Disclosure: The views and opinions expressed herein belong solely to the authors and do not represent the views and opinions of crypto.news editorials.
You tried to bridge $50 from Ethereum (ETH) to a ZK rollup like zkSync, expecting to pay out a few pennies. Instead, they charge a fee of $0.15 to $0.50. This is hard to accept when you know that Optimistic Rollups and ZK Rollups are often 3x cheaper, and the Dencun upgrade reduced data costs by over 90%. The bottleneck is not network congestion. That in itself is proof.
summary
ZK rollup pricing is driven by proof, not data or execution. GPU inefficient cryptography consumes 60-70% of fees for proof generation, creating a ~$97 million concentrated prover market and keeping ZK fees high despite Ethereum’s Dencun upgrade. This creates a centralization and security trap. Over 90% of ZK-L2 relies on a small number of Prover-as-a-Service providers, introducing censorship, MEV extraction, outages, and Web2-style rent extraction, undermining decentralization and adoption. The solution is specialized hardware and an open prover market. ZK’s unique ASIC and competitive on-chain prover auctions will ultimately make scaling ZK viable by reducing costs by orders of magnitude, decentralizing proofs, and unlocking sub-cent fees.
Generating zero-knowledge proofs for transaction batches is an intensive process. This involves trillions of mathematical operations, especially elliptic curve multiplication, performed on hardware not designed for the task. According to L2Beat data, the certification process alone accounts for a staggering 60-70% of ZK-L2’s total fees.
This has created a concentrated market of “prover farms” estimated to be worth more than $97 million. This is a multi-billion dollar bottleneck. This is a centralized, power-hungry system that relies on mismatched hardware, and is why ZK scaling still feels more like a promise than a reality.
There are two solutions. We need domain-specific ASICs and an open prover market. Introducing these could make sub-cent transactions the default rather than the exception. This is no fantasy. This is an engineering reality waiting to be accepted.
Where does the fee actually go?
ZK rollup transactions have several costs. L2 execution is almost free. Posting data to the mainnet is also cheaper thanks to BLOBs, costing a tenth of a cent per transaction. The real beast is proving it. Generating a single proof for a batch of 4,000 transactions takes 2 to 5 minutes on a high-end A100 GPU.
Based on Brevis benchmarks, cloud computing fees alone cost between $0.04 and $0.17. If you amortize that cost, it’s about 1 to 4 cents per transaction. However, under heavy loads, the value can easily jump to more than 10 cents.
ZK proofs are not the same as matrix calculations used in AI. It relies on multiscalar multiplication on elliptic curves (MSM) and number-theoretic transformations (NTT). These operations are challenging for GPUs designed for parallel floating-point operations. One Groth16 proof requires approximately 1012 field operations.
According to testing by Ingonyama, approximately 80% of GPU cycles are idle during this process. The hardware just isn’t compatible. zkSync Era’s report for Q1 2025 showed that attestation accounted for 65% of its $2.3 million in revenue.
This amounts to $1.5 million paid directly to the centralized cluster. Rollup teams, on the other hand, pay AWS $1,000 to $5,000 per rig per month, and their costs increase linearly with volume. No wonder the total amount locked in the top ZK L2 remains at $3.3 billion while the optimistic rollup is over $40 billion.
However, even optimistic rollups (Base, Arbitrum, Optimism, etc.) are not a complete solution, as they involve significant trade-offs. Optimistic rollups require a withdrawal delay of at least 7 days to allow monitoring validators to dispute invalid state transitions. In the fast-paced world of DeFi, this latency is too long for most use cases. These networks also require users to trust these monitoring validators. This may be fine for low-value transactions like Web3 gaming and social media, but it leaves a lot to be desired when it comes to real-world assets and other powerful financial use cases.
Therefore, the path to truly scalable, cheap, and secure transactions lies not in avoiding proof generation, but in revolutionizing it.
This is the core of inefficiency. You’re paying a Web2 premium for both cost and centralization to create trust in a decentralized network. Certification fees are not just expenses. It is an economic anchor that prevents the adoption of ZK. Until generating proofs becomes as cheap as running them, ZK rollup will remain locked into a system where its greatest strength, cryptographic security, is also its most costly bottleneck. And that cost goes beyond just inflating fees. It builds a centralized trap that we must now escape from.
The trap of centralization
More than 90% of ZK rollups are outsourced to a small number of “proof-as-a-service” companies. This creates a large single point of failure. In March 2025, Blast operators froze the attacker’s transactions and 12,000 innocent users for 48 hours. The central prover verifies the plaintext batch, extracts the MEV, and rejects the transaction. dYdX uses 3-of-5 multisig. This means three insiders could potentially hijack the chain. This reflects hardware dependency issues in other blockchain systems, such as Solana’s consensus mechanism.
This centralization creates vulnerabilities. Although the top provers achieve 99.2% uptime, a single AWS outage can cause a cascading effect. In Q2 2025, Starknet experienced a 20% drop in throughput due to a single node failure. These providers have earned 80% margins, and while rollup rents remain tough, Ethereum’s L1 provides security. It’s an architectural contradiction. We have “decentralized” the L2 that relies on Web2 data centers. As Vitalik Buterin said at Devcon 2024, “Outsourcing proof doesn’t scale anything, you just trust a new oracle.”
The fix is included
Solving this problem requires a one-two punch. We need silicon built for ZK, and we need a marketplace to distribute it. First up is the hardware. GPUs are like race cars on the highway. These are great for tensor operations used in AI, but terrible for curve operations used in ZK. This is where domain-specific ASICs come into play.
The winners of the 2025 ZPrize competition were able to time their STARK proofs on FPGAs in 6-8 seconds. This is 10 to 100 times faster per watt than a GPU because it is wired with MSM and NTT. Cysic has already developed an ASIC that can produce a proof of a billion-gate circuit in less than two seconds, and does so with a 50x reduction in energy consumption. This is the same evolution seen in Bitcoin (BTC) mining. We’ve moved from CPUs to ASICs, but this time it’s about truth, not hashes.
Second is the market. We need to do away with the SaaS model and treat provers as a commodity. Succinct’s 2025 mainnet exemplifies this. Rollup posts jobs on Ethereum-based auctions, and professional rigs place bids in seconds. Brevis ProverNet achieved 20x CPU speed via a GPU cluster with a slash of lazy nodes.
The result is a 40% reduction in costs, fair distribution of value, and resistance to censorship. Anyone can prove it, but Ethereum will. ZPrize 2025’s recursive aggregation verified proofs on-chain in less than 200ms. A $10,000 DeFi bot can run private ZK-ML on L2 for 1/1000th of a cent per transaction.
unlock the floodgates
According to the zkSync Atlas roadmap, cheaper proofing will result in predictable charges of less than $0.01 for 15,000 TPS. This will unlock pay-per-pixel NFTs, real-time gaming economies, and AI agents that settle transactions without front-running. L2 becomes a neutral infrastructure rather than the province of the prover.
Developers focus on optimizing VMs rather than managing large server farms. This will drive 10x more innovation in apps. ZK TVL will grow from $28 billion to more than $100 billion in 2025. The real opportunity is holistic.
When proof is cheap and decentralized, the entire value proposition crystallizes. Users get speed and cost. Developers now have a platform where they don’t have to trust a small number of operators. Ethereum gets a scaling solution that maintains security guarantees without compromise.
the battle is here
The biggest barrier to ZK adoption is not cryptocurrencies or games, but engineering economics. Centralized infrastructure forecloses scalable truths. The war did not end with an agreement. It’s on the market with verifiable computing hardware. We ship decentralized and accelerated provers, bringing cheap, trustless transactions to billions of people.
We don’t need a prover overlord. We need a global bazaar for proof where everyone contributes, everyone verifies, and the open web is secure.
