This is a segment of the 0xResearch newsletter. To read the complete edition, Subscribe.
It all starts with Ethereum. L1 is too late. All nodes must redundantly re-execute transactions within the block. This limits scalability.
By 2020, the Ethereum community has merged around the rollup Scaling solution. A groundbreaking 2021 blog post on Vitalik’s rollups, we’re optimistic and explain how ZK rollups can expand Ethereum.
ZK rollups are technically good at scaling, but the technology is still not available and was far behind (later).
So optimistic rollups first invaded the market. Optimistic rollup assumes that every transaction is valid until it challenges it with evidence of fraud within a seven-day challenge period.
The optimistic rollup worked, but came with hidden costs. The challenge period meant the finality of longer transactions before users could withdraw funds. Locked liquidity meant capital efficiency and generally poor UX, particularly due to chain interoperability.
Meanwhile, ZK was catching up. At ETHCC 2022, Polygon, Zksync and Scroll all announced Zkevms. This allows Solidity Dev to write code and prove that EVM runs.
By 2023, ZK rollups were beginning to gain real traction.
Why is ZK proof superior to optimistic fraud proof? In other words, ZK proofs are much smaller (~1-10 kb) compared to raw transaction data (size of megabytes).
By using ZK encryption to prove Ethereum transactions, these highly compressed proofs provide lower data availability costs and increased scalability.

How Zero Knowledge Works in One Word
The ZK had taken off, but the proof generation was still expensive. Based on ZKSTATS.IO, the average cost to generate ZK proofs in December 2023 was $80.21.
Fast forward to 2025. Certification costs decreased to $1.3 per proof, with an improvement of about 98.4%.
What has changed?
Today, each of the core pieces of the ZK rollup stack is split.
First of all, ZKVMS is here. These specialized virtual machines speed up the ZK development experience and make validity proof generation more efficient. Before ZKVM, developers had to write complex mathematical “circuits” to prove the execution of EVM.
ZKVMS such as SP1, RISC Zero, Nexus, OpenVM effectively democratize ZK development for all developers (C++, Rust) without ZK encryption expertise. In the past, ZKEVMS only enabled robustness development in ZK rollups. Think of ZKVMS as a more general concept than Zkevms.
Second, the cost of generating proofs for market competition is declining. Today there are RISC Zero, Cysic, Lagrange and many competitive markets operating in concise. Some are still on the testnet, while others are operational.
ZK L2S demonstrates aggregation techniques to amortize verification costs. The way this works roughly is to batch many proofs within one. This will verify the final proof more quickly.
These markets are not permitted either. This means anyone with a GPU rig can sign up, post bonds, and generate ZK proofs. Previously, ZK rollups used “centralized prover.” In other words, I rented GPU/FPGA hardware from Google or Amazon.
Better proof systems are also constantly being released. These proof systems (e.g. GROTH16, HALO2-KZG, STARK, PLONK, EXPANDER) define algorithmically the rules for how ZK proofs are constructed and verified. They’re getting better, which means they’re verifying the ZK proofs smaller and faster. This means improved ZKVM performance.