Roughly 90% of Ethereum blocks since the Merge have been built by someone other than the validator who proposed them. That uncomfortable fact — that block production is effectively outsourced to a handful of off-chain builders via MEV-Boost — is the problem Glamsterdam, Ethereum's H1 2026 hard fork, was designed to address.
In Part 1 we walked through Pectra and the Fusaka roadmap. In Part 2 we covered PeerDAS and how data availability sampling cut L2 fees. Glamsterdam is the next hard fork in the queue, and it is the one that finally drags Maximal Extractable Value (MEV) inside the protocol while quietly making blocks faster to execute.
The scope was locked in early April 2026 after Devnet-4 stabilized, and Devnet-5 is now running with client teams. The fork name is a portmanteau: "Gloas" for the consensus-layer changes and "Amsterdam" for execution. The successor, already pencilled in for H2 2026, is called Hegota.
The MEV problem in one paragraph
MEV is the profit a block producer can extract by reordering, inserting, or censoring transactions. After the Merge, validators were not equipped to run sophisticated searcher pipelines, so Flashbots' MEV-Boost emerged as a sidecar: validators outsource block construction to specialist builders via relays, take the highest bid, and propose the resulting block. Cumulatively, post-Merge MEV has crossed 330,000 ETH — more than $1.8 billion — and roughly 80% of DeFi transactions now route through private RPCs to avoid being front-run. Flashbots Protect alone has shielded around $43 billion of DEX volume across about 2.1 million users since 2021, and CowSwap, the leading batch-auction DEX, is past $35 billion lifetime volume.
The market that grew up to handle this is concentrated. As of April 2026, the top five MEV-Boost relays deliver about 94% of payloads: Ultra Sound at 32%, Titan at 20%, bloXroute Max Profit at 19%, bloXroute Regulated at 14%, and Aestus at 9%. Three of the seven major relays enforce OFAC compliance. On the builder side, three firms — beaverbuild, Titan Builder, and rsync-builder — together win more than 80% of blocks.
This is the oligopoly Glamsterdam has to confront.
EIP-7732: Enshrined Proposer-Builder Separation
EIP-7732, authored in June 2024 by Francesco D'Amato, Barnabé Monnot, Mike Neuder, Potuz, Justin Traglia, and Terence Tsao, brings the MEV-Boost auction into Ethereum's consensus layer. Instead of trusting an off-chain relay to hold the builder's payload until the proposer commits, the protocol itself coordinates the handoff.
The flow is a three-phase commit-reveal:
- Bid. Builders submit signed bids for the next slot directly to the proposer, committing to a payment and a payload hash.
- Reveal. The winning builder reveals the actual execution payload. A new committee, the Payload Timeliness Committee (PTC), attests on-chain to whether the payload arrived on time.
- Validate. Execution clients verify state transitions as usual. If the builder withholds the payload, the PTC's attestation triggers a missed-slot penalty against the builder rather than the proposer.
The practical consequence is that relays — the trusted intermediaries everyone has been quietly uncomfortable with — become optional. A solo staker can compete for builder bids without trusting a Flashbots-operated server. Validator payments are enforced by consensus, not by a relay's word.
What ePBS does not do is dissolve the builder oligopoly. The same three firms that win 80% of blocks today have the same searcher relationships, order flow deals, and infrastructure tomorrow. ePBS removes a trust assumption and a censorship vector (OFAC-compliant relays lose their gatekeeping role), but it does not magically democratize block-building. That requires either inclusion lists — which proposers can use to force certain transactions into the next block, and which Glamsterdam ships in a basic form — or much harder work on encrypted mempools, which is years away.
If you stake at home or via a pool, the mechanics largely vanish into your client. We covered the staking landscape in our validators and LST guide; after Glamsterdam, that guide's MEV-Boost section becomes a historical footnote.
EIP-7928: Block-Level Access Lists
The second headline EIP is less politically charged but arguably bigger for performance. Block-Level Access Lists (BALs), authored in March 2025 by Toni Wahrstätter, Dankrad Feist, Francesco D'Amato and others, attach a manifest to every block declaring exactly which storage slots and accounts each transaction will read or write.
This is not a new concept at the transaction level — EIP-2930 access lists have existed since Berlin in 2021 — but doing it at the block level changes the economics. With a BAL, an execution client knows the full read/write set for the block before it starts executing. That enables three things:
- Parallel execution. Empirical analysis of mainnet shows that 60–80% of transactions in a typical block touch disjoint storage. With BALs, those transactions can be executed simultaneously across CPU cores instead of serially.
- Faster state access. Clients can prefetch storage trie nodes in parallel before execution begins, eliminating a major source of latency.
- Lighter stateless clients. Future stateless clients (Verkle, in Hegota or later) get a clean dependency manifest for free.
Average BAL overhead measured across mainnet replays is about 70 KiB per block — meaningful but well within bandwidth budgets, especially after PeerDAS landed in Fusaka. This is also why Glamsterdam can credibly raise the gas limit. The current target of 60M (recently bumped from 36M post-Pectra) is expected to move to 100M after Glamsterdam, with 200M as a longer-term ambition. Parallel execution turns gas-limit increases from a synchronous-CPU bottleneck into a much more tractable problem.
The supporting cast of EIPs
Glamsterdam is not just ePBS and BALs. The full bundle includes:
- EIP-7917 — deterministic proposer lookahead, so the network knows the proposer schedule further in advance. Useful for ePBS bid markets and for L2 preconfirmations.
- EIP-7708 — emit a log on every ETH transfer, finally giving indexers a reliable way to track native ETH movement without trace_call.
- EIP-8045 — exclude slashed validators from committees immediately rather than waiting for the exit queue.
- EIP-8037 / EIP-8038 — repricing of state-access opcodes (SLOAD, SSTORE, BALANCE, EXTCODE*) to better reflect post-BAL costs.
- EIP-7904 — broader opcode repricing aligned with parallel execution.
Notably absent: EIP-7782, which would have shortened slot times from 12 seconds to 6. It was deferred from Glamsterdam during the April scope lock — client teams wanted ePBS and BALs to bake before stacking another consensus-layer change on top. Expect it in Hegota.
What this means for users
If you are a regular user, the visible effects of Glamsterdam are modest but real. Gas should trend cheaper as the limit rises and parallelization absorbs more throughput. Censorship resistance should improve at the protocol layer because proposers can attach inclusion lists without relying on a non-OFAC relay. And private-RPC products like Flashbots Protect will keep working — ePBS does not eliminate the value of pre-trade privacy — but the trust surface shrinks.
For self-custody users, none of this changes how you hold ETH or interact with dapps. If you store ETH in a non-custodial wallet, Glamsterdam is invisible at the application layer; we wrote a separate guide on holding ETH safely in Zelcore that remains accurate.
For solo stakers, ePBS is a quiet upgrade in autonomy. You no longer need to trust a relay operator to hold your block hostage; the protocol does it for you.
The honest limits
Three caveats are worth stating plainly.
First, ePBS does not break the builder oligopoly. The same firms with the best searcher relationships and order-flow deals will keep winning bids. Glamsterdam shifts who settles the auction (the chain, not Flashbots) without changing who wins it.
Second, BALs help block construction but do not reduce L1 fees on their own. They enable a higher gas limit, and the gas limit raise is what users will feel. The two are linked but distinct.
Third, a 100M gas limit on L1 is still tiny compared to the throughput L2s deliver via blobs. For most user activity, the right answer remains a rollup. Glamsterdam makes L1 a better settlement layer, not a competitive execution layer.
Onward to Hegota
Glamsterdam is the medium-term fork. The long-tail roadmap — Single-Slot Finality, Verkle trees and stateless clients, the EVM Object Format (EOF), and the deferred 6-second slots — lands in Hegota and beyond. That is where Ethereum stops being a chain that can be run by individuals in principle and starts being one that is genuinely cheap to run in practice.
In Part 4 we will work through that long-tail: SSF, Verkle, EOF, and what 6-second slots actually feel like for users and L2s.
