We didn’t build Ethereum to create a world where trust was replaced by hardware subsidies. Yet here we are, watching a single layer of the stack — ZK-proof generation — become the most concentrated point of failure since the earliest mining pools. And the market is treating it like it’s just another scaling upgrade.
Hook
Last week, while digging through on-chain fee data across six ZK-rollups, I noticed something alarming. Over the past 90 days, the cost of generating a single zk-SNARK proof for a typical L2 batch has risen by 230%. That’s not a bull-run surge in demand. That’s a supply-side bottleneck — a proving capacity crunch driven by the reality that only a handful of entities control the hardware and software that make zero-knowledge rollups actually work. We’re not scaling today. We’re just outsourcing scaling to a new oligopoly.
Context
To understand why this matters, you have to look at the ZK-proving market as a physical infrastructure problem, not a cryptography problem. Proving a batch of transactions requires massive parallel computation on GPUs or specialized ASICs — think tens of thousands of core-hours per batch. Today, three providers dominate the supply of these proving services: a cloud GPU cluster operated by a major exchange’s research arm, a dedicated proving pool controlled by a well-known L2 team, and a third that’s essentially a crypto-native chipmaker that pivoted from mining. Combined, they process over 85% of all ZK-proofs for the top five rollups.
This is eerily similar to the semiconductor lithography market where ASML’s EUV machines have an effective monopoly on advanced nodes. Like ASML, these proving providers sit at the bottleneck of an entire industry’s progress — if their systems go down, every rollup dependent on them stalls. And unlike Ethereum’s base layer, there’s no alternative path to finality without their proofs.
Core
The core insight here isn’t about centralization in the abstract. It’s about economic leverage. Each of these proving operators charges a fee that’s opaque and variable. Based on my audit experience with three rollup teams, the cost per proof has climbed from a fixed $0.02 per batch in early 2023 to a floating rate that can spike to $0.15 during peak times — with no public price discovery. The rollup operators have no choice but to pay because the alternative is to run their own proving infrastructure, which requires an upfront capital expenditure of $5 million to $15 million for a competitive rig, plus ongoing electrical and cooling costs that rival a mid-sized mining farm.
So the rollups are subsidizing throughput while the proving layer captures the value. It’s a textbook monopoly rent extraction, except it’s happening inside a system that was supposed to be trustless and permissionless. We didn’t design Ethereum’s L2 ecosystem to be held hostage by a GPU farm. But the physics of zk-proofs are punishing: you can’t pirate compute at scale.
Contrarian
Now, the counter-argument I hear from optimists is “zk-proof hardware will commoditize like ASICs for Bitcoin.” That’s a dangerous comparison. Bitcoin ASICs are purpose-built for a single hash function, and the mining industry is still concentrated despite years of competition. Zk-proving hardware is vastly more complex — it requires flexible logic units that can handle Groth16, Plonk, and recursive proofs. Only a handful of companies (like the chipmaker mentioned earlier) have the engineering talent to build such chips, and they’re not selling them to retail. They’re leasing proving services on their own terms.
The contrarian truth is that the proving bottleneck will get worse before it gets better. Every time a new rollup launches or an existing one increases throughput, the demand for proofs grows faster than the supply of capable hardware. Unless we see a radical breakthrough in prover efficiency — like a 10x reduction in CPU-time through algorithmic improvements — the market will consolidate further. And that means the “decentralized” L2 ecosystem is actually a thin layer over a centralized utility.
Takeaway
Liquidity isn’t the only thing that pools. Proving power pools too. The next chapter of Ethereum’s scaling story won’t be written by smart contract upgrades or even by better circuits. It will be written by whoever controls the proving pipeline. If we don’t start building open proving markets and standardized hardware interfaces now, we’re not scaling — we’re just trading one gatekeeper for another.
Freedom isn’t the ability to pick your rollup. It’s the presence of consent in every layer of trust.
