A Rollup-Centric Ethereum Roadmap
Executive summary
Buterin's six-page forum post in October 2020 is the most consequential strategic document Ethereum has ever produced: it announces the abandonment of the original sharding-first scaling plan in favour of a rollup-centric roadmap where the L1 scales data availability and execution moves to L2s. The post is technically thin — it is a strategic memo, not a specification — but it correctly forecasted the next five years of Ethereum's development: the rise of optimistic and ZK rollups, EIP-4844 (proto-danksharding), the deprecation of execution sharding, and the ecosystem's bifurcation into a settlement-and-data L1 with execution L2s. DI's assessment: the post is unusually short and unusually right. It belongs in a small genre of strategic documents that materially redirect a major protocol, alongside Satoshi's whitepaper and Ethereum's original yellow paper. Its limitations are mostly limitations of what it chose not to argue — the post does not analyse L2 fragmentation, sequencer centralisation, or cross-rollup composability, all of which became central problems by 2024.
Background and motivation
By October 2020 Ethereum had been working on its scaling roadmap — the multi-phase 'Ethereum 2.0' programme — for approximately three years. The original plan, announced in 2018 and refined through 2019-2020, envisioned a sharded execution layer with 64 (or in some versions 1024) execution shards each running their own EVM-equivalent state transition, plus a separate beacon chain for consensus. Phase 0 (beacon chain) was scheduled for late 2020. Phase 1 (data availability shards) and Phase 2 (execution shards) were forecast for 2022 and 2023 respectively. By mid-2020, however, several developments had complicated this plan. First, optimistic-rollup designs (Optimism's OVM, Arbitrum's AnyTrust/Nitro precursors) had matured to near-deployment, demonstrating that L2 execution scaling was viable on existing Ethereum L1. Second, ZK-rollup research had advanced considerably, with StarkWare's STARK proving and Matter Labs's zkSync producing operational testnets. Third, the cross-shard communication problem within Ethereum 2.0's execution-sharding model remained unsolved: tightly-coupled state across 64+ shards required either complex cross-shard transactions or aggressive synchrony assumptions, neither of which had a clean design. Buterin's post is best read as a public reckoning with these constraints. Rather than continuing to pursue execution sharding while rollups solved the same problem more pragmatically, the post argues that Ethereum should abandon execution sharding and reposition the L1 as a data-availability and settlement layer for rollups. The intellectual move is recognisably economic: it accepts that the optimal division of labour is between a security-and-data-rich L1 and execution-rich L2s, rather than a single layer that does everything. This framing was new to Ethereum's public planning even though it had been emerging in informal community discussions through 2020.
Core technical contributions
The post's technical contributions are sparse by paper standards but consequential by roadmap standards. First, it explicitly endorses rollups — both optimistic and ZK — as the primary scaling primitive, with rollup data published to L1 and rollup execution offloaded to L2. Second, it argues that L1 throughput should be measured in data bytes per slot rather than transactions per second, because rollup-using transactions consume L1 data but not L1 execution. Third, it forecasts that L1 data-shard scaling can support rollups processing approximately 100k TPS in aggregate, an order of magnitude beyond what execution sharding could plausibly deliver. Fourth, it implicitly deprecates execution sharding by reframing the L1's role as security and data, with applications living entirely on L2. Fifth, it identifies data availability as the binding constraint on L1 scaling and points toward what would later become danksharding: a design where validators sample data shards probabilistically rather than each validator needing to download the full data layer. The post does not specify these mechanisms in any detail — it is six pages, and most of those pages are strategic argument rather than technical specification. The actual mechanism specifications were developed in subsequent Ethereum Research posts and EIP processes through 2021-2024 (most importantly EIP-4844 by Buterin et al, 2022, and the danksharding research programme led by Dankrad Feist). The post's role is to commit to the strategic direction; the specifications are downstream artefacts of that commitment.
Methodology and rigor
By academic standards, the post has minimal methodology: there is no formal model, no proof, no empirical analysis, and no quantitative comparison of alternatives. It is a strategic memo, not a research paper. What rigor it has is the rigor of strategic argument: the post identifies the relevant tradeoffs (L1 execution vs L2 execution, data availability vs execution capacity, rollup latency vs L1 finality) and argues which side of each tradeoff Ethereum should take. The argument is correct but not deeply justified: Buterin asserts that 100k TPS is achievable with rollups plus data sharding without showing the calculation; he asserts that cross-shard communication is too hard without surveying the open research; he asserts that rollups will mature on their existing trajectory without analysing the failure modes. Subsequent papers have filled these gaps, but the original post itself is not the source of the rigorous case. From a methodological standpoint, the post's most important contribution is the framing rather than the analysis. It introduces the concept of L1-as-settlement-and-data, L2-as-execution as the governing mental model for Ethereum scaling, and that mental model has structured every subsequent design decision. The post does not prove that this framing is optimal; it asserts that it is the correct direction and lets the subsequent five years of research validate or reject the assertion. By 2026, the assertion has been broadly validated, although the validation is still ongoing — full danksharding has not yet shipped (as of April 2026), and the L2 ecosystem has produced fragmentation problems the post did not anticipate.
Strengths
The post's strengths are concentrated in strategic framing. First, it correctly identified that Ethereum's competitive position would be determined by the L2 ecosystem rather than L1 execution capacity, at a moment when most of the protocol-research community was still pursuing execution sharding. This was a non-obvious call at the time and has been validated by subsequent events. Second, it gave the L2 ecosystem permission to scale aggressively: by publicly committing the L1 to a supporting role, the post legitimised the substantial L2 investment that followed (Optimism, Arbitrum, zkSync, StarkNet, Polygon zkEVM, Base, and dozens of smaller rollups). Without this permission, L2 development would likely have been slower and more fragmented. Third, the post is decisive: it does not hedge or propose alternatives. It commits to a direction. This is unusually direct for protocol governance and is part of why the post had the influence it did. Fourth, the post is correctly humble about uncertainty: it acknowledges that data sharding mechanism design is incomplete, that rollup designs are still maturing, and that the roadmap may need to be revised as research progresses. This calibrated certainty (committed direction, acknowledged uncertainty) is characteristic of Buterin's best technical writing and is part of why the post has aged well. Fifth, the post is short. Six pages is the right length for a strategic memo; longer would have dissipated focus. The brevity forces the argument to be clear, and the clarity is a substantial part of the post's persuasive effect.
Weaknesses and limitations
The post has three important limitations. First, it does not analyse L2 fragmentation or cross-rollup composability. By committing to a multi-rollup ecosystem, the post implicitly accepts that liquidity and state will partition across rollups, but the post does not address how that partition will be managed. Subsequent years have made it clear that rollup fragmentation is a major UX and economic problem: bridging between rollups is slow, expensive, and trust-laden; liquidity is bifurcated; cross-rollup atomic transactions are not yet routine; and the user experience of using multiple rollups is materially worse than using a single chain. The post does not address any of this. Second, the post does not analyse sequencer centralisation. Optimistic rollups in 2020 used centralised sequencers, and the post does not require or discuss decentralisation as part of the L2 ecosystem's design. By 2024 sequencer centralisation had become a major critique of the rollup ecosystem and a focal point for shared-sequencer research (Espresso, Astria), which is downstream of the gap left by the original post. Third, the post does not consider the alt-L1 competitive landscape. By concentrating Ethereum's strategy on rollups, the post implicitly cedes the integrated-L1 segment to Solana, Sui, Aptos, and others; whether this was the right strategic choice depends on whether the Ethereum-plus-rollups stack can match an integrated L1's UX, which is an open question even in 2026. Fourth, the post says nothing about the economic relationship between L1 and L2: how rollup fee revenue should accrue to ETH stakers, whether L2 sovereignty is acceptable, and how MEV would distribute across the stack. These are now major unresolved governance questions.
Subsequent influence
The post's influence on Ethereum's actual development trajectory is hard to overstate. EIP-4844 (proto-danksharding, shipped March 2024) is a direct descendant: it introduces blob-carrying transactions specifically designed for rollup data, exactly as the post forecast. The full danksharding research programme is the realisation of the post's data-sharding direction. The deprecation of execution sharding — never explicitly cancelled but quietly removed from the roadmap — happened because the post made it possible to do so without political cost. The L2 ecosystem itself can be read as a downstream consequence: the post legitimised L2 investment, and the resulting capital flows produced Optimism, Arbitrum, Base, zkSync, StarkNet, Polygon zkEVM, Linea, Scroll, and dozens of smaller rollups. By April 2026, the L2 ecosystem hosts an aggregate TVL exceeding $80B and processes more transaction volume than Ethereum L1. Outside Ethereum, the post influenced the broader L1 strategic conversation. Cosmos's interpretation of L2-as-appchain is recognisably different from Ethereum's L2-as-rollup framing but is shaped by the same observation that monolithic execution layers do not scale. Polkadot's parachain model predates the post but converged on a similar division of labour. Avalanche's subnet strategy explicitly competes with Ethereum's rollup strategy on similar architectural ground. Academically, the post has been cited in dozens of subsequent rollup-design papers, danksharding mechanism designs, and rollup-economic-analysis papers. It is the canonical reference for the rollup-centric framing even though it provides almost no technical detail.
How it holds up in 2026
Five and a half years post-publication, the post's strategic thesis has been broadly vindicated. EIP-4844 has shipped. The L2 ecosystem is operational at scale. Rollup throughput substantially exceeds what execution sharding could have delivered. Ethereum L1 has retained its security-and-settlement role and has not been displaced by alt-L1s in the high-value asset categories (BTC-equivalent stores of value, large institutional flows, RWA). The post's forecast of 100k TPS via rollups remains aspirational rather than realised, but cumulative L2 throughput in April 2026 is in the 4-8k TPS range and growing. The strategic gamble — committing Ethereum to a rollup future and accepting the L2 fragmentation cost — has produced a competitive ecosystem rather than an integrated-L1 collapse. On the other side: the L2 fragmentation problem is real and growing. By April 2026 there are 60+ active L2s with material TVL, cross-rollup UX is poor, and the Ethereum-plus-rollups stack still has materially worse user experience than Solana for most retail use cases. Sequencer centralisation remains unresolved despite years of shared-sequencer research. Full danksharding is not yet live (PeerDAS is the latest milestone). The MEV-and-fee-distribution question across L1 and L2 is unsettled, with growing tension between Ethereum core and L2 incentives. The post's silences — on fragmentation, sequencer decentralisation, cross-rollup composability — have become Ethereum's open problems. DI's holistic assessment: the post is one of the small handful of strategic memos in computing history that materially redirected a major protocol and was vindicated by subsequent events. Its limitations are limitations of what a six-page forum post can cover, not errors in what it argued. As of 2026, the rollup-centric roadmap is the consensus view of Ethereum's path, and that consensus traces back to this post. A nine on enduring importance, and arguably the most influential blog post in cryptocurrency history.
Strengths
- Correctly identified data availability rather than execution capacity as L1's binding scaling constraint
- Decisive strategic commitment legitimised L2 investment and accelerated rollup ecosystem development
- Calibrated certainty: committed direction with acknowledged uncertainty about specific mechanisms
- Implicit deprecation of execution sharding solved a major roadmap dead-end without political cost
- Six-page brevity forced argumentative clarity and increased the post's persuasive effect
- Forecast of 100k TPS via rollups was directionally correct even if not yet fully realised
Weaknesses
- L2 fragmentation problem unaddressed; cross-rollup UX and liquidity remain unresolved in 2026
- Sequencer centralisation not flagged as a design constraint for the L2 ecosystem
- Economic relationship between L1 and L2 (fee accrual, MEV distribution, sovereignty) unanalysed
- Alt-L1 competitive landscape not engaged; integrated-L1 segment ceded to Solana et al
- Technical specifications absent; relied on subsequent EIPs (4844, danksharding) to provide rigor
- 100k TPS forecast remains aspirational; full danksharding not yet shipped as of April 2026
Key contributions
- Strategic pivot from execution sharding to data-availability sharding as Ethereum's primary scaling axis
- Explicit endorsement of rollups (both optimistic and ZK) as the medium-term scaling solution
- Forecast of L1 throughput at ~100k TPS via rollups operating on top of data-shard L1
- Identification of data availability as the binding L1 scaling constraint, not execution capacity
- Argument that L1 should optimise for security and decentralisation, not application throughput
- Implicit deprecation of phase-2 execution sharding (later confirmed and executed)
- Framing of L2 ecosystem as a competitive market rather than a planned partition
TL;DR
Six-page forum post that redirected Ethereum's entire scaling strategy in 2020. Vindicated by EIP-4844, the L2 ecosystem, and the deprecation of execution sharding. Its silences (L2 fragmentation, sequencer centralisation) are Ethereum's current open problems.
Get DeFi Intel research in your inbox
Weekly long-form coverage of papers, incidents, jurisdictions, chains, tokens and the people building them. Free tier covers headlines; Pro adds the analyst-grade breakdowns.
Sources
External references gathered from the body of this brief. Last reviewed 2026-05-03.