The L2 Endgame: Why Most Rollups Are Already Dead, and What Comes After
Introduction: the consolidation that finally happened
In February 2026, Vitalik Buterin published a post arguing that the original rollup-centric Ethereum roadmap "no longer makes sense" in its 2020 form, a remarkable concession from the architect of that roadmap. The market had already arrived at the same conclusion. Base, Arbitrum and OP Mainnet now control roughly 83% of all L2 DeFi TVL: Base around 46.6%, Arbitrum 30.9%, OP Mainnet around 6%. The remaining fifty-plus rollups fight over single-digit shares, and most of them lost active users sharply after their token-generation events. The Block's 2026 Layer 2 Outlook describes the dynamic in clinical terms: "most new launches have become ghost towns shortly after airdrop farming cycles".
For enterprise architects who have spent two years being told the L2 choice would matter forever, the more honest 2026 framing is this: the L2 wars are over, the winners are obvious, and the next architectural decision is not which L2. It is how to abstract the L2 layer entirely.
Where the data sits in 2026
A working snapshot as of Q1 2026:
- Arbitrum One holds approximately £13 billion bridged TVL, with 250,000 to 300,000 daily active addresses. It remains the deepest DeFi liquidity venue and the home of perpetuals, derivatives and complex on-chain finance.
- Base is the consumer-onboarding leader at around £8 billion TVL and over one million daily active addresses, fed by Coinbase's distribution. It is the chosen home of the consumer agent economy (CLANKER, Virtuals, Zora, Farcaster frames).
- OP Mainnet sits at around £4.8 billion TVL and 82,000 daily actives. Optimism's strategic value is no longer in OP Mainnet itself but in the OP Stack, which now powers Base, Worldchain, Ink, Mode and dozens of other chains under the Superchain umbrella.
- zkSync Era, Linea, Scroll, Polygon zkEVM and Starknet collectively hold around 10% of L2 TVL. ZK rollups have not won on retail volume; they are positioning as the institutional and high-value-transaction layer (zkSync's Prividium for enterprises, Starknet for ZK-native UX, Polygon zkEVM for permissioned deployments).
- Stage classification. As of January 2026, Arbitrum One, OP Mainnet and Base have all reached L2BEAT Stage 1: permissionless fraud proofs are live. Most other rollups remain Stage 0, meaning users still depend on a trusted operator.
The structural lesson is brutal: distribution beat technical differentiation. Base won because Coinbase has 100 million plus verified users. Arbitrum won because it captured DeFi's compounding liquidity flywheel before zkEVMs were ready. Optimism won the "many-chains" architectural bet. Almost everything else is dead-chain-walking.
The data availability layer is now the second front
If the execution layer has consolidated, the data availability layer is where the next round of competition is happening. EIP-4844 (March 2024) reduced L2 DA costs by 50% to 90% by introducing blob space. Throughout 2025 the major rollups optimised aggressively for blob-based posting; in January 2026 Ethereum increased blob capacity again, with PeerDAS on the roadmap to scale data availability sampling materially further.
External DA layers have specialised:
- Celestia holds roughly 50% of the modular DA market and processed over 160 GB of rollup data through early 2026. The Matcha upgrade will lift block size to 128 MB; the experimental Fibre Blockspace protocol targets 1 Tbps blockspace. At end-2024, Celestia generated around £180/day in blob fees; that figure has grown roughly 10x.
- EigenDA has reached around 100 MB/s throughput using a Data Availability Committee secured by EigenLayer restaking, attractive to Ethereum-aligned rollups that want inherited security without on-chain blob costs.
- Avail has secured integration commitments from Arbitrum, Optimism, Polygon, StarkWare and zkSync, positioning as the multi-ecosystem neutral DA.
Cost differentials are not marginal. When Eclipse posted to Celestia in 2025, it paid around £0.06 per megabyte versus £3.1 per megabyte for Ethereum blobs, 55x cheaper. That structural advantage is why most non-Ethereum-aligned rollups now post to a modular DA layer; that structural disadvantage is why Vitalik's recent posts have argued for accelerating Ethereum's native DA scaling.
Chain abstraction: the layer the architects should actually be building for
The technically interesting story of 2026 is not which chain wins. It is the abstraction stack making the chain question disappear. Three primitives are converging:
- Account abstraction. ERC-4337 is now mature, with over 50 million UserOperations recorded across chains by April 2026 (BundleBear data). EIP-7702, shipped in Ethereum's Pectra upgrade in May 2025, lets any existing externally-owned account temporarily delegate to smart-account code, bringing batched operations, gas sponsorship, session keys and social recovery to the existing wallet base without forcing migration.
- Cross-chain intents. ERC-7683, finalised in late 2024 and authored by Across and Uniswap Labs, defines GaslessCrossChainOrder and OnchainCrossChainOrder, structured intents that solver networks fulfil across chains. Across, Eco and LI.FI all support ERC-7683; UniswapX and CowSwap operate on similar intent-based models.
- Solver and liquidity networks. Pre-positioned market-maker inventory across chains, increasingly coordinated by shared sequencing efforts like Espresso, lets a user sign a single intent ("send 1,000 USDC from Arbitrum to Base") and have it executed cheaply on whichever path is optimal.
The combined effect: a 2026 user, or a 2026 enterprise treasury, signs an intent. The solver network handles bridging, gas, chain selection. The application has no idea which chain settled the operation. For the architect, this is the layer that matters. A new enterprise integration in 2026 should not be designed against a specific L2; it should be designed against an intent layer that abstracts the L2 layer entirely.
Three architect-level decisions for 2026 to 2027
Decision one: build for ERC-7683 intents, not RPC endpoints. The 2024 architecture pattern (a wallet, an RPC provider, a chain selection screen) is being replaced by a wallet, a paymaster, an intent and a solver. Enterprise Web3 integrations that bake chain selection into UX will look as dated by 2027 as websites that asked you to choose a server.
Decision two: pick a settlement chain on regulatory and liquidity grounds, not technical ones. The technical performance gap between Arbitrum, Base and OP Mainnet is now within margin. The DAU gap between Base and the rest is not. The institutional KYC posture of Coinbase-backed Base, the depth of DeFi liquidity on Arbitrum, and the public-goods governance of Optimism's Superchain are the actual differentiators in 2026.
Decision three: assume rollup decentralisation is still incomplete. Despite Stage 1 progress at the top three, sequencer centralisation remains the rule, and ZK provers are still mostly centralised. For mission-critical enterprise workflows, this is a real risk that must be priced. Shared-sequencing networks (Espresso, Astria) and based rollups (which inherit Ethereum sequencing directly) are the two paths to honest decentralisation; both are still maturing.
The contrarian call
The conventional 2024 to 2025 narrative was that "thousands of app-specific rollups" would proliferate, each with its own token. The 2026 reality is that fewer than ten general-purpose rollups will absorb almost all the value. App-specific chains will exist (Immutable X for gaming, dYdX v4 for derivatives, Hyperliquid for perps, Plume for RWAs) but as a category the "RaaS-as-default" thesis has underdelivered. The next 18 months will see continued consolidation, not proliferation. Enterprise architects pitched "your own appchain" should ask: against this consolidation, where does the volume actually come from?
Conclusion
The L2 question has answers now. Base for consumer distribution, Arbitrum for DeFi depth, OP Stack for many-chains coordination, zkSync's Prividium for enterprise privacy, Starknet for ZK-native execution. The architectural bet for 2026 to 2028 is not on a winner among them. It is on the abstraction layer that lets enterprises stop choosing.
Designing your L2 architecture? Our architects can model the trade-offs for your specific workload. Talk to our team.