A rollup is a Layer 2 blockchain that runs its own execution environment, processes user transactions, and periodically “rolls up” a summary of its activity to a base layer like Ethereum. The summary includes enough information for anyone observing the base layer to reconstruct the rollup’s state and verify that it is consistent with the rules — the specific form of this verification is what distinguishes optimistic rollups from zero-knowledge rollups, and the two are the main flavours of rollup in use today.
The critical feature of a rollup, compared to other scaling approaches, is that transaction data is posted to the base layer. This means anyone can reconstruct the rollup’s state from L1 data alone, without trusting the rollup operator. If the rollup operator disappears, goes offline, or tries to censor users, the data for every transaction is still available on L1, and a new operator (or the users themselves) can rebuild the state and continue from where things left off. This “data availability” property is what makes rollups genuinely secure in a way that sidechains or plasma chains are not.
How Rollups Save Money
A rollup’s value comes from the fact that L1 block space is expensive and L2 execution is cheap. Instead of running every user transaction directly on Ethereum mainnet (where each transaction might cost several dollars), the rollup runs transactions off-chain and only posts a compressed summary to L1. The summary includes the transaction calldata (or, after EIP-4844, blob data) and the new state root, but not the full transaction history with all the intermediate details that the EVM normally requires to process each step.
The compression ratio is substantial. A rollup can pack hundreds of transactions into the same L1 cost as a single mainnet transaction, which is roughly where the order-of-magnitude cost reduction comes from. After the March 2024 Dencun upgrade introduced dedicated blob space for rollups, the cost per rollup transaction dropped by another 10x, bringing most L2 transactions to single-cent territory. This is the infrastructure-level reason Ethereum’s usable capacity has grown dramatically over the last few years without changing the mainnet’s block size.
Optimistic vs ZK
Optimistic rollups (Arbitrum, Optimism, Base, Unichain) post their state roots to L1 and assume they are correct unless someone proves otherwise. There is a challenge window — typically 7 days — during which anyone can submit a fraud proof showing that the state transition was invalid. If the challenge succeeds, the bad state is reverted and the challenger gets a reward. If no challenge arrives within the window, the state is considered final and users can withdraw freely. The 7-day delay is the main user-facing cost of the optimistic design.
ZK rollups (zkSync, Starknet, Linea, Scroll, Polygon zkEVM) post a cryptographic proof of validity alongside the state root. The L1 contract verifies the proof immediately and accepts the state root as correct without waiting for any challenge. The tradeoff is that generating the proofs is computationally expensive, which historically made ZK rollups more complex to build and less EVM-compatible. This gap has been closing rapidly — modern ZK rollups support most Solidity contracts with minimal changes, and the prover costs have dropped by multiple orders of magnitude since 2021 — but optimistic rollups still have a meaningful head start in production usage and total value locked.
Long-term, most observers expect ZK to dominate because instant withdrawals and stronger trust assumptions are structurally better, but the timeline is measured in years and optimistic rollups will likely remain the dominant architecture for the next several.
The Rollup-Centric Roadmap
Vitalik Buterin outlined what is now called the “rollup-centric roadmap” in a 2020 blog post: instead of scaling Ethereum L1 by making blocks bigger or shardings the base chain, the plan is to keep L1 as a secure but slow settlement layer and push all high-throughput activity to rollups. L1 becomes a kind of settlement spine; L2s become the places where everyday transactions happen; users interact with L2s directly for most things and only occasionally touch L1 for bridging or high-value settlement.
This has been the explicit strategy of Ethereum development since 2021, and it has mostly worked. A large fraction of Ethereum activity now happens on L2s, the L2 ecosystem has matured substantially, and Dencun’s introduction of blob space was specifically designed to support the rollup-centric approach. The main remaining concerns are about fragmentation (different rollups have different user bases, and moving assets between them has UX friction) and about decentralisation of individual rollups (sequencers are still mostly centralised on most rollups, though decentralised sequencer designs are in progress).
For a user today, the practical meaning is that if you are transacting on Ethereum for amounts under a few thousand dollars, you should probably be on an L2 rather than on mainnet, and the specific L2 you pick matters less than the general decision to use one. Base, Arbitrum, and Optimism are the biggest by activity; OP Stack chains are growing faster than standalone rollups; and zkSync/Starknet/Linea are catching up on the ZK side. The category as a whole has become the default home for Ethereum usage, and mainnet is increasingly reserved for the specific cases where L1 settlement is worth the cost.