Ethereum’s L1 throughput debate, decided in three diagrams
Three charts settle a debate Ethereum researchers have been having since EIP-1559: gas-per-block, blob inclusion under EIP-4844, and MEV-Boost relay share tell the real story.
The argument about Ethereum L1 throughput is older than the network’s current consensus client. It runs from EIP-1559 in August 2021 through The Merge in September 2022 to the Dencun fork in March 2024 that introduced blob-carrying transactions through EIP-4844. Each fork was accompanied by a fresh round of research posts on ethresear.ch insisting that L1 capacity had to rise, fall, or stay flat. For most of that period the debate was decided by social-graph consensus — whoever wrote the most readable Mirror post that month set the narrative. As of the Pectra activation last May and the data accumulated since, the debate can now be settled empirically. Three diagrams do it. This piece walks through each.
What is at stake is whether Ethereum’s roadmap — a deliberately L2-centric one, codified in Vitalik Buterin’s writings on the rollup-centric future — is delivering on its premise. The premise was that L1 would settle, L2s would execute, and the aggregate throughput of the system would scale even as L1 gas limits stayed conservative. The premise is now testable. It is also, the data suggests, partly wrong: L1 is doing more work than the rollup-centric story implies, blobs are not consumed at the rate the EIP-4844 designers projected, and the MEV-Boost market has consolidated to a point that has implications for protocol-level decisions like inclusion lists. We will take the diagrams in order.
Diagram one: gas-per-block over time
Ethereum’s gas limit is a soft parameter, voted on by validators at the margin of each block. Before The Merge the limit hovered around 15 million gas; after EIP-1559 the target was set at 15 million with a 30 million ceiling. In November 2024 a validator coordination push lifted the target to 18 million; in February 2026 it lifted again to 22.5 million as part of Pectra’s social rollout. The actual gas consumed per block tracks the target almost perfectly because the 1559 base-fee mechanism makes any deviation expensive. The interesting figure is not the gas limit but the work done by that gas: how much of each block is L2 calldata, how much is rollup state-root submission, and how much is genuine L1 execution.
| Period | Gas target | L1 execution (avg) | L2 calldata/blobs (avg) | Other (avg) |
|---|---|---|---|---|
| Q1 2024 (pre-Dencun) | 15M | 62% | 34% | 4% |
| Q3 2024 (post-Dencun) | 15M | 71% | 13% | 16% |
| Q1 2025 | 18M | 69% | 15% | 16% |
| Q1 2026 (post-Pectra) | 22.5M | 65% | 22% | 13% |
The naive reading — that Dencun freed up L1 capacity for native execution because rollups moved their data to a parallel resource — is correct, but only partially. What the diagram shows is that the share of L1 execution dedicated to genuine L1 activity (DEX swaps, staking deposits, native NFT transfers) rose from 62% to 71% within a quarter of Dencun’s activation, then drifted back toward 65% as Pectra raised the gas target and rollups found new ways to use calldata for their own purposes. L1 is busier in absolute terms, doing more native work, and the rollup-centric story has not removed L1 demand — it has changed its composition.
Diagram two: the blob inclusion curve
EIP-4844 introduced a parallel resource — the blob — with its own market, its own base fee, and a target of three blobs per block with a ceiling of six. Pectra’s EIP-7691 raised the target to six and the ceiling to nine. The designers assumed that L2s would saturate the blob target almost immediately and the blob base fee would rise to a non-trivial equilibrium. That has not happened in the way they expected. For most of 2024 and 2025 the blob base fee sat at one wei — the protocol minimum — because rollup demand fluctuated wildly and aggregate consumption rarely met the target. Only since the Pectra raise did sustained equilibrium emerge.
The diagram that tells the story is a stacked area chart of blob consumption by rollup, sampled every epoch. Base accounts for roughly 38% of all blob consumption; Arbitrum for 24%; Optimism for 11%; Linea, Scroll, zkSync Era, Taiko, and Blast split the remainder. The shape is concentrated: two rollups consume more blob space than the next dozen combined. When Base launches a new partner integration or a token event, the blob market reacts within hours. When activity is quiet, the blob base fee falls back to one wei within two blocks. The market is bistable. The EIPs repository contains active discussion of a new pricing mechanism — EIP-7918 — that would smooth this behaviour by adding inertia to the base-fee update rule.
Diagram three: the MEV-Boost relay share
The third diagram is the most politically uncomfortable. MEV-Boost is an out-of-protocol auction system that allows validators to outsource block construction to a market of specialized builders, with a small set of trusted relays operating as the auction venue. The diagram is a pie chart of relay share over the last quarter. Flashbots’ relay accounts for roughly 41% of all blocks delivered through MEV-Boost; bloXroute’s two relays (regulated and max-profit) account for 28%; Titan Builder’s relay for 18%; the remainder is split among Agnostic, Aestus, Ultra Sound, and Manifold. That distribution understates the consolidation: the top three relays together deliver 87% of all auctioned blocks.
| Relay | Share of MEV-Boost blocks | Censorship profile | Notable builders |
|---|---|---|---|
| Flashbots | 41% | OFAC-compliant | Flashbots, beaverbuild, rsync |
| bloXroute Max-Profit | 17% | Non-filtering | Titan, beaverbuild |
| bloXroute Regulated | 11% | OFAC-compliant | Flashbots, Titan |
| Titan | 18% | Non-filtering | Titan Builder |
| Aestus / Agnostic / Ultra Sound | 13% | Mostly non-filtering | Various smaller builders |
Why these three together end the debate
Take the three diagrams together and the rollup-centric roadmap can be evaluated against its claims. The first chart shows L1 is doing more native work, not less, even with the gas-target ladder we have climbed. The second shows the blob market is structurally undersaturated, which means EIP-4844 succeeded at moving rollup data off L1 calldata but failed to produce the predicted blob-fee equilibrium. The third shows that the out-of-protocol auction has consolidated to a point that protocol-level interventions — EIP-7547 inclusion lists, the various FOCIL drafts, the long-discussed enshrined PBS — are necessary if Ethereum’s neutrality story is going to survive the next bull cycle.
What the three diagrams do not show, but which any honest interpretation must include, is the role of EIP-7702 in shifting the composition of L1 native execution. Account abstraction at L1 has changed how wallets bundle calls, which has raised the gas-per-transaction figure and lowered the transactions-per-block figure simultaneously. That bifurcation makes apples-to-apples comparisons across forks subtle. Vitalik addressed this in a January post on vitalik.eth.limo and the analytics community is still digesting the implications. Our gas tracker has been updated to show the breakdown by transaction-type tag.
The throughput plateau argument, restated
Critics of the rollup-centric roadmap have argued for at least two years that Ethereum is approaching a throughput plateau: that as long as L1 gas limits rise only modestly and the blob market remains undersaturated, the network’s aggregate transactions-per-second figure is bounded by L2 sequencer capacity and bridge security assumptions rather than by L1 itself. The three diagrams in this piece neither prove nor disprove that argument. They reframe it. The plateau, if there is one, is not at the L1 layer but at the bridging-and-trust layer. L1 has room to grow gas; blobs have room to grow consumption; what has not happened is a coordinated change to the trust model that lets rollups treat each other — and the L1 settlement layer — as a shared environment rather than a federation of isolated sovereigns.
- Native rollup proposals (the “based rollup” thread on ethresear.ch) would change the picture by making sequencing an L1-shared resource.
- Shared sequencers like Espresso and Astria sit between the two models; their share of L2 traffic is still under 4% as of Q1 2026.
- Restaking via EigenLayer adds a new trust dimension that the three diagrams above do not measure but which materially affects the bridging-layer plateau.
What the relay-share chart implies for protocol design
The MEV-Boost chart deserves a closer reading because it is the diagram with the most direct protocol-design implications. The three-relay concentration is not a market failure in the usual sense; it is the natural equilibrium of a market in which trust is the scarcest resource and relay reputation has compounding returns. A validator deciding which relays to accept blocks from is making a security decision, not a pricing decision, and security decisions favour the incumbent. The result is a stable oligopoly that no individual validator has an incentive to disrupt. That is precisely the kind of structural problem that protocol designers historically solve by enshrining the function in-protocol, which is what the various enshrined PBS proposals are trying to do.
The shorter-term intervention — inclusion lists — addresses a narrower problem: validators that delegate block construction to a censoring relay should still be able to force the inclusion of specific transactions that the relay declined to include. EIP-7547 codifies that mechanism. The argument against it is that it adds complexity to the validator client without solving the underlying centralisation question; the argument for it is that it preserves Ethereum’s neutrality property at the transaction level even while the block-construction market consolidates. The all-core-devs debate on this point has been the loudest of any ACD call in 2026, and the resolution — if there is one — will be embedded in the Glamsterdam fork specification rather than in any single Vitalik post.
What comes next
The Glamsterdam fork — the working name for the consensus-layer upgrade tentatively scheduled for Q4 2026 — is where the three open questions will be addressed in protocol. The blob-pricing fix sits as a candidate EIP; an inclusion-list mechanism sits as a candidate EIP; a gas-target raise sits as a candidate social rollout. None of these are settled. The all-core-devs calls have been unusually loud about the trade-offs, and the consensus team’s coordination has been visible on github.com/ethereum/pm for anyone who wants to read the meeting notes. For the cycle’s macro context against L2 token flows, our market dashboard tracks the L1/L2 fee ratio weekly, and the next ACD call is on our events calendar.