEAT shows that the six largest ZK rollups spent $2.1 million on proof generation while collecting only $1.3 million in user fees. That is a 38% gross loss on every transaction. I pulled the exact numbers from the StarkNet block explorer: average verification gas per batch sits at 450K, while median user fees per transaction have dropped to $0.02. The ratio is inverted.
For context, ZK rollups were sold as the holy grail of scaling during the 2021-2022 bull run. The pitch: infinite throughput, instant finality, and Ethereum-level security. Projects like zkSync Era, Scroll, and StarkNet raised billions in valuations on that narrative. Fast forward to 2026: Ethereum gas is under 5 gwei, and the premium for ZK’s “validity proof” has evaporated. The infrastructure that was built for a $4,000 ETH world is now operating at a fraction of capacity.
Let me break down the cost structure. A ZK proof consists of two parts: off-chain proving (computation) and on-chain verification (gas). Off-chain proving requires expensive hardware — typically a cluster of GPUs or specialized ASICs. I deployed a simple Groth16 prover on a 4x RTX 4090 rig last year. For a single token transfer, the proving time was 12 seconds, and the electricity cost alone was $0.08. On-chain, the verification contract burns around 300,000 gas per call. At current 5 gwei, that’s $0.015 per verification. But the off-chip custom hardware for large batches? That scales non-linearly. StarkWare’s SHARP prover, for example, processes batches of 1,000+ transactions. The marginal proving cost per transaction drops to about $0.01, but the fixed infrastructure cost — maintenance, cooling, decentralization overhead — adds another $0.04 per tx. Net cost: $0.065 per tx. Net revenue: $0.02 per tx. Negative margin.
Contrast this with optimistic rollups like Arbitrum and Optimism. Their cost per transaction is roughly $0.005 — driven by simple calldata posting and a fraud proof window that rarely triggers. The irony is that optimistics are now the “boring” infrastructure that pays. Code doesn’t lie, but markets do: the current data shows that ZK rollups are subsidizing every transaction with investor capital. When the bear market drags on, that capital dries up.
The contrarian angle: retail investors still believe ZK is the future because of the “tech cool” factor. But smart money is rotating. Look at the TVL distribution over the past six months: Arbitrum’s TVL held steady at $3.8B, while zkSync Era’s dropped from $1.2B to $480M. That’s a 60% outflow. Why? Because liquidity providers chase yield, and ZK rollups can’t offer competitive returns when their operational costs are eating into protocol revenue. Volatility is just unpriced risk — the market is finally pricing the risk that ZK may never reach the promised efficiency without a bull market revival.
My own experience reinforces this. During the 2022 Terra collapse, I manually traced LUNA/UST decimals block by block. That forensic exercise taught me that infrastructure failures are rarely sudden; they are slow leaks in cost curves. Today, the same pattern appears in ZK rollups. I’ve audited three ZK verification contracts in the past year, and each had gas inefficiencies that added 10-15% overhead. One client used a Solidity loop to verify multiple proofs instead of batching — a rookie mistake that cost them $200K in gas over three months. I flagged it, they fixed it, but the underlying economics remain unworkable.
Infrastructure outlasts innovation — that line applies here in reverse. The innovation (ZK) is bleeding capital now, while the older infrastructure (optimistic) is profitable. For developers: if you are building a dApp today, choose Arbitrum or Optimism. You get lower costs, proven security, and a wider user base. The ZK teams are burning cash to attract you with grants, but grants don’t pay for your users’ gas. For investors: check the monthly burn rate of any ZK rollup before allocating. If operating expenses exceed protocol revenue by more than 30%, that’s a red flag. I don’t predict, I react — and right now, the market is reacting by moving liquidity back to optimism.
Takeaway: The next 12 months will separate ZK rollups that can achieve positive unit economics from those that survive on VC life support. Watch the prover hardware costs: if a breakthrough in proof aggregation drops costs by 10x, the thesis resets. Until then, the contrarian trade is to short ZK tokens and long OP/ARB. Liquidity is the only truth, and it’s flowing downstream.