Glamsterdam Enters Final Phase as the quest to Push Ethereum L1 Toward 10000 TPS Draws Closer

Glamsterdam is Ethereum's most substantial hard fork since the Merge, currently running its full EIP slate across multi-client developer networks ahead of a mainnet activation targeted for the second half of 2026, most plausibly around the end of August. Rather than ceding all throughput to rollups, it attacks the base layer's own execution ceiling through three changes: EIP-7928 (Block-Level Access Lists), which declares each transaction's data dependencies upfront to enable parallel execution and support a block gas-limit increase from 60 million toward 200 million; EIP-7732 (Enshrined Proposer-Builder Separation), which moves block-building into the consensus layer and reduces reliance on centralized off-chain relays; and EIP-7904, which rebenchmarks gas costs against real computational resources to lower fees and discipline state growth. The associated targets; a path toward 10,000 transactions per second and an approximately 78.6% reduction in baseline Layer 1 gas are design goals rather than guaranteed outcomes, and both depend on workloads and validator choices that real-world conditions may not favor. Here we examine the architecture behind each change and the reasons the headline figures may prove optimistic.

For several years the scaling answer was treated as settled doctrine. Layer 1 would remain slow, secure, and decentralized, while throughput migrated to rollups. Dencun gave those rollups inexpensive blob space, and Fusaka extended data availability further through PeerDAS. The rollup-centric thesis hardened into orthodoxy.

Glamsterdam complicates that account. It does not abandon rollups so much as confront the base layer's own execution ceiling, the constraint the orthodoxy had quietly written off as unfixable. Ethereum Foundation developers are presently running the full slate of proposed EIPs across multi-client developer networks glamsterdam-devnet-5 and devnet-6 as of June 2026. One Foundation devops engineer characterized the effort as probably the largest fork since the Merge and described the current stage as the last phase before the work of hardening and shipping to public testnets.

That framing deserves a precise reading. The characterization of Glamsterdam as the largest fork since the Merge is well supported. The suggestion that mainnet is imminent is less so. The heavy multi-client devnet work represents the step before public testnets, not the step before mainnet, and the schedule has already slipped once, from an original June 2026 goal to a second-half target now centered on roughly the end of August. No mainnet block height has been locked. The ambition is real; the countdown is shorter in the telling than in the engineering.

The headline targets attached to this fork are aggressive: a path toward 10,000 transactions per second on Layer 1 and a baseline gas-cost reduction of approximately 78.6%. Both are design targets rather than demonstrated results. What follows is the architecture intended to achieve them, together with the reasons it may fall short of the round figures.

The Ethereum Virtual Machine executes transactions one at a time, in a single queue. When activity spikes, a simple transfer waits behind a complex DeFi liquidation, which in turn waits behind everything else, and gas costs rise across the network because there is one congestion point that all participants share.

This single-threaded design is a safety property rather than an oversight. A node cannot execute two transactions in parallel without first knowing whether they touch the same storage, because if both write to the same slot, ordering matters and concurrent execution would corrupt state. The EVM has therefore discovered each transaction's data dependencies during execution, step by step, which enforces the sequential order.

EIP-7928, which introduces Block-Level Access Lists, removes that uncertainty. The block producer includes an upfront map of every storage slot and account that each transaction will access. With dependencies declared before execution rather than discovered during it, the structural justification for the single lane no longer holds.

Three consequences follow. Validating nodes can identify transactions that touch entirely separate state and execute them concurrently across CPU cores without risk of a write conflict, because the map demonstrates that they do not overlap. Nodes can preload the required state from disk into memory in a single batch rather than incurring disk-read latency mid-execution, which matters because disk input-output, rather than raw computation, is frequently the genuine bottleneck on a validating node. And non-proposer nodes can verify block state differentials rather than replaying every transaction in sequence, which reduces the burden of syncing.

That structural headroom is what justifies raising the block gas limit from 60 million toward 200 million, and the gas limit is the variable that maps most directly to throughput. More gas per block, processed in parallel, is the mechanism underlying the 10,000 TPS figure.

The qualification is important. Parallel execution delivers parallel speedup only when transactions genuinely fail to overlap, and real Ethereum blocks are not composed of random independent transfers. They are dense with contention, because a large share of transactions interact with the same handful of heavily used contracts—major DEX routers, the dominant stablecoins, popular lending pools. When transactions contend for the same state, Block-Level Access Lists allow nodes to detect the conflict cleanly, but they cannot cause conflicting writes to execute simultaneously. The 10,000 TPS ceiling is consequently a best-case figure predicated on favorable workloads. Realized throughput under typical DeFi-heavy load will sit below it, possibly considerably below, and the extent of that gap is precisely what the devnets exist to measure. The figure is best understood as a target the architecture permits rather than a baseline it guarantees.

Speed achieved at the expense of decentralization would not constitute a gain for Ethereum, and the centralization risk the network currently faces is concrete. Block production runs largely through MEV-Boost, an out-of-protocol arrangement, and by most counts between 80% and 90% of blocks are assembled by a small set of off-chain builders and routed to validators through third-party relays. That pipeline embeds trust assumptions, grants those intermediaries censorship leverage, and compresses the slot timeline, leaving validators little time to receive, verify, and propagate a block.

EIP-7732, which introduces Enshrined Proposer-Builder Separation, moves the auction into the consensus layer. The division between builders, who optimize transaction ordering for maximal extractable value, and proposers, the validators who commit blocks to the chain, becomes protocol-enforced rather than dependent on a relay's good behavior.

The change to the slot timeline carries the longest-reaching implications. ePBS divides a single slot into separate phases, providing the network a dedicated and extended window for payload verification and propagation. That additional window is what makes Ethereum's zero-knowledge future viable on the base layer, because it allows validators to receive and verify ZK proofs without risking penalties for missing a slot deadline, aligning the consensus layer with eventual ZK-EVM integration.

A measured assessment is warranted here as well. ePBS reduces the trust placed in relays and builders, but it does not abolish MEV or dismantle the builder market. Block-building remains specialized and capital-intensive, so a small number of sophisticated builders will likely continue to win the auction. What ePBS achieves is to make that competition trustless and to remove the relay as a censorship chokepoint. It does not, on its own, decentralize block production; it renders the centralization that remains less dangerous, which is a different and more defensible claim.

The third cluster of changes, EIP-7904, is the least conspicuous and likely the most immediately felt. Many gas costs were set years ago against hardware that no longer reflects how nodes actually operate, and EIP-7904 rebenchmarks them against real computational cost.

Resource-light operations, including ZK proving and cryptographic verification, become cheaper, which is what makes lightweight clients and privacy applications more viable on Layer 1. Operations that permanently expand Ethereum's global state become more expensive, deliberately, in order to discipline state growth toward a sustainable target on the order of 120 GiB per year. State bloat is the slow-moving centralization threat, because an ever-growing state raises the hardware floor for running a node, and a higher floor reduces the number of independent operators. Pricing state creation closer to its true long-term cost is the lever against that pressure.

The projected net effect is the approximately 78.6% reduction in baseline Layer 1 gas. The same caution applies as with the throughput figure: it is a projection tied to a particular workload mix and to the 200 million gas limit being safely reachable. A fee reduction of that magnitude assumes the gas-limit increase holds in production without degrading propagation or displacing smaller validators. Because the gas limit is ultimately a choice made by the validator set rather than mandated by the specification, a more conservative increase would yield a correspondingly smaller fee reduction.

The significant claim in Glamsterdam is not any single figure but the underlying bet: that the base layer's execution ceiling was a software limit sequential execution, mispriced gas, out-of-protocol block production rather than a hard physical one, and that the ceiling can be raised without increasing the cost of running a node enough to thin the validator set. Block-Level Access Lists unlock parallelism, ePBS hardens block production, and EIP-7904 disciplines state growth. Together they target higher throughput and lower fees while holding the line on who can afford to participate.

If the upgrade succeeds at the targeted scale, the tidy division of labor between a slow secure Layer 1 and fast Layer 2s becomes less distinct, and native DeFi on the base layer becomes economically plausible once again. If parallel speedup underdelivers against contended real-world workloads, or if validators raise gas limits cautiously, the headline figures will compress toward something more modest but still meaningful.

The concrete signals deserve more weight than the abstract ones. Public testnet activations on Sepolia and Hoodi are the clearest evidence that mainnet is approaching, and the benchmarks emerging from the current multi-client devnets will indicate how much of the 10,000 TPS target survives contact with realistic load. The round numbers belong to the marketing; the devnet data is the substance.

 

Disclaimer: This article is for informational purposes only and does not constitute financial, investment, or trading advice. Digital assets are volatile and you may lose capital. Conduct your own research before making any decision.