Why Rollups Exist
The gap that makes rollups necessary is staggering: Ethereum does roughly 15 transactions per second on L1, whilst Visa’s network can theoretically handle up to 65,000 TPS at peak capacity , albeit actual throughput sits far lower. You don’t optimize your way out of a gap like that. DeFi protocols, gaming, social apps, anything that needs responsive user interaction smashes straight into the same wall: L1 throughput is scarce, and scarcity makes everything expensive.
The scaling debate has been churning for years: sharding, state channels, sidechains, plasma, each with its own faction and manifesto. Rollups cut through all of that because they solve throughput without abandoning L1 security, which is the trade-off that sidechains and alt-L1s quietly accept and that marketing materials conveniently skip over. A caveat worth flagging early: “inheriting L1 security” is conditional on active fraud proofs (or validity proofs), correct liveness assumptions, and the absence of malicious admin upgrade keys. Not all rollups fully satisfy those conditions today.
The core idea is elegant: execute transactions off-chain where compute is cheap, then post compressed transaction data back to L1 where security is strong. L1 becomes a settlement and data availability layer rather than an execution environment. The rollup inherits the base layer’s security properties because all the data needed to reconstruct state lives on L1. Think about that for a second: you get cheap execution and Ethereum-grade security guarantees, at least in the design.
The sidechain comparison sharpens the distinction. A sidechain has its own validator set and security budget, so it’s only as secure as those validators are honest. A rollup posts its data to Ethereum, meaning that in principle, even if every rollup operator vanishes tomorrow, anyone can reconstruct state from L1 data and recover their funds. In practice, as of early 2023 both Arbitrum and Optimism have upgrade key multisigs that could alter bridge contracts, and Optimism hasn’t yet deployed active fraud proofs, so this guarantee is aspirational rather than fully enforced. Still, the ability to exit without trusting the operator is what elevates rollups beyond “just another place to park your transactions.”
Optimistic vs ZK Rollups
This is where things get really interesting. The rollup world splits into two camps, and the split comes down to a beautifully fundamental question: how do you prove that off-chain execution was correct?
Optimistic rollups take the innocent-until-proven-guilty approach. They assume transactions are valid by default and only check them if someone raises a dispute. When a batch is posted to L1, a challenge window opens (typically seven days) during which anyone can submit a fraud proof demonstrating an invalid state transition. A valid fraud proof rolls back the incorrect batch and slashes the dishonest operator’s stake. The system is “optimistic” because it assumes good behavior and only burns verification effort when challenged. There’s something philosophically satisfying about that.
ZK rollups take the opposite bet: trust math, not people. They generate a cryptographic validity proof (a zero-knowledge proof, specifically a SNARK or STARK) for every batch, verified on L1 before the batch is accepted. There’s no challenge window because the math itself guarantees correctness. If the proof verifies, the state transition is valid, no challenge window required.
The trade-offs here are fascinating and often poorly understood. Optimistic rollups are simpler to build and support general-purpose smart contract execution with minimal modifications, which is exactly why Arbitrum and Optimism shipped to production well before any general-purpose ZK rollup could. That seven-day challenge window is the price you pay for that simplicity: withdrawals from an optimistic rollup to L1 take a week unless you use a liquidity bridge, which introduces its own trust assumptions. Seven days just to move your own funds back to L1, which is a striking constraint for any application needing responsive settlement.
ZK rollups offer faster finality because validity proofs can be verified in minutes rather than days, albeit generating those proofs is computationally expensive. Supporting general-purpose EVM execution inside a ZK circuit remains one of the hardest engineering problems in the space. zkSync and StarkNet have made remarkable progress, but developer experience and tooling maturity still lag behind optimistic rollups as of early 2023.
The ZK landscape as of January 2023 is moving fast, and the pace of progress is remarkable. zkSync Era has been running on mainnet in a limited capacity and is preparing for broader public access, whilst StarkNet has been live on mainnet alpha since late 2021 with a growing application ecosystem, albeit developer tooling and language barriers (StarkNet uses Cairo rather than Solidity) remain real adoption hurdles. Both projects represent serious engineering bets that general-purpose ZK-EVM execution is within reach, and the progress over the past year has been substantial enough that writing ZK rollups off as purely theoretical would be a mistake.
Arbitrum vs Optimism: Where They Diverge
OK, this is where the rabbit hole goes deepest. Arbitrum and Optimism are the two dominant optimistic rollup implementations, and the differences between them are far more substantive than the casual “they’re basically the same” dismissal you’ll hear in most conversations.
Fraud proof design. The divergence here goes deeper than any other architectural difference between the two chains. Optimism uses single-round fraud proofs: when a dispute is raised, the entire transaction is re-executed on L1 to determine correctness. Arbitrum uses multi-round interactive fraud proofs, where disputing parties engage in a bisection protocol to narrow disagreement down to a single instruction, and only that instruction executes on L1.
Think about what that means in practice: Arbitrum’s dispute resolution is cheaper on L1 because you only execute a single instruction on-chain rather than replaying an entire transaction. Optimism’s approach is simpler to reason about and implement, but it imposes a hard cap on transaction complexity. If a transaction is too expensive to re-execute on L1, it can’t be disputed, which effectively limits individual transaction size. Two completely different engineering philosophies, both internally consistent, both with real consequences.
EVM compatibility. Optimism originally built the OVM (Optimism Virtual Machine), a modified execution environment that required contract developers to make adjustments, then moved toward EVM equivalence with their Bedrock upgrade. Arbitrum took a different path with ArbOS and the Arbitrum Virtual Machine (AVM), albeit they also provide strong EVM compatibility. In practice, both chains run most Solidity contracts without modification as of early 2023. Arbitrum’s Nitro upgrade brought them closer to true EVM equivalence by compiling contracts to WASM and running them against a Geth-based state transition function, which is a clever piece of engineering.
Transaction sequencing and finality. Both chains use a centralized sequencer for transaction ordering and soft confirmations, with decentralization on both roadmaps. The sequencer gives users near-instant confirmations, albeit those confirmations are only as trustworthy as the sequencer operator until transaction data is posted to L1 and either the challenge window passes or sufficient L1 blocks confirm the data.
Arbitrum’s sequencer provides around 0.25-second soft confirmations, with Optimism’s in a similar range. The UX difference is negligible for most applications, but the architectures diverge in how they handle sequencer failures and plan to decentralize sequencing, a distinction that will matter significantly as these networks mature.
The fraud proof design is where Arbitrum and Optimism made divergent engineering bets, and those bets cascade through every other architectural decision.
Ecosystem and adoption. Arbitrum held roughly $3 billion in TVL as of late 2022, compared to Optimism’s roughly $800 million, according to L2Beat. Arbitrum’s larger ecosystem reflects its earlier launch and stronger developer experience. Optimism’s OP Stack, though, represents a different strategic bet entirely. Rather than winning by being the single best rollup, Optimism is positioning itself as the default infrastructure for building rollups. The OP Stack’s open design means any team, including major exchanges and institutions, could use it to launch their own chain with minimal overhead. That’s a bold play with sweeping implications for the entire L2 landscape.
Trade-offs That Matter in Production
Once you move past the architectural fascination and start evaluating rollup infrastructure for actual production use, marketing comparisons fall away and a few concrete questions remain.
Withdrawal latency. The seven-day challenge window on optimistic rollups is a real constraint for applications needing fast L1 settlement. Liquidity bridges like Hop Protocol and Across can provide faster withdrawals, but they add counterparty risk and fees. ZK rollups will win decisively on this axis once they’re production-ready for general-purpose use, and that day is coming faster than most people expect.
Developer tooling and ecosystem maturity. Both chains have strong tooling, but Arbitrum’s larger TVL means battle-tested contracts, deeper liquidity, and broader composability options. For a team shipping to production today, that ecosystem depth is a practical advantage that outweighs most theoretical architectural comparisons, however elegant those comparisons might be.
Decentralization roadmap. Both chains rely on centralized sequencers, which means you’re trusting their operators for transaction ordering and censorship resistance. Both have decentralization on their roadmaps, but the details and timelines differ. A team evaluating rollup infrastructure should weight the credibility of those commitments heavily, because a rollup with a centralized sequencer and no clear path to decentralization is, in practice, a hosted service with extra steps.
Operational and bridge risks. Beyond the protocol-level trade-offs, rollups carry practical risks that deserve weight in any production evaluation. Both Arbitrum and Optimism rely on upgrade key multisigs that can modify core bridge contracts, sequencer centralization means a single operator controls transaction ordering and inclusion, and cross-chain bridges remain a persistent attack surface, as the string of bridge exploits in 2022 painfully demonstrated. These are solvable problems on both roadmaps, but as of early 2023 they represent real trust assumptions that shipping teams need to account for.
Data availability costs. Both optimistic and ZK rollups post transaction data to Ethereum L1, so L1 gas prices directly affect rollup operating costs and user fees. Ethereum’s upcoming EIP-4844 (proto-danksharding) will introduce blob transactions that could dramatically reduce L1 data posting costs for all rollups, which should reshape the cost structure for every rollup. The timeline and impact remain uncertain as of early 2023, but the direction is clear.
What This Means for Infrastructure Decisions
After spending real engineering time with both Arbitrum and Optimism, my take is that the choice between them is driven by your specific constraints rather than any absolute technical ranking. Arbitrum’s interactive fraud proofs are a more elegant engineering solution, but Optimism’s OP Stack modularity opens strategic possibilities that a single-chain approach can’t match. Both are credible production infrastructure handling real value.
The broader pattern is convergence. Optimistic rollups are getting faster finality through improved bridge designs and upcoming proof systems, whilst ZK rollups are gaining better EVM support and developer tooling. The differences that feel sharp today will blur over the next two years, and the teams building on rollups right now will have a head start when that convergence arrives.
The rollup ecosystem has produced useful patterns for batching, proving, and bridging that extend well beyond Ethereum. In Part 2 of this series, I’ll dig into how Arbitrum’s rollup mechanics integrate with Stacks Subnets, and why that combination might offer something that neither approach provides alone.