Ethereum Scaling in 2025: zk-Rollups, Optimistic Rollups, and Privacy

Ethereum Scaling in 2025: Layer-2 Solutions and the Path Forward

Ethereum’s mainnet remains the most widely used smart-contract platform, but its base layer processes only about 15 transactions per second. For decentralized applications to serve mainstream users, that throughput must increase by orders of magnitude without sacrificing security or decentralization. Layer-2 solutions are the primary strategy for achieving this scaling. By moving transaction execution off the main chain while retaining its security guarantees, Layer-2 networks have already pushed Ethereum’s effective throughput past 2,000 transactions per second in some configurations.

This article examines the two dominant Layer-2 approaches (zk-Rollups and Optimistic Rollups), the trade-offs they introduce, and the emerging infrastructure that connects them. It draws on reporting from NPR and WIRED to explain both the technical mechanics and the practical implications for developers and users.

How Layer-2 Scaling Works

Layer-2 solutions batch transactions together, process them off-chain, and submit a compact proof or summary back to Ethereum’s mainnet. The key insight is that Ethereum’s security (its validator set, consensus rules, and data availability) still protects the batched transactions. Users do not need to trust the Layer-2 operator. They rely on the same cryptographic guarantees that secure the base layer.

To understand the difference between the two main approaches, consider a concrete example. Suppose Alice wants to trade 10 ETH for USDC on a decentralized exchange. On Ethereum mainnet, that transaction would compete with every other transaction for block space, potentially taking minutes and costing tens of dollars in gas fees. On a Layer-2 network, Alice’s transaction is bundled with thousands of others, processed in seconds, and settled on Ethereum in a single batch. The cost drops to a few cents.

The two approaches differ in how they prove the correctness of those batched transactions. This distinction has significant consequences for speed, cost, and developer experience.

zk-Rollups: Zero-Knowledge Proofs for Instant Finality

Zero-knowledge rollups generate a cryptographic proof (a validity proof) for each batch of transactions. This proof mathematically verifies that every transaction in the batch was executed correctly. Ethereum’s mainnet only needs to check this short proof, not re-execute the thousands of individual transactions. The result is instant finality: as soon as the proof is submitted and verified on Ethereum, the batch is considered final. Users do not need to wait for a challenge period.

Leading zk-Rollup projects include zkSync, StarkNet, and Scroll. Each uses a different proof system (zkSync uses PLONK, StarkNet uses STARKs, Scroll uses a custom proof system compatible with the Ethereum Virtual Machine). The common thread is that all validity proofs provide mathematical certainty about batch correctness.

However, generating these proofs is computationally intensive. A single batch might require hours of computation on specialized hardware. This creates a practical bottleneck: proof generation must be fast enough to keep up with transaction inflow. Projects are investing in hardware acceleration (FPGAs, GPUs, and custom ASICs) to reduce proof generation time from hours to minutes.

For developers, zk-Rollups present a challenge. The Ethereum Virtual Machine (EVM) was not designed with zero-knowledge proofs in mind. Writing smart contracts that are efficient to prove in zero-knowledge requires different optimization strategies. Some zk-Rollups (like Scroll) aim for EVM equivalence, meaning existing Solidity contracts run without modification. Others (like StarkNet) use a different virtual machine (Cairo) that requires rewriting contracts.

Optimistic Rollups: Fraud Proofs and Challenge Periods

Optimistic rollups take a different approach. They assume transactions are valid by default (hence “optimistic”) and only check correctness if someone challenges the batch. A challenger can submit a fraud proof that shows a specific transaction in the batch was executed incorrectly. If the fraud proof is valid, the batch is reverted and the sequencer (the entity that submitted the batch) loses its staked bond.

This design introduces a challenge period (typically one to seven days) during which anyone can submit a fraud proof. During this period, users cannot withdraw funds from the Layer-2 back to Ethereum mainnet with full security guarantees. They must either wait for the challenge period to expire or use a liquidity provider that offers faster withdrawals (at a fee).

Arbitrum and Optimism are the two dominant Optimistic rollups. Arbitrum uses a multi-round fraud proof system called “Arbitrum Nitro” that reduces the cost of fraud verification by narrowing the dispute to a single disputed instruction. Optimism uses a single-round fraud proof system called “OP Stack” that requires re-executing the entire disputed transaction on Ethereum mainnet. The multi-round approach is more gas-efficient for the verifier but adds complexity to the challenge protocol.

For developers, Optimistic rollups offer a smoother path. Both Arbitrum and Optimism are EVM-equivalent at the bytecode level. A Solidity contract deployed on Ethereum mainnet can be deployed on Arbitrum or Optimism without any code changes. This compatibility has made Optimistic rollups the early leader in total value locked (TVL) and deployed applications.

Comparison: zk-Rollups vs Optimistic Rollups

Both approaches have produced working systems, but they optimize for different priorities. zk-Rollups offer faster finality and stronger security guarantees at the cost of higher computational overhead and reduced EVM compatibility. Optimistic rollups offer easier developer onboarding and lower operational costs at the cost of delayed withdrawals and a trust assumption in the challenge mechanism.

Cross-Rollup Infrastructure: Bridges and Asset Transfers

As the number of Layer-2 networks grows, moving assets between them becomes a critical problem. A user who deposits ETH into Arbitrum cannot directly send it to a zkSync wallet. They must first withdraw from Arbitrum back to Ethereum mainnet (waiting through the challenge period) and then deposit into zkSync. This process is slow and expensive.

Cross-rollup bridges attempt to solve this by maintaining liquidity pools on multiple Layer-2 networks and routing transfers through these pools. A bridge locks assets on the source rollup and mints equivalent assets on the destination rollup. The bridge operator takes a fee for providing this service.

The economics of these bridges are driven by blob data costs. Ethereum’s EIP-4844 (implemented in March 2024) introduced “blobs” (temporary data blocks) that Layer-2 networks use to publish batch data. The cost of posting a blob depends on the blob gas market, which fluctuates with Layer-2 activity. Cross-rollup transfers that require data publication on both the source and destination rollups are sensitive to these costs. Blob data economics and cost optimization for cross-rollup asset transfers is a topic we have explored in detail elsewhere on this site.

Security is the primary risk in cross-rollup bridges. A bridge that holds millions of dollars in liquidity is an attractive target. If the bridge’s smart contract has a vulnerability, an attacker can drain the entire pool. Several high-profile bridge hacks (including the Ronin bridge and the Wormhole bridge) have demonstrated that bridge security is harder to get right than individual Layer-2 security.

Emerging solutions include atomic swaps (where both sides of the transfer execute or neither does) and canonical bridges (where the Layer-2 network itself natively supports cross-rollup transfers). Neither is widely deployed as of 2025.

Security Considerations Across Layer-2 Networks

Layer-2 networks inherit Ethereum’s security for transaction execution, but they introduce new attack surfaces. The sequencer (the entity that orders transactions within a batch) is a centralization point. If the sequencer is malicious or compromised, it can censor transactions, reorder them for profit (MEV extraction), or submit invalid batches (though these would be caught by the proof system).

Most Layer-2 networks currently use a single sequencer operated by the project team. Decentralized sequencer sets, where multiple independent entities compete to propose batches, are an active research area but not yet deployed in production. The centralization of sequencers is a known weakness that the Ethereum community is working to address.

Another risk is the dependency on the Layer-1 Ethereum mainnet. If Ethereum experiences a consensus failure or a reorganization, Layer-2 networks that depend on Ethereum’s data availability and finality could be affected. This is a systemic risk that applies to all Layer-2 solutions equally.

For users, the practical takeaway is to verify the security assumptions of any Layer-2 network they use. Does the rollup have a working fraud proof system? Is the sequencer decentralized? Has the bridge contract been audited by multiple reputable firms? These questions matter more than the marketing claims of any project.

Developer Experience: Building on Layer-2

For developers building decentralized applications, the choice between zk-Rollups and Optimistic rollups depends on their specific requirements. If instant finality is critical (for a high-frequency trading application, for example), zk-Rollups are the better choice despite the higher development cost. If the application is a standard DeFi protocol (lending, swapping, yield aggregation), Optimistic rollups offer a faster path to deployment with full EVM compatibility.

The tooling ecosystem has matured significantly. Both Hardhat and Foundry support deployment to Arbitrum, Optimism, zkSync, and StarkNet. Ethers.js and Web3.js work with all major Layer-2 networks through the standard JSON-RPC interface. The main difference is in debugging: Optimistic rollups support the same debug_traceTransaction and eth_call semantics as Ethereum mainnet, while zk-Rollups often require custom debugging tools because the EVM execution traces are not directly available.

Gas costs on Layer-2 networks are typically 10x to 100x lower than Ethereum mainnet, but they are not zero. A simple token transfer might cost $0.01 on Arbitrum compared to $1.50 on Ethereum mainnet. A complex DeFi interaction (swap, deposit, and stake) might cost $0.05 on Optimism compared to $5.00 on Ethereum mainnet. These savings make it economically viable to build applications that would be prohibitively expensive on the base layer.

What This Means for Users and Developers in 2025

Layer-2 scaling is no longer theoretical. Arbitrum and Optimism together hold over $15 billion in total value locked. zkSync and StarkNet are growing rapidly, with dedicated user bases and growing application ecosystems. The competition between the two approaches is driving innovation in proof systems, bridge design, and developer tooling.

For users, the practical benefit is lower fees and faster transactions. For developers, the benefit is access to a larger user base without the constraints of Ethereum mainnet’s limited throughput. The trade-offs (delayed withdrawals on Optimistic rollups, higher development costs on zk-Rollups) are manageable with careful design.

The next frontier is smooth interoperability between Layer-2 networks and between Layer-2 and Layer-1. DevOps practices for 2024, including platform engineering and security-first approaches will be essential as the infrastructure matures. Projects that invest in cross-rollup infrastructure today will be well positioned as the ecosystem grows.

Sources and References

This article was researched using a combination of primary and supplementary sources:

Supplementary References

These sources provide additional context, definitions, and background information to help clarify concepts mentioned in the primary source.

• Census.gov | U.S. Census Bureau Homepage
• The Republican Plan to Reform the Census Could Put Everyone’s Privacy at Risk
• 2030 Census