Ethereum mainnet does maybe 15-30 transactions per second. Gas fees $20-50 on a normal day. Hundreds during congestion.

Not usable for most people at that cost. The technology works. The throughput doesn't.

Layer 2 is the fix. Separate network sitting on top of Ethereum. Handles transactions off the main chain fast and cheap. Periodically bundles everything and settles back to Ethereum for final security.

Inherits Ethereum's security. Doesn't inherit Ethereum's fees. That's the whole idea.

Optimistic vs ZK Rollups

Two approaches to the same problem. Both work. Different tradeoffs.

Optimistic rollups. Arbitrum, Optimism, Base. Assume all transactions valid unless someone challenges them. Post data to Ethereum without a proof. Challenge window runs for 7 days. Nobody disputes, transactions finalize.

Fast to build. Captured most liquidity early because of it. Downside nobody loves: withdrawing back to Ethereum mainnet officially takes 7 days. Third-party bridges like Across get around this but add their own risks.

ZK rollups. zkSync, Starknet. Generate cryptographic proof that everything in a batch is valid. Submit proof to Ethereum. Ethereum verifies the proof not each transaction individually. Instant finality. No challenge period. Better long-term properties theoretically.

Harder to build. More expensive to run. Got to market later. Still catching up on ecosystem depth.

Optimistic rollups won the liquidity battle so far. ZK rollups have the more interesting technology. Both getting used.

Dencun Changed Everything

March 2024. Ethereum's Dencun upgrade introduced blob-based storage for rollup data.

Before it, L2s posted transaction data to Ethereum using expensive calldata. After, blobs. Much cheaper. Gets pruned after a few weeks since L2s only need recent data available briefly.

Fees on Arbitrum, Base, Optimism dropped over 90% overnight. Transactions that cost $0.50 suddenly cost $0.005. Entire category of small trades that made no economic sense became viable instantly.

Single upgrade that probably did more for real L2 adoption than years of marketing.

Where the Activity Actually Is

Base. Coinbase built it. Fastest growing L2 in 2025 by almost every metric. Around 46% of all L2 DeFi TVL by late 2025. $10 billion sitting there. Retail adoption driven by Coinbase's user funnel. Real usage, not just farming. Main criticism: sequencer run by Coinbase. Centralized. Not decentralized yet and Coinbase hasn't committed to a timeline.

Arbitrum. $16.6 billion TVL. Largest overall. Deep DeFi liquidity. Uniswap, Aave, serious protocol presence. More technically mature. Community governed through ARB token. Generally considered more serious about decentralization than Base even if neither has solved it.

Optimism. $6 billion TVL. Built the OP Stack that Base runs on. Also used by dozens of other L2s. Superchain vision: multiple chains sharing security and interoperability through OP Stack. More of an infrastructure play than a destination chain now.

zkSync and Starknet. ZK rollup camps. Lower TVL than optimistic competitors. Growing. zkSync did a large controversial airdrop in 2024. Starknet uses Cairo language instead of Solidity which is a real barrier for developers used to the Ethereum ecosystem.