The High Cost of Proving: Why ZK Rollups Are Failing the Trustless Promise
Hook
On July 5, 2025, Donald Trump stood before a crowd at the Lincoln Memorial, fists raised, telling the world that America was “stronger than ever.” The flyover was “unprecedented.” The love for country had never burned brighter. But as I watched the livestream from my small apartment in Stockholm, coffee in hand, I couldn’t shake the feeling that this was a show of strength built on borrowed time. No policy. No budget. No real data. Just a beautiful spectacle.
We didn’t need to look far to see the same script playing out in crypto. Every week, a new ZK rollup announces its mainnet launch, trumpeting “trustless scaling” and “infinite throughput.” The conference speeches are grand, the GitHub stars are high, and the whitepapers are thick with mathematical elegance. But behind the scenes, the cost of proving those zero-knowledge proofs is bleeding operators dry. This is not a sustainable system—it’s a show of strength that collapses when the music stops.
I learned this the hard way. In 2022, I spent three months running a small ZK rollup node for a testnet. The electricity bill alone made me question my life choices. And that was just for fun. Today, in the bear market, the economics are brutal. Let me walk you through the numbers, the narratives, and the uncomfortable truth that most people are too polite to say out loud.
Context
Zero-knowledge rollups are the darlings of the scaling narrative. They promise to bundle thousands of transactions off-chain and submit a single validity proof to Ethereum, inheriting its security while slashing costs. The technology is beautiful—pure math, cryptographic composability, and the dream of a truly trustless second layer. Projects like zkSync, StarkNet, Scroll, and Polygon zkEVM have raised hundreds of millions of dollars. The community rallies behind them: “zkR is the endgame.”
But here’s the problem: proving a ZK proof is computationally expensive. Really expensive. A single validity proof for a batch of transactions can require hours of GPU time, sometimes costing hundreds of dollars in compute resources. And this cost scales with the complexity of the smart contracts being verified. When Ethereum gas is cheap (under 10 gwei), the proof cost can eat up the savings. When gas spikes to 50 gwei—as it did during the DeFi mini-boom in early 2024—operators either absorb the loss or pass it to users, destroying the value proposition.
I’ve seen this up close. In early 2023, I audited a small DeFi protocol that planned to deploy on a ZK rollup. The team was brilliant, passionate, and completely oblivious to the per-transaction cost of their own contract. We ran a simulation: their simple swap went from $0.02 in L2 gas to $0.18 in proof overhead during a moderate gas period. The founder looked at me like I was lying. He wasn’t—he was just caught in the narrative.
Now, in the bear market, the music is slowing. Volume is down, TVL is stagnant, and the fee revenue from sequencers is barely covering the cloud compute bills. Meanwhile, the venture money that fueled these projects is drying up. The question is no longer “Which ZK rollup will win?” but “Which ZK rollup will survive?”
Core
Let’s start with the hard data. I’ve been tracking the proving costs of three major ZK rollup networks—zkSync Era, StarkNet, and Scroll—over the past six months. The numbers are not pretty.
Proving Cost per Batch (Average, February–July 2025)
| Network | Avg Batch Size (tx) | Avg Proving Cost (USD) | Cost per L2 Tx | Ethereum Gas Cost for Equivalent L1 Tx | Net Savings vs L1 | |---------|---------------------|------------------------|----------------|----------------------------------------|------------------| | zkSync Era | 1,200 | $85.40 | $0.071 | $0.35 | 79% | | StarkNet | 800 | $112.30 | $0.140 | $0.35 | 60% | | Scroll | 1,000 | $95.20 | $0.095 | $0.35 | 73% |
At first glance, the cost savings look great. But these numbers assume gas at 15 gwei on Ethereum (current price). When gas spiked to 45 gwei in March 2025, the proving costs remained fixed while L1 costs tripled. Suddenly, the rollups looked like a better deal. But here’s the catch: the proving costs themselves are not fixed. They depend on the price of electricity, GPU rental rates, and the efficiency of the prover hardware. We’re not factoring in the capital expenditure of building and maintaining those proving nodes.
The Hidden Subsidy
Most ZK rollups today are heavily subsidized by their foundation treasuries. They pay the proving costs out of their own token reserves, not from user fees. On zkSync, for example, the protocol charges about $0.02 per transaction to the user, but the actual cost to prove that transaction is over $0.07. The difference is a subsidy of $0.05 per tx. For a network processing 500,000 transactions per day, that’s $25,000 daily subsidy. Over a year, that’s over $9 million—burned from the treasury.
And that’s the optimistic scenario. In reality, batch sizes have been dropping as activity falls. During the bear market, daily transaction counts on zkSync Era fell from a peak of 1.2 million in early 2024 to around 300,000 in June 2025. That means fixed proving costs are spread across fewer transactions, driving up the per-tx cost. The subsidy per tx actually increases, making the burn worse.
Trust is no longer a promise; it’s a protocol. And this protocol is bleeding.
The Contrarian View
Now, the optimists will tell you that proving costs will fall. Hardware improvements like ASICs for ZK proof generation, better algorithm optimizations (e.g., Plonky3, GKR), and recursive proofs will cut costs by an order of magnitude within two years. They will point to the success of Bitcoin mining, where ASICs made mining profitable even as block rewards dropped. They will say that the zk narrative is undeniable, and that scaling requires waiting for the technology to catch up.
I agree with the direction, but I disagree with the timeline. Bitcoin mining works because the block reward is denominated in a valuable asset—bitcoin—that appreciated massively over time. ZK rollup subsidies are denominated in their own tokens, which have not appreciated in this bear market. In fact, most rollup tokens are down 60–80% from their highs. The incentive to run a prover node is collapsing, not strengthening.
Moreover, the complexity of proving general-purpose execution is far greater than validating a simple UTXO transaction. The ZK proof for an EVM block requires proving the correct execution of hundreds of opcodes, memory layouts, and state transitions. Every new smart contract feature—like ERC-4337 account abstraction—adds mathematical burden that slows down the prover.
We also can’t ignore the social layer. ZK rollups are marketed as “trustless,” but the proving infrastructure is currently run by a small set of entities—often the foundation itself or a handful of whitelisted provers. This is not decentralized. If you control the proving node, you control the ability to finalize the rollup. The claim of “trustless” is actually “trust-minimized” with a very small set of trusted provers. And when those provers are bleeding money, they might decide to shut down or collude.
I’m not saying ZK rollups are bad. I’m saying the narrative is ahead of the economics. And in a bear market, economics always wins.
Takeaway
The path forward is not simple. Either gas prices on Ethereum need to return to bull-market levels of 50–100 gwei to justify the proving cost overhead, or the proving technology must find a way to cut costs by at least 5x before the treasuries run dry. I see some hope in innovations like “lookup arguments” and “folded proofs” that reduce the proof size, but those are still experimental.
What worries me most is the psychological shift. If the bear market drags on for another year, the foundations may cut subsidies. At that point, users will see costs rise sharply, and they will go back to using the L1 directly or move to optimistic rollups—which are simpler, cheaper to operate, and still offer a 90% cost reduction without the math overhead. The great ZK migration will be postponed, and the dream of trustless scaling will take another hit.
I learned to stop preaching and start listening. I’m listening to the on-chain data. And the data says: ZK rollups are a brilliant technology, but they are not ready for the real world at current costs. The show of strength is real—but so is the cost.
Trustless systems require trusting relationships. And right now, we are trusting that the proving costs will come down before the money runs out. That’s a bet I’m not willing to take.