Skip to main content

Glamsterdam Slips: Ethereum's MEV Reform Hits Engineering Reality as ePBS Runs Late

· 11 min read
Dora Noda
Dora Noda
Software Engineer

For the first time in Ethereum's accelerated 2026-2027 fork cadence, the roadmap has blinked. In mid-April 2026, core developers publicly acknowledged what client teams have whispered for weeks: Enshrined Proposer-Builder Separation — the single most ambitious piece of the Glamsterdam hard fork — is "trickier than anticipated," and the original May-June mainnet window is almost certainly out of reach. The slip pushes Glamsterdam toward Q3 or Q4 2026, narrowing the gap with the already-scheduled Hegota fork and reopening a question Ethereum thought it had settled: can a five-client base layer still upgrade at the pace a post-Pectra L2 economy demands?

The answer matters far beyond Ethereum's own ecosystem. Every week that ePBS stays in devnet is a week that the current Flashbots-relay oligopoly keeps mediating tens of billions in annual MEV flows, a week that L2s keep pricing blobs against a supply curve the fork was supposed to loosen, and a week that Solana — which locked in 98.27% validator approval for its Alpenglow consensus overhaul in September 2025 — extends its lead on a measurably faster monolithic upgrade cadence. Glamsterdam was never just another hard fork. It was Ethereum's answer to the "Fast L1" thesis. The answer is now late.

The Two EIPs That Define Glamsterdam​

Glamsterdam — a portmanteau of Gloas (the consensus layer codename) and Amsterdam (the execution layer codename, following the Devcon-host-city tradition) — is anchored by two headliner EIPs that would together rework how Ethereum produces blocks.

EIP-7732 (ePBS) splits the consensus block from the execution block at the protocol level. Today, validators running MEV-Boost hand off block construction to a marketplace of off-protocol builders via centralized relays. Flashbots' own December 2024 migration of all builders and orderflow to BuilderNet was itself an admission that the relay architecture creates concentration risk. ePBS enshrines the split: proposers and builders become first-class protocol participants that interact directly, no relay required.

EIP-7928 (Block-level Access Lists) makes blocks declare their read/write footprint — accounts and storage slots touched — up front. Clients can then parallelize execution and verification, cutting the time a block sits in propagation. Combined with ePBS, the pair targets a ~50% reduction in effective block latency.

On paper, these complement each other cleanly. In implementation, they are interacting with every other piece of the post-Pectra stack — including EIP-7251 MaxEB validator consolidation, already shipped in Pectra — in ways the original specifications did not fully anticipate.

What Actually Slipped​

The April 16, 2026 All Core Developers Consensus call (ACDC #177) made the engineering reality plain. Until this month, testing was bifurcated across two separate networks: epbs-devnet for proposer-builder separation, and bals-devnet for access lists. The "generalized devnet" that merges them — the first environment where every Glamsterdam component coexists — was only greenlit after Lighthouse and a handful of other consensus clients produced stable builds during the April interop window.

"ePBS Devnet Zero is less about performance and more about interoperation," one client engineer noted in the call summary. Translation: we are still discovering what breaks when five consensus clients and five execution clients each interpret the spec independently. Early runs produce mostly empty blocks — not a structural flaw, but a signal that the happy path is still being paved, let alone the adversarial one.

The timeline consequences compound:

  • Devnet Zero must stabilize across Lighthouse, Prysm, Teku, Nimbus, and Lodestar on the consensus side, plus Geth, Nethermind, Besu, Erigon, and Reth on the execution side.
  • Devnet One and Two need to surface interaction bugs between ePBS, BALs, gas repricing, and the 25+ non-headliner EIPs under consideration.
  • Public testnets (Holesky successor, Sepolia) need at least 6-8 weeks of shadow-fork activity before any mainnet target date can be credibly set.
  • Client releases must be cut, audited, and distributed to stakers with enough lead time that a fork-choice issue on mainnet doesn't become a liveness incident.

Each step is measured in weeks, not days. A May-June launch required Devnet Zero to be stable in February. It wasn't. June-July required stability in March. It wasn't. By April, the math simply doesn't close — hence the quiet acknowledgment that the headline fork is sliding into H2 2026.

The Client-Diversity Paradox​

Ethereum's five-client execution diversity is arguably its most valuable credibly-neutral property — a single-client bug cannot take down the network. But that diversity is also why Glamsterdam is hard.

Every implementation team must independently build, test, and stabilize ePBS. Each surfaces different edge cases. The April ACDC minutes describe "pending pull requests" across multiple clients and note that consensus-specs changes may block ePBS Devnet Two entirely until resolved. This is precisely the coordination tax that a single-implementation chain doesn't pay.

Compare with Solana's cadence. Alpenglow, Solana's consensus replacement targeting 100-150ms deterministic finality, passed a validator vote with 98.27% support — one of the strongest mandates in the network's history — in September 2025. The rollout plan is simple: Anza's Agave client ships the upgrade in Q3 2026; Firedancer, Jump Crypto's second implementation, follows. Because Solana effectively had one production client until Firedancer crossed 20% mainnet stake earlier this year, its coordination surface is a fraction of Ethereum's.

This is not a neutral observation. Monolithic chains ship faster. Multi-client chains ship safer. Glamsterdam's slip is the price of Ethereum's architectural choice — and for the first time since the Merge, the market is weighing whether that price is still worth paying.

What the Delay Means for L2s and MEV​

The Pectra fork shipped in 2025 on schedule. Fusaka added PeerDAS and expanded blob capacity soon after. L2 teams built their 2026 capacity plans around Glamsterdam landing in Q2, with further blob expansion and MEV redistribution arriving to meet surging rollup demand.

Those plans now need to be redrawn.

Blob pricing. Glamsterdam was expected to lift blob-per-block counts further — potentially to 72 or more — giving optimistic and ZK rollups significantly more cheap data availability. A Q3-Q4 slip means 2-4 more months of tighter blob supply, which matters most for rollups that have already saturated their fee lanes.

MEV redistribution. Today's MEV-boost architecture rewards the largest builders disproportionately. Super-linear returns on orderflow aggregation mean that once a builder secures a lead, the economics push toward monopoly. ePBS doesn't eliminate MEV, but it removes the relay as a coordination chokepoint — which should, in theory, redistribute a fraction of today's extraction back to proposers and, downstream, to stakers. Every month of delay is a month that existing MEV dynamics persist.

L2 competition. Base has already hit 2-second block times. Arbitrum and Optimism are shipping their own sequencer and DA improvements on independent cadences. The longer the L1 bottleneck persists, the more L2 roadmaps diverge from any unified L1 upgrade — and the more the "rollup-centric roadmap" becomes N separate rollup roadmaps that happen to settle to the same base layer.

The FOCIL Question and the Hegota Squeeze​

One of the April ACDC's most consequential decisions was moving Fork-Choice Inclusion Lists (FOCIL) out of Glamsterdam and into Hegota, where it has been formally selected as the consensus-layer headliner for late 2026.

FOCIL is Ethereum's censorship-resistance answer — a mechanism that lets a committee of proposers collectively enforce transaction inclusion, so no single builder can systematically exclude payloads (OFAC-sanctioned addresses, for example). The Base engineering team had publicly warned that bundling FOCIL with ePBS would push Glamsterdam past 2026 entirely. Pulling it out buys schedule relief for Glamsterdam — at the cost of compressing the window between Glamsterdam's eventual mainnet activation and Hegota's.

If Glamsterdam ships in Q4 2026 and Hegota targets late Q4, the gap could be as little as 6-8 weeks. That's a short runway for validators, block builders, staking providers, and L2 teams to absorb one protocol change before the next arrives. The Hegota roadmap also includes significant state-bloat work — early discussions have focused on Verkle Trees, which would reduce node hardware requirements but require their own extensive testing.

The scenario Ethereum is now quietly planning for: two major forks landing within a quarter of each other, on a test matrix that Solana ships in a single coordinated release.

Why This Isn't a Crisis — Yet​

It would be a mistake to read Glamsterdam's slip as existential. Ethereum's fork process is working exactly as designed. ePBS is being delayed not because coordination failed, but because coordination caught a problem — spec ambiguities and implementation edge cases — before they reached mainnet. The alternative, shipping on schedule and firefighting live fork-choice bugs with $400B+ in settled value at stake, is immeasurably worse.

The deeper question is whether Ethereum's current cadence is sustainable as the per-fork complexity grows. Glamsterdam is not just two EIPs; it's two headline features plus 25+ smaller changes, each interacting with a post-Pectra validator set that already has maxEB consolidation, PeerDAS-enabled blobs, and a mature L2 ecosystem on top. Each new fork makes the test matrix wider.

Proposals to split Glamsterdam itself — shipping BALs in Q3 and ePBS in Q1 2027 — have been floated and so far rejected. The counter-argument is compelling: BALs without ePBS delivers marginal value, because the parallel-execution gains are bottlenecked by the current builder architecture. The features are genuinely complementary, and unbundling them saves schedule at the cost of the upgrade's coherence.

What to Watch in the Next 90 Days​

For developers, stakers, and anyone pricing L2 block space for the second half of 2026, three signals matter:

  1. Devnet Zero stability. If the generalized devnet runs 30+ days without a critical incident by early June, a Q3 2026 target becomes credible. If it doesn't, Q4 is the floor.
  2. Public testnet activation. Holesky's successor and Sepolia need to fork at least 8 weeks ahead of mainnet. No testnet fork date = no mainnet fork date.
  3. Hegota EIP selection. The February 2026 Hegota headliner was FOCIL; the execution-layer headliner is still being debated. Whatever is chosen will further constrain how tightly the two forks can be sequenced.

Ethereum's 2026-2027 roadmap was always ambitious. Glamsterdam's slip is the first concrete reminder that ambitious does not mean on-time. For a network whose credibility rests on doing hard things carefully, a measured delay is not a failure. It is the feature.

BlockEden.xyz operates enterprise-grade Ethereum infrastructure across Pectra, Fusaka, and every testnet staging Glamsterdam's fork rules. If you're building against a moving upgrade target and need nodes that track mainnet, testnets, and devnets in parallel, explore our Ethereum API services — infrastructure designed for teams who can't afford a forked state on the wrong side of a spec change.

Sources​

Share on Twitter