Optimistic rollups made a pragmatic bet: assume transactions are valid by default, and only run computation if someone challenges them. It worked well enough to get Arbitrum and Optimism to significant scale. But the seven-day withdrawal window that defines the optimistic model - the delay required to allow fraud proofs to be submitted - has always been the thing users hate most about it.
Validity proofs, the mechanism underlying ZK rollups like zkSync Era and Starknet, take the opposite approach. Every batch of transactions comes with a cryptographic proof that the state transition is correct. The L1 doesn’t wait and hope; it verifies. Withdrawals settle in minutes rather than days.
The counterargument for years was cost. Generating a ZK proof is computationally expensive, which made ZK rollups slower to finalize and more costly to operate at the prover level. That tradeoff made the optimistic model genuinely competitive during 2022 and 2023, when ZK proving times were still measured in hours for complex EVM operations.
The Prover Problem Is Largely Solved
What changed is prover hardware and proof system design. Specialized ZK proving hardware - both custom ASICs and GPU-accelerated provers - has cut proving times dramatically. Meanwhile, proof systems like Plonky2 and STARKs have become more EVM-compatible without sacrificing efficiency. zkSync’s Boojum proving system, for instance, was specifically built for GPU parallelization.
At the same time, Optimism’s OP Stack only launched its fault proof system on mainnet in mid-2024, meaning Optimism operated without on-chain fraud proofs for years - relying instead on a centralized sequencer as the practical security backstop. That gap matters when comparing the two models.
Arbitrum Is Still Fighting
Arbitrum’s BOLD (Bounded Liquidity Delay) protocol, designed to make fraud proofs permissionless and resistant to delay attacks, represents a genuine technical effort to shore up the optimistic model. It’s not vaporware. But it also illustrates how much engineering overhead the fraud proof mechanism requires to function safely at scale.
ZK rollups don’t need that overhead. The proof either verifies or it doesn’t.
None of this means optimistic rollups are going away. They still have larger ecosystems, more deployed contracts, and better tooling in many respects. But the architectural advantage has shifted. Validity proofs were supposed to be the future - at this point, they’re increasingly the present.