Ethereum's Glamsterdam upgrade — the protocol's largest change since the 2022 Merge — has slipped three times since June 2026 and remains stuck in devnet testing. The Sepolia testnet activation, originally scheduled for August 3, now targets October 6 at 13:53 UTC (Epoch 351232). Mainnet deployme...
"You could get the whole network to stop." — Stefan Starflinger, Ethereum Foundation DevOps Engineer, describing a consensus-layer bug discovered on Glamsterdam Devnet-8
Ethereum's Glamsterdam upgrade — the protocol's largest change since the 2022 Merge — has slipped three times since June 2026 and remains stuck in devnet testing. The Sepolia testnet activation, originally scheduled for August 3, now targets October 6 at 13:53 UTC (Epoch 351232). Mainnet deployment, once aimed at H1 2026, has drifted to Q4, with December discussed informally among core developers.
The upgrade bundles two headline EIPs: EIP-7732 (Enshrined Proposer-Builder Separation, or ePBS) and EIP-7928 (Block-Level Access Lists). Together, they replace Ethereum's dependence on off-chain MEV relay infrastructure with protocol-native block construction, and introduce parallel transaction execution. The technical payoff is a path to a 200-million gas-limit floor — more than 3x the current 60 million — and theoretical throughput approaching 10,000 TPS on Layer 1.
The stakes are measurable. Five builders currently control 96.7% of all MEV blocks. MEV-Boost relay infrastructure routes over 97% of Ethereum block proposals through a handful of trusted intermediaries. ePBS moves that trust on-chain. Whether the testing delays signal healthy caution or structural complexity depends on Devnet-11, scheduled to launch September 14 with 84,000 validators.
Glamsterdam's testing history illustrates the tension between Ethereum's new faster-upgrade cadence and the complexity of protocol-level changes.
| Milestone | Original Target | Current Target | Status | |-----------|----------------|----------------|--------| | Final Devnet (Devnet-11) | — | Sept. 14, 2026 | Pending launch | | Gloas Fork (Devnet-11) | — | Sept. 16, 2026 | Pending | | Sepolia Testnet | Aug. 3, 2026 | Oct. 6, 2026 | Conditional | | Hoodi Testnet | TBD | Post-Sepolia | Not scheduled | | Mainnet | H1 2026 | Q4 2026 (~Dec.) | Tentative |
The delays stem from concrete failures. Devnet-8 exposed a consensus-layer vulnerability where blocks repeated parent hashes — a bug that Starflinger said could halt the entire network. Devnet-9, running 1,000 validators, failed to finalize blocks normally. Execution-layer EIP-8037 required implementation patches from each client team. Developers skipped Devnet-10 entirely, jumping to Devnet-11 with 84,000 validators — 84x the count on Devnet-9 — to approximate mainnet conditions.
The ACDC #186 developer call on September 3 set the October 6 Sepolia date, contingent on Devnet-11 stability. If Devnet-11's Gloas fork on September 16 finalizes cleanly, October 6 becomes credible. If not, the date will slip again.
No All Core Developers meeting has locked a mainnet block height. Developers have stated that correctness takes priority over any calendar target.
EIP-7732 enshrines Proposer-Builder Separation directly into Ethereum's consensus layer. Under the current system, block production relies on MEV-Boost — off-chain software maintained primarily by Flashbots — to mediate between validators (proposers) and block builders. The proposer selects the highest-value block from builders via a trusted relay, without seeing the block's contents.
ePBS moves this exchange on-chain. Builders submit sealed bids and block-header commitments directly to the network. The proposer picks a bid; the builder then reveals the execution payload. The protocol enforces the exchange, eliminating the need for trusted relay middleware.
The technical change expands the data propagation window from approximately 2 seconds to roughly 9 seconds. This additional time enables the network to handle larger blocks safely, which directly supports the gas-limit increase.
The ePBS builder market is strictly opt-in. The existing MEV-Boost relay infrastructure continues to function as-is. There is no forced migration. This design choice reduces coordination risk but raises the question of adoption speed — builders and proposers must voluntarily switch to the on-chain system for its decentralization benefits to materialize.
EIP-7928 introduces Block-Level Access Lists (BALs), which declare in advance which accounts and storage slots a block's transactions will touch. This allows validator clients to preload state data and execute unrelated transactions in parallel rather than sequentially.
Historical analysis of Ethereum transaction data shows that 60–80% of transactions access disjoint storage slots, making them candidates for parallelization. The remaining 20–40% can be processed using post-transaction state diffs included in Execution BALs.
Average BAL size is approximately 70 KiB per block, according to historical data analysis — compact enough to avoid significant bandwidth overhead.
World Chain, a Layer 2 network, has already deployed EIP-7928 access lists in production. Its implementation streams access-list data incrementally every 200 milliseconds through its flashblock architecture, achieving processing capacity of up to 1 gigagas per second on standard cloud infrastructure.
The current MEV ecosystem is concentrated. As of September 9, 2026, relay market share stands at:
| Relay | Market Share | |-------|-------------| | Ultra Sound Money | 33.03% | | Titan Relay | 29.70% | | BloXroute Regulated | 25.09% | | Aestus | 8.06% | | Flashbots | 2.23% | | Agnostic Relay | 0.97% | | EthGas | 0.93% |
At the builder level, concentration is more acute: five builders control 96.7% of all MEV blocks as of January 31, 2026, with the top builder holding 27.9% market share.
Flashbots launched BuilderNet in December 2024, migrating to a shared-order-flow model using Trusted Execution Environments (TEEs). By January 2026, BuilderNet produced 25.5% of Ethereum blocks. However, the TEE requirement creates hardware barriers to entry, and the cooperative design removes competitive bidding pressure among builders.
ePBS does not solve builder centralization — capital-intensive builders with sophisticated infrastructure may still dominate block construction. What it does is remove a single point of trust failure at the relay layer. Relay operators currently process block headers and payments without protocol-level enforcement. Under ePBS, those guarantees are cryptographic and consensus-enforced.
Glamsterdam's combined improvements — the extended data propagation window from ePBS and the parallel I/O from BALs — are designed to support a gas-limit increase from the current 60 million to 200 million.
This is not an automatic change. Ethereum's gas limit is set by validator consensus — each validator votes on the target gas per block. Glamsterdam removes the technical bottlenecks that make higher limits unsafe; validators must then coordinate to raise the actual limit.
Developers are targeting approximately 120 GiB of annual state growth at a 200-million gas limit, a level intended to keep node operation within reach of consumer-grade hardware. The current gas limit produces substantially less state growth, so the 200M target represents a deliberate trade-off between throughput and decentralization.
At 200 million gas, Ethereum Layer 1 would approach theoretical throughput of 10,000 TPS, though actual throughput depends on transaction composition (simple transfers consume less gas than complex smart-contract calls).
Ethereum currently has approximately 38.9 million ETH staked across more than 897,000 active validators — about 32% of circulating supply. Base staking APR has compressed to roughly 2.78%, down from over 4% in earlier years, as increasing participation thins per-validator rewards.
MEV adds 10–30% on top of base staking yields. MEV-Boost validators earn approximately 3.3–3.8% all-in, according to Ethereum staking data. The validator entry queue ballooned to 3,589,414 ETH with a 62-day wait time as of May 2026.
Glamsterdam affects institutional stakers through two channels, according to Figment's analysis:
Block-building trust assumptions. ePBS standardizes the block-production pipeline at the protocol level, giving institutional operators a more transparent process than the current relay-dependent system. For staking providers managing validator fleets on behalf of allocators, reduced trust dependencies lower operational risk.
Exit queue improvements. EIP-8061, also included in Glamsterdam, increases exit and consolidation churn rates. For institutional operators, faster exit processing reduces time-to-liquidity risk — a consequential cost for large positions. According to Figment, Glamsterdam "will materially improve exit liquidity" for institutional validators.
Glamsterdam's gas-limit expansion and improved blob handling affect Layer 2 economics. Average fees on Arbitrum, Base, and Optimism declined 99.16% from $0.18 in Q1 2024 to $0.0015 in Q1 2026. Ethereum mainnet median fees fell 99.68% over the same period, from $3.79 to $0.012.
As of May 2026, 73 active Layer 2 networks hold more than $40 billion in user funds, up from under $4 billion in 2023. Blob capacity currently sits at approximately 375 KB per block.
A 200M gas limit would further reduce blob-posting costs for Layer 2 operators by increasing available block space. The expanded data propagation window under ePBS also provides more time for blob data transmission, reducing the bandwidth pressure that currently constrains blob capacity increases.
Glamsterdam is Ethereum's most consequential protocol change since the Merge, and its testing struggles reflect that scope. The upgrade addresses two measurable structural problems: concentrated relay infrastructure that mediates 97%+ of block proposals, and sequential transaction execution that constrains throughput at 60 million gas.
The technical architecture is sound in specification. ePBS removes an off-chain trust layer; BALs enable parallelism that real-world data supports. World Chain's production deployment of EIP-7928 provides empirical evidence for the parallel-execution thesis.
The open question is not whether these changes improve Ethereum. It is whether the implementation complexity — evidenced by consensus bugs on Devnet-8, finality failures on Devnet-9, and three schedule slips — can be resolved before the testing window narrows further. Devnet-11 on September 14 is the next data point. December mainnet remains achievable but unconfirmed. No block height has been locked.
The economic value redistribution is clear: relay operators lose protocol-enforced monopoly rents; validators gain a native builder market; Layer 2 operators benefit from cheaper blob posting. Whether that value materializes in Q4 2026 or later depends on what happens when 84,000 validators try to finalize a Gloas fork on September 16.