What a Layer 2 rollup actually is
A Layer 2 (L2) rollup is a separate execution environment that batches many transactions, runs them off Ethereum, and then "rolls up" the results into a compressed package posted back to the base chain. Ethereum stays the settlement and data-availability layer; the rollup borrows its security rather than minting its own validator set. That distinction is what separates a rollup from a sidechain — a sidechain has its own consensus and its own trust assumptions, while a rollup posts enough data to L1 that anyone can reconstruct and verify its state.
There are two dominant designs. Optimistic rollups (Arbitrum, Base, OP Mainnet) assume every batch is valid and open a challenge window — typically 7 days — during which a fraud proof can claw back an invalid state. ZK rollups (zkSync Era, Linea, Scroll, Starknet) generate a cryptographic validity proof for every batch, so withdrawals can finalize in 1–24 hours with no trust in watchdogs.
How the economics work
The cost of a rollup transaction is dominated by what it pays to publish data on L1. Before March 2024, rollups wrote data as permanent calldata — rollups spent roughly 15,000 ETH on calldata in December 2023 alone. EIP-4844 (Dencun) introduced blobs: fixed-size data chunks (128 KiB each) that nodes prune after ~18 days. Because blobs aren't stored forever, they're far cheaper, and the switch cut data costs by 90–99%.
The blob supply keeps expanding. Pectra (2025) lifted the per-block blob target from 3 to 6. The Fusaka upgrade, live on mainnet since December 2025, shipped PeerDAS — peer data-availability sampling — so nodes only download a slice of blob data instead of all of it. Staged BPO forks pushed targets to 10, then 14 blobs per block, with core devs aiming higher through 2026. A typical L2 transaction that cost ~$0.50 in late 2025 dropped toward $0.20–$0.30 afterward.
Why it matters
As of May 2026, more than 70 active rollups hold over $48 billion in TVL, with Arbitrum, Base, and OP Mainnet alone commanding the majority. For a regulated venue dealing in tokenized real-world assets and stablecoin settlement, the appeal is concrete: cheap, fast finality without abandoning Ethereum's audit trail. Under MiCA and similar frameworks, provable settlement data matters as much as throughput, and a rollup that posts its data to L1 keeps that trail intact.
The risks people underweight
Most rollups are not as trustless as the marketing implies. L2Beat's stage framework — Stage 0 through 2 — grades each chain on state validation, data availability, exit windows, and sequencer/proposer failure. The uncomfortable data point: across analyzed projects, 86% lack an exit window because of instant upgradeability, meaning a multisig can change the rules before users can leave. Most chains still run a single, centralized sequencer; if it goes offline or censors you, the only recourse is a forced-inclusion path on L1 — and ~13% of projects have no credible one. Even where a forced path exists on paper, parameters often make it impractical for an ordinary user during an actual outage.
In my view, this is the gap to scrutinize before you trust a rollup with size: not the headline TPS, but whether you can exit when the operator misbehaves or disappears.
Where it's heading (2026)
The combined L2 ecosystem processes roughly 5,600 TPS today, with projections of 24,000+ as blob capacity ramps. ZK proving is the variable to watch — proof-generation hardware is maturing fast, and the validity-proof model removes the multi-day withdrawal delay that optimistic designs still carry. The harder problem isn't speed. It's whether the operators behind these chains are willing to give up the upgrade keys and decentralize their sequencers — the part of the roadmap that doesn't ship itself.