727 FinanceEthereum scaling is rewriting the underlying logic, and the once-hot Rollup narrative is now facing reassessment.
Written by William M. Peaster, Bankless
Compiled by Saoirse, Foresight News
The Ethereum Name Service (ENS) has just completed the largest upgrade in its history. Notably, this upgrade was deployed directly on the Ethereum mainnet, rather than on a Layer 2 network that had been in preparation for years.
On Tuesday, October 27, ENS v2 officially launched on L1. Yet as early as nine months prior, ENS had already abandoned Namechain, the Rollup chain originally intended to host this upgrade.
ENS is not an isolated case. Over the past year, an increasing number of projects have chosen to exit Layer 2 networks, shelve plans to build their own L2s, or shut down dedicated chains already launched, migrating their core contracts back to Ethereum L1.
In short, this is a wave of return to the mainnet.
The underlying backdrop to this trend: Ethereum's Gas limit (the amount of computation a single block can carry) was raised from 30 million to 60 million in 2025, and L1 transaction fees fell sharply as a result.
Ethereum raised the block Gas limit from 30 million to 60 million in phases in 2025, while flagging key events such as Synthetix's post, Vitalik's article, and the launch of ENS v2 on L1; the Glamsterdam upgrade has already reached a 200 million Gas limit on the Sepolia testnet, but the mainnet launch date remains undecided.
Meanwhile, the cost of building a public chain from scratch has proven to be very high. This month alone, two Layer 2 networks, Blast and Abstract, announced their shutdowns within four days of each other.
Let us walk through the full picture, starting with this latest domino to fall.
Why ENS chose to stay on Ethereum
As you all know, ENS can turn long wallet addresses into readable, memorable domain names like "william.eth".
ENS v2 is a complete reconstruction of the entire protocol. Simply put, every ENS domain can now establish an independent registration system for its subdomains; domain holders gain granular permission controls, and can pay related fees using stablecoins such as USDC.
But the most worth discussing aspect of this article is where v2 was deployed. Back in May 2024, I wrote "Getting Started with ENS v2," introducing ENS Labs' plan to migrate the protocol to a Layer 2 network, which later materialized as the dedicated Rollup — Namechain. At the time, I believed this upgrade might open a new phase for ENS going mainstream.
ENS v2 did open a new phase — just not on L2. On February 6, ENS founder Nick Johnson explained the pivot: "Ethereum L1 is scaling, and scaling far faster than most people expected two years ago."
The facts bear this out: alongside Ethereum's scaling, the Gas cost of ENS domain registration has dropped 99% over the past year. In addition, Nick argued that Layer 2 networks introduce additional trust assumptions that do not exist on L1 itself.
Vitalik Buterin also publicly endorsed the decision the same day: "This is a wise decision!"
The dominoes falling one after another
ENS is the latest member of this "return list," but by no means the first.
A summary of projects that migrated back to the Ethereum mainnet from Layer 2 networks (or originally planned self-built L2s) over the course of 2026, noting the platform migrated from, the timing, and migration progress.
Synthetix was an early mover. It was among the first to launch on an Ethereum Rollup back in 2021, yet in June 2025 it published the article "L2s Are Dead, Long Live Ethereum," which pointed directly at pain points such as liquidity fragmentation.
Synthetix then gradually wound down its L2 deployments that summer, and its new perpetual futures exchange launched on L1 in December. Founder Kain Warwick later admitted: "I spent two years pushing the L2 narrative, but that path didn't work for Synthetix, so we chose to return to the mainnet."
Then in February, Vitalik Buterin published an article arguing that while L1 itself continues to scale, the path of Layer 2 networks toward full decentralization is far slower and harder than originally envisioned. He concluded that the original vision of Layer 2 networks, and their positioning within the Ethereum ecosystem, no longer holds.
And three days later, ENS halted the Namechain project.
This summer, Superseed became another typical case. Its original core mechanism relied on fees from its own Layer 2 chain to offset user loans. But the project officially deprecated that chain on August 15, and is currently migrating its entire ecosystem back to the Ethereum mainnet, replacing the chain-fee model with an application-layer yield strategy.
In September, Trueo, the prediction market protocol behind the DefiLlama market, announced its "return to Ethereum" from Base. Vitalik expressed support for the plan.
October also saw a concentrated wave of similar moves:
ShapeShift: migrated its rFOX staking program from Arbitrum to L1 on October 1, because the vast majority of FOX tokens are themselves custodied on the mainnet.
Blast: announced a gradual shutdown on October 2, as on-chain revenue could not cover operating expenses.
Nillion: launched the Blacklight network on L1 on October 5, with a roadmap showing the phased retirement of its own L2.
Abstract: officially announced on October 6 that it would shut down on December 15, judging that a standalone public chain model aimed solely at ordinary users is not sustainable.
Some NFT projects within the Abstract ecosystem, such as Gigaverse's ROMs and BEARISH, have also announced migration to L1.
Core state returning to L1?
Behind these migrations lies a deeper trend: projects are separating state storage from computation execution.
Take Synthetix's new perpetual contracts as an example: orders are matched off-chain, but user funds are custodied on Ethereum L1, and trades are ultimately settled on L1.
Then look at Derive V3, a project I analyzed earlier this month: to ensure trading speed, order matching is done off-chain; risk calculations are verified via zero-knowledge proofs; and user funds are held in Ethereum L1 contracts.
ENS v2 follows the same approach: the domain NFTs themselves are stored on L1; subdomains can achieve zero-Gas resolution on L2 or off-chain via the CCIP-Read standard, lowering usage costs. CCIP-Read allows domains to point to data stored elsewhere.
In other words, the high-value core components (asset custody, ownership verification, transaction settlement) return to L1, while computation-intensive heavy workloads run in the lowest-cost, fastest environments. This is fundamentally different from the traditional Rollup model: traditional Rollups carry both state and computation, and users must go through a cumbersome exit process to withdraw their assets.
A comparison of three architectures: traditional Rollups store both state and execution computation on Layer 2; the new hybrid applications place core state such as funds and settlement on Ethereum L1 while moving complex computation off-chain; and L1.5 ephemeral Rollups keep private state in the user's wallet, with L1 storing only state commitments and validity proofs, with no persistent data storage in between.
Of course, some technical approaches are gradually dissolving the boundary between L1 and other environments.
Take the Ethereum Economic Zone (EEZ), which I introduced back in March: its goal is to achieve synchronous composability between Ethereum and its Rollups — that is, contracts on different chains can interact directly within the same transaction.
EEZ is not about stripping applications away from L2, but about making Rollups feel like part of L1. Related plans are being implemented: in August, GnosisDAO voted to convert Gnosis Chain from an independent L1 into an EEZ-architecture Rollup.
There are also newer, not-yet-battle-tested ideas. This week, researcher namnc proposed ephemeral rollups, also called L1.5, on the Ethereum Research forum, taking the state-computation separation idea to its extreme.The logic of this design is: L1 only stores user state commitments and verifies all proofs; user wallets custody private state themselves; computation consists of short-lived tasks, and no state persists after execution ends. Since L1.5 does not persistently store state, there is naturally no asset exit process either.
However, this wave of return has one precondition: Ethereum L1 must continue to maintain sufficient compute capacity. If projects flood in faster than L1 scales, Gas fees will surge again, and the entire situation will reverse.
Ethereum's major scaling upgrade Glamsterdam had its testnet launch plan postponed last week, and the Ethereum Foundation said mainnet activation timing is not yet determined.
But Glamsterdam has already raised the Gas limit on testnet to 200 million, more than three times the current 60 million limit. Judging by the fee-reduction effect of the previous Gas limit doubling, if this is subsequently implemented on mainnet, L1's capacity will be considerable.
No matter how things evolve, the trend is clear: more and more projects are putting their most core state data back on L1, and ENS is just the latest member.