The Layer 2 Scaling Landscape

Research question and scope

This article compares the major technical approaches Ethereum Layer 2 networks use to scale transaction throughput, specifically on the dimension of what each approach actually trades off to achieve that scaling, and separately, how close each approach's real-world deployments actually are to the trustlessness their design intends. It is a comparison of publicly documented designs, not a live ranking by usage, market share, or total value locked — those figures change continuously and are covered separately by Kaavon's coin research profiles, not duplicated here.

Background and context

Ethereum's own roadmap has been explicitly "rollup-centric" since 2020: rather than scale the base layer's own transaction capacity indefinitely, the strategy is to push execution onto Layer 2 networks while the base layer focuses on providing security and data availability for them. The March 2024 Dencun upgrade (EIP-4844, "blobs") was a direct product of this strategy -- it added a dedicated, cheaper data-storage lane for rollups to post their transaction data to Ethereum, which materially reduced the cost rollups pay to inherit base-layer security. By late 2026, dozens of Layer 2 networks are active, built on a handful of shared technology stacks (notably the OP Stack and the ZK Stack), rather than each being built from scratch.

Methodology

L2Beat's project-level type, stage, and stack classifications in this article were directly read from l2beat.com/scaling/summary on November 4, 2026. L2Beat did not display total-value-secured (TVS) figures on that summary page in this pass, so this article does not cite an L2Beat TVS number; DefiLlama's chains dashboard, which would otherwise provide a comparable TVL cross-check, returned an HTTP 403 error on direct fetch and could not be read either. Market-share and TVL figures cited below therefore come from secondary market-commentary sources (cited individually), several of which disagree with each other by double-digit percentage points depending on measurement date and methodology -- this is disclosed explicitly rather than presented as a single precise figure.

Key findings

• There is no single "Layer 2 security model" — optimistic rollups, zero-knowledge rollups, and sidechains inherit the base layer's security to meaningfully different degrees.
• The fraud-proof window that gives optimistic rollups their name is the direct cause of their longer base-layer withdrawal delay — it is a deliberate design trade-off, not an oversight.
• Zero-knowledge rollups' faster cryptographic finality has historically come with more complex, more slowly-evolving tooling for general-purpose smart contracts, though this gap has been narrowing.
• Sidechains are the clearest trade-off case: higher throughput and lower cost in exchange for security that depends on the sidechain's own validator set, not the base layer's.

Why Layer 2 networks exist

Base-layer blockchains like Ethereum have limited transaction throughput, which can make fees rise significantly during periods of high demand. Layer 2 networks process transactions off the base layer while relying on it for underlying security, aiming to increase throughput and reduce cost without abandoning the base layer's security guarantees.

Optimistic rollups

Optimistic rollups (such as Arbitrum and Optimism) assume transaction batches are valid by default, with a challenge window during which a fraud proof can dispute an incorrect batch before it's treated as final. This generally offers strong compatibility with existing Ethereum applications, at the cost of a longer withdrawal delay back to the base layer.

Zero-knowledge rollups

Zero-knowledge (ZK) rollups instead generate a cryptographic proof that a batch of transactions was processed correctly, verified on the base layer, which can allow faster finality than the optimistic model's challenge-window approach, though historically with more complex, evolving tooling for general-purpose smart contracts.

Sidechains

• A sidechain runs its own independent consensus mechanism, rather than deriving security directly from the base layer's validators.
• This can allow higher throughput and lower cost, but means security depends on the sidechain's own validator set rather than inheriting the base layer's security directly.

L2Beat's Stage framework: how trustless a rollup actually is

A rollup's category (optimistic, ZK, sidechain) describes its intended design, not how much of that design is actually live in production. L2Beat, an independent research outfit that specializes in rollup risk analysis, classifies each network into Stage 0, Stage 1, or Stage 2 based on how much of its security actually runs without a trusted operator being able to override it -- Stage 0 generally means a trusted party (often the project's own team) can still unilaterally control the chain's state or upgrade its contracts with no delay; Stage 1 requires a functioning fraud/validity proof system plus a real user exit path if the operator misbehaves; Stage 2 requires removing privileged-party control almost entirely.

Based on a direct read of L2Beat's project summary page on November 4, 2026: Arbitrum One, Base Chain, OP Mainnet, Unichain, Ink, and Starknet were each classified Stage 1. Several other large, well-known rollups -- including Linea, ZKsync Era, Scroll, Mantle, and Katana -- were classified Stage 0. Stage 2 (the framework's most decentralized tier) appeared on L2Beat's list only for smaller or archived projects at the time of this reading, not on any of the largest rollups by usage. A number of large networks that market themselves loosely as "Layer 2s" -- including Polygon PoS, Gnosis Chain, and several OP Stack chains such as Ronin, Soneium, and Blast -- were categorized by L2Beat as "Other" rather than as a rollup type at all.

Comparing the three approaches

Independent critical analysis

The marketing framing around rollups -- that they "inherit" base-layer security -- is technically accurate about design intent but can overstate how much of that inheritance is actually live today. L2Beat's own Stage data, read directly for this article, shows that even several of the largest rollups by usage (Linea, ZKsync Era, Scroll, Mantle) remain Stage 0 as of November 2026 -- meaning a trusted operator retains the practical ability to override the chain's state. This is not a defect specific to any one project; it reflects a genuine, industry-wide engineering reality that trust-minimized proof systems and permissionless exit mechanisms are harder to ship than the base rollup architecture itself, years after the rollup-centric roadmap was first articulated. The practical implication: a reader evaluating a specific rollup's security should check its current Stage classification directly rather than assume "it's a rollup" implies a settled level of trust-minimization -- the category name describes an architecture, not a completed migration away from operator trust.

A second observation: the wide disagreement between secondary sources on basic figures like total L2 TVL (estimates in this research pass ranged from roughly $35 billion to $48 billion depending on source and date, within weeks of each other) is itself informative about the state of data infrastructure in this space -- it suggests a reader should treat any single cited TVL or market-share figure for Layer 2 networks as a rough, dated snapshot rather than a precise, cross-source-verified figure, independent of which specific number is cited.

Limitations of this comparison

• This table describes documented design intent, not measured real-world performance, cost, or decentralization of any specific network at any specific point in time.
• "Example networks" are illustrative of the category, not an exhaustive or ranked list, and a specific network's implementation can differ in detail from the general category description.
• Rollup and sidechain designs continue to evolve; a comparison like this should be periodically reviewed against each network's current documentation rather than treated as permanently fixed.
• L2Beat's own TVS dashboard and DefiLlama's chains dashboard were not independently readable in this research pass (no TVS figures shown on the summary page read; DefiLlama blocked the fetch with an HTTP 403 response) -- no single verified TVL/TVS figure is cited as a result, only the range reported across secondary sources.
• Stage classifications change as projects ship further decentralization work; the Stage 0/1 figures above are a November 4, 2026 snapshot of L2Beat's own classification, not a permanent rating.

Implications and conclusion

The Layer 2 landscape includes several genuinely different technical approaches, each trading off security model, finality time, and compatibility differently -- and, separately from the architecture itself, each network's actual deployment sits somewhere on a spectrum from "trusted operator still in control" (Stage 0) to "largely trust-minimized" (Stage 1/2), which the category label alone does not reveal. Evaluating a specific network on its own documented design and its current Stage classification, rather than assuming uniformity across everything labeled "Layer 2," produces a materially more accurate picture of how much security a given network's users are actually relying on the base layer for today, versus trusting the network's own operators.

References

• Primary, directly read: L2Beat, Scaling Summary (l2beat.com/scaling/summary, read November 4, 2026) -- project type, Stage classification, and technology stack for each listed network.
• Secondary, via search synthesis (not directly fetched; figures disagree across sources and are cited as a range for this reason): Yellow.com, "Ethereum L2s Are Splitting Into Winners And Dead Weight In 2026" (yellow.com/research/ethereum-l2-winners-dead-weight-2026); BlockEden, "Layer 2 Consolidation War: How Base and Arbitrum Captured 77% of Ethereum's Future" (blockeden.xyz/blog/2026/02/11/layer-2-consolidation-war-base-arbitrum/); SpotedCrypto, 2026 L2 market-overview articles (spotedcrypto.com) -- combined L2 TVL estimates ranging roughly $35-48 billion across these sources as of their respective 2026 publication dates.
• Attempted but not retrievable in this pass: DefiLlama chains dashboard (defillama.com/chains) -- fetch blocked with HTTP 403.