How ZK rollups work
Execution happens on the L2; the magic is the proof:
- The sequencer orders transactions and produces L2 blocks.
- A prover (often a specialized party or hardware cluster) generates a succinct validity proof showing the execution was correct.
- The proof + compressed transaction data is posted to L1.
- L1 verifies the proof in ~300-500k gas (constant time regardless of how many transactions were in the batch).
- Once the proof is verified, the L2 state is final on L1. No fraud-proof window.
The proof technology is usually a SNARK (Succinct Non-interactive ARgument of Knowledge) or STARK (the Scalable Transparent variant). Both reduce a multi-transaction computation to a single cryptographic object small enough to verify on-chain cheaply.
ZK vs optimistic rollup tradeoffs
| ZK | Optimistic | |
|---|---|---|
| L1 verification cost | Higher per batch | Lower per batch |
| Withdrawal time | Minutes | 7 days |
| Proof generation time | Minutes to hours | N/A |
| EVM equivalence | Historically partial; improving | Near-full |
| Compute overhead | High (dedicated provers) | Low |
| Upgrade flexibility | Harder (changing the circuit is a big deal) | Easier |
Optimistic rollups dominate TVL today because they shipped earlier and hit EVM-equivalence faster. ZK rollups are gaining share as proof systems mature and costs come down.
Current ZK rollup landscape
- zkSync Era — the most-used ZK rollup by activity; its own EVM-like VM (EraVM) that’s close-but-not-identical to Ethereum.
- Starknet — uses Cairo (its own language) rather than Solidity; differentiates on novel applications.
- Polygon zkEVM — targets full EVM equivalence at the bytecode level.
- Linea — Consensys-built, also targets bytecode equivalence.
- Scroll — EVM-equivalent, positioned as a “canonical” ZK implementation.
Risks and considerations
- Proof system complexity — ZK rollups are orders of magnitude more complex than optimistic designs. Bugs in the circuit logic can produce valid proofs for invalid states — a catastrophic failure mode, though none have been exploited in production to date.
- Prover centralization — most ZK rollups have a small number of provers. A prover outage halts finalization.
- Upgrade authority — same as other rollups: multisigs control upgrades; compromise is a real tail risk.
- New primitives, less battle-testing — the technology is ~5 years old in production. Expect more failure modes to be discovered as TVL grows.
For users, the current practical state is: use a ZK rollup if the app you need is there. zkSync’s native DEXs (SyncSwap, iZiSwap) and Starknet’s DeFi (Ekubo, Nostra) are the most-developed ecosystems. EVM-equivalent ZK rollups (Linea, Scroll, Polygon zkEVM) have the smoothest migration from Ethereum mainnet patterns.