The House Edge

Appearance

10. Ethereum and Its L2s: The Most Studied, Least Settled ​

The chain where the extraction surface is most explicitly architected — and where the value capture dispersed along architectural lines but concentrated along operator ones. ​

Cold open ​

In Q1 2026, Ethereum's chain-level revenue fell to #3 behind Hyperliquid (~$144.8M) and Solana (~$89.5M) at approximately $82 million — the first quarter in the chain's history in which both a single-application L1 and a general-purpose L1 outearned it.[1] On the same chain in the same quarter, Titan built approximately 52% of all mainnet blocks, BuilderNet built approximately 25%, Quasar approximately 15% — three firms responsible for over 91% of block construction (Chapter 5).[2] On the L2 layer above, Base captured approximately 62% of all L2 fee revenue — roughly $67 million annualised through its single-sequencer architecture, of which approximately 65% came from just 250 addresses in one year.[3] On the restaking layer parallel to both, EigenLayer's total value locked sat at approximately $18 billion across 4.6 million ETH delegated to ~1,900 operators, with slashing live since April 2025 and an active-validated-services roster that included EigenDA, AltLayer MACH, Hyperlane, and Witness Chain.[4] The compression at Ethereum L1 is real and is the chain's clearest economic story: blob fee revenue from L2s collapsed from approximately $113 million in 2024 to approximately $10 million in 2025 — a more than 90% decline.[5] The L2 architecture worked. The L2 architecture moved the fee economy off L1 and onto Base, Arbitrum, Optimism, and the long tail of seventy-some rollups. But the same handful of firms — Coinbase, Consensys, Flashbots, Lido — appear at multiple layers of the resulting stack. Chapter 8 was Solana (vertical integration that compressed the extraction surface). Chapter 9 was Hyperliquid (architecture that eliminated it). Chapter 10 is Ethereum + L2s: the chain where the extraction surface is most explicitly architected, most extensively measured, and most concentrated by named operators even though the architecture deliberately disperses it.

What this chapter answers ​

  • What happened to Ethereum's L1 fee economy when EIP-4844 and Pectra moved most transaction volume to L2s?
  • What does the PBS extraction surface look like in 2026 with ePBS still ~12 months out — and who owns it?
  • Why did Base capture 62% of all L2 fee revenue, and what does that tell us about the L2 sequencer economy?
  • What is cross-chain MEV — and why is the L2 fragmentation problem producing a new extraction surface measured at ~$10M+/year?
  • How does the PBS + L2 fragmentation + restaking trifecta produce a value-capture pattern different from Solana's and Hyperliquid's — and why are the same five-to-ten firms operating across all three surfaces?

The setup ​

Ethereum's architectural premise in 2026, stated cleanly: the chain has deliberately dispersed its value capture across three architectural surfaces — PBS at L1, L2 sequencers and bridges, and restaking via EigenLayer — under the design assumption that explicit separation of concerns produces a more competitive operator landscape than vertical integration (Solana) or in-consensus matching (Hyperliquid). The empirical question this chapter develops: did the dispersion produce operator competition, or did the same set of well-capitalised firms simply colonise each architectural layer in turn? The answer the chapter assembles, drawing on Chapters 5 (PBS architecture), 6 (the infrastructure layer), and 7 (exclusive flow): the dispersion is real along architectural lines — the L1 builder is not the L2 sequencer is not the restaking AVS operator — but the dispersion is not real along operator lines. Coinbase Cloud is a top Ethereum staker (~5–12% of all staked ETH, depending on metric), the Base sequencer operator, an OFAC-filtering-optional validator (Chapter 5), and an EigenLayer restaking operator. Consensys operates the Linea sequencer, MetaMask, Infura, and — post-January 2026 transfer per Chapter 6 — MEV-Blocker. Flashbots co-operates BuilderNet, runs MEV-Boost / Flashbots Protect / Flashbots Relay, and operates SUAVE/Flashnet R&D. The same five-to-ten firms operate at multiple architectural layers; the architecture's explicit separation of concerns has not produced a corresponding separation of operators.

The chapter previews three structural findings the worked example and the mechanics develop. First: the L1 fee economy contracted by design. ETH is now mildly inflationary (~0.23% annually), L1 daily fees fell from approximately $23 million to approximately $6.3 million, and blob payments from L2s to L1 collapsed from approximately $113 million in 2024 to approximately $10 million in 2025.[6] Second: the value did not disappear when it moved off L1. It now flows through Base's sequencer (~$67 million per year, ~$870 million cumulative since Base's August 2023 launch), Arbitrum's Timeboost (~$3 million per year, with two firms winning more than 90% of express-lane auctions), and the L2 fee waterfalls of Optimism, Linea, Scroll, and others. Third: a new extraction surface emerged. Cross-chain MEV was measured at approximately $10 million of searcher revenue across nine chains in twelve months by Maire et al., with the top five addresses executing more than 50% of all cross-chain arbitrages and a single address capturing approximately 40% of daily post-Dencun volume.[7]

The worked example ​

Alice — the trader the book has followed since the prologue — makes her $10,000 USDC → ETH swap through the Ethereum 2026 stack twice: once on Ethereum L1, and once on Base. The chapter's central illustration of the trifecta is the two-leg comparison.

The published and partially-illustrative take at each layer for Alice's $10,000 swap on each path:

Layer (L1)MechanismTake (illustrative)Cumulative
WalletMetaMask + Flashbots Protect (free protected routing)~$0~$0
IntentCoW Swap intent submission~$0 (CoW is intent-based)~$0
SolverSCP or Barter wins the solver auction; ~80–90% of CoW solver share between two firms~5–10 bps spread captured by solver and shared with user via better execution~$5–10
BuilderTitan ~52% builder share; Wu et al. average retention from exclusive flow ~17.75% margin under Banana-Gun-style contracts~$5–10 captured at the builder layer~$15
RelayUltrasound ~36% / Titan Relay ~26% / others~$0–1 relay fee (most relays run at near-zero margin)~$15
ProposerLido validator (~24% of stake) receives MEV-Boost payment; ~10% commission to Lido + ~5% to node operator + remainder to staker~$3–5 to staker; minimal to proposer at the per-trade level~$15–20
L1 gasEIP-1559 base fee + priority fee; post-Pectra L1 fees ~$6.3M/day system-wide~$5–10 depending on congestion~$20–25
L1 total~$15–25 cumulative take
Layer (Base)MechanismTake (illustrative)Cumulative
WalletCoinbase Wallet (Base-native)~$0~$0
DEXAerodrome swap (Base's largest DEX)~5–10 bps DEX fee; LP-paid~$2–4
SequencerBase sequencer (Coinbase); priority-fee surcharge plus base sequencer fee<$1 sequencer fee at typical L2 gas levels; ~$185K/day aggregate across all Base trading~$2–5
Blob settlementBase's L2-batch settles to L1 via blob fee<$0.10 per trade at Fusaka blob pricing~$2–5
Base total~$2–5 cumulative take

Two acknowledgements. First: several per-layer figures are illustrative rather than published — the Lido staker-share economics depend on the specific node operator's commission and Lido's ongoing fee mechanics; the SCP/Barter solver retention is composed from Chapter 7's solver-layer analysis. Second: the L1-vs-Base 6–8× difference in cumulative take is real and is the L2 architecture's clearest improvement to retail welfare — but the fee flow concentrates into a single sequencer operator (Coinbase) on the Base path, where the L1 path dissipates across multiple competing builders and relays. Whether higher visibility plus lower magnitude on Base is structurally preferable to lower visibility plus higher magnitude on L1 is the chapter's clinical comparison rather than a moral judgement.

The cross-chain searcher pattern Maire et al. documented is the worked example's third reference point: a single address that executes ~40% of daily post-Dencun cross-chain arbitrage volume captures a meaningful fraction of the ~$10 million annualised cross-chain MEV pool — a concentration ratio comparable to the L1 builder oligopoly Chapter 7 documented, one architectural level out. The chapter develops the cross-chain MEV mechanics in §4d.

The mechanics, in detail ​

L1 in 2026 — the chain that moved its fee economy ​

Ethereum L1's economic compression since EIP-4844 (Dencun, 13 March 2024) and Pectra (7 May 2025, EIP-7251 raising max effective balance to 2,048 ETH) is the chapter's first structural fact. Validator count in May 2026: approximately 1.06–1.1 million, compressing as 32-ETH validators consolidate into up-to-2,048-ETH consolidated validators under Pectra. Total staked ETH: approximately 35.86 million (~28.91% of supply). Daily L1 gas revenue: approximately $6.3 million in 2025, down from a peak of approximately $23 million. ETH is now approximately 0.23% annually inflationary for the first sustained period since the Merge; the daily ETH burn fell to approximately 3.26 ETH per day post-Pectra (a roughly 71% decrease). Ethereum's Q1 2026 chain-level fee revenue: approximately $82 million, placing the chain at #3 by chain-level REV behind Hyperliquid (~$144.8M) and Solana (~$89.5M) for the first time in its history.[8]

The chapter develops this not as decline narrative but as deliberate trade-off: the L2 architecture worked, the fee economy moved to L2s, and Ethereum has not (yet) recovered the value capture. Blob payments from L2s to L1: approximately $113 million in 2024 → approximately $10 million in 2025 — a more than 90% decline.[9] Industry estimates suggest blob fees could contribute 30–50% of total ETH burn by 2026 depending on L2 activity. The Fusaka upgrade (3 December 2025) introduced PeerDAS (EIP-7594) and EIP-7918, reducing per-node blob bandwidth by approximately 8×; two Blob Parameter Only (BPO) forks pushed the per-block blob target from 3 to approximately 14 and the maximum from 6 to 21 across December 2025 to January 2026. The long-term Danksharding goal is approximately 128 blobs per block. The L1 fee economy is not coming back to its 2021–2023 levels; the chain has architected itself into the settlement-and-data-availability layer for a competitive L2 economy that runs above it.

PBS in 2026 — Titan, BuilderNet, Quasar, and the new entrant ​

Live MEV-Boost telemetry for early-to-mid 2026 (relayscan.io 24-hour snapshots, refreshed at draft date): builder shares Titan ~52.16%, BuilderNet ~24.63%, Quasar ~15.06%, Eureka ~2.41% (the new entrant), Beaverbuild ~1.81%, Bombora ~1.31% — top three approximately 92%.[10] Relay shares: Ultrasound ~35.71%, Titan Relay ~26.09%, BloXroute Max-Profit ~13.38%, BloXroute Regulated ~12.30%, Aestus ~7.32%, Flashbots ~2.09%. The PBS oligopoly Chapter 5 developed in May 2025 has consolidated, not loosened, through Q1 2026.

Eureka Labs is the chapter's one structurally new builder entrant since Chapters 5 and 6: founded December 2024, raised a $6.7 million seed in Q2 2025 from Spark Capital and Collider Ventures, currently fourth-largest builder at approximately 1.5–2.4% share.[11] The firm's product positioning is "programmable blocks" — a builder service that exposes more configurable transaction-ordering rules to its searcher and bot counterparties than Titan's exclusive-flow model or BuilderNet's open-source non-exclusive design (Chapter 6). The structural question Eureka's entry poses: can a new builder gain meaningful share against the Titan-BuilderNet-Quasar oligopoly without an exclusive-flow contract of the Banana-Gun type? Eureka's slow share growth through 2025–2026 suggests the answer is approximately "yes, slowly" — meaningful but bounded entry is possible, but it does not appear to dent the incumbents.

The chapter's structural argument lands on the ePBS timing. EIP-7732, the in-protocol replacement for the out-of-protocol MEV-Boost relay system, has slipped past Glamsterdam's original H1 2026 target. Per the Ethereum Foundation's April 2026 Checkpoint #9 publication: "Glamsterdam in Q2 seems to me to be unlikely." The implementation has proved "trickier than anticipated" due to two-party consensus coordination and "partial block" handling.[12] Justin Drake's February 2026 "Strawmap" outlines a seven-fork roadmap through 2029 with explicit one-consensus-headliner + one-execution-headliner pacing per fork. Realistic mainnet activation for ePBS: Q3/Q4 2026, with Base team warning of potential slip to 2027 if FOCIL (Fork Choice Inclusion List, EIP-7805) is added alongside ePBS in the same hard fork.

The structural finding the subsection lands: MEV-Boost — the out-of-protocol PBS — has been the de facto enshrined PBS for approximately four years and counting; the in-protocol replacement is still 12+ months out. The exclusive-flow concentration Chapter 7 documented (Wu et al.'s 75 EOF arrangements accounting for approximately 71% of trading-related Ethereum builder revenue; Pahari and Canidio's 77.2%–84% of fees from exclusive transactions; Banana Gun + Titan and Maestro + Beaverbuild as the two named contracts) is the chapter's PBS-layer empirical anchor and is referenced rather than re-litigated here.

The L2 sequencer economy — Base wins, Arbitrum monetizes, the rest catch up ​

Total L2 TVL in May 2026: approximately $48 billion across approximately 73 active rollups.[13] Arbitrum + Base + Optimism collectively process approximately 90% of L2 transaction volume. TVL distribution: Arbitrum One ~$14.9–16.9B (40–44% of L2 TVL); Base ~$10.7–11.2B (28–33%); OP Mainnet ~$5.6B; zkSync Era ~$4.1B; Linea ~$3.4B; Scroll ~$2.1B; Starknet ~$1.5B; Unichain ~$0.88B. Daily transactions: Base ~12.89 million per day (~1 million+ daily active addresses, the highest of any L2); Arbitrum One ~4.3 million per day; Starknet ~585,000; Linea ~211,000. The L2 ecosystem as a whole now processes more transactions than Ethereum mainnet by a wide margin. L2BEAT Stage classification as of May 2026: Stage 1 (permissionless fraud proofs + trustless exit) achieved by Arbitrum One, Base, OP Mainnet, Starknet, Scroll, and Ink. No L2 has reached Stage 2 (fully decentralised governance) as of May 2026.

Every major L2 in production in May 2026 operates with a single centralised sequencer.[14] Arbitrum One: sequencer operated by Offchain Labs. Base: sequencer operated by Coinbase. OP Mainnet: sequencer operated by OP Labs. Unichain: sequencer operated by Uniswap Labs. Linea: sequencer operated by Consensys (the Maru/QBFT upgrade replaced the earlier fully centralised sequencer with multi-node consensus, but the operators remain Consensys-affiliated). The single-sequencer architecture concentrates transaction ordering, MEV extraction, and fee revenue at the operator — structurally analogous to a single block builder with no proposer-side competition.

Combined L2 sequencer revenue in 2025–2026 runs $150–250 million annually, with Base capturing approximately 62% of all L2 fee revenue.[15] Specifically: Base averages approximately $185,291 per day of sequencer revenue (~$67 million annualised); the full-year 2025 cumulative figure was approximately $870 million; one recent month posted $51.1 million; the past-30-day on-chain fee revenue ran approximately $2.72 million; Base was the only L2 profitable in 2025 at approximately $55 million net. Base's priority-fee component alone averages approximately $156,138 per day, accounting for approximately 86.1% of daily sequencer revenue — and 64.9% of all priority fees on Base came from just 250 addresses in one year, the most extreme retail-vs-power-user concentration ratio the book has documented. Arbitrum's gross DAO income runs approximately $26 million annualised as of Q4 2025; Timeboost generated approximately $3 million in its first three months and now contributes approximately $3 million per year run-rate (approximately 26% of Arbitrum DAO income).

Arbitrum Timeboost is the chapter's L2-side exclusive-flow anchor. Launched in April 2025, Timeboost is a sealed-bid second-price auction selling 200-millisecond "express lane" priority for a 60-second slot. Bidders compete to win the express lane; the auction's revenue routes 97% to the Arbitrum DAO and 3% to the auctioneer. The express-lane auction now accounts for 20–30% of daily DEX volume on Arbitrum. Castro et al. (arXiv 2509.22143, September 2025) document the empirical concentration: two entities win more than 90% of Timeboost auctions; competition has declined over time; revenue has trended downward as auction concentration tightened.[16] The structural pattern is the chapter's L2-side EOF echo: the same exclusivity-by-stake-and-relationship dynamic Chapter 7 documented at the L1 builder layer has re-emerged at the L2 sequencer layer one architectural level down.

Shared and decentralised sequencers are in production but not yet at material scale. Espresso Systems' Espresso Sequencer launched mainnet on 12 February 2026 with HotShot consensus and sub-second finality; first integrations with Caldera and Arbitrum Orbit chains; throughput target 25 MB/s (from current 5+ MB/s).[17] Astria's shared sequencer is in production for modular EVM rollups. Optimism's Superchain shared sequencer (in collaboration with Espresso + OP Labs) is targeted for 2026 production alongside native interop. As of May 2026 no major standalone L2 has decentralised its sequencer in production at scale; the realistic horizon is late 2026 to 2027. The single-sequencer architecture is, for the moment, the L2 economy's structural shape.

Cross-chain MEV — the new extraction surface the L2 fragmentation produced ​

The chapter's load-bearing empirical anchor for the cross-chain surface: Maire, Sviridov, Capponi, and Wattenhofer (Flashbots + TU Munich; ACM SIGMETRICS 2024 / arXiv 2501.17335) identified 242,535 cross-chain arbitrages across nine blockchains between September 2023 and August 2024, totaling approximately $868.64 million of trading volume and generating approximately $10.05 million in searcher revenue / $8.65 million net profit.[18] Activity grew approximately 5.5× over the study period and surged after Dencun (March 2024) cut rollup costs. Composition: 58.35% L1-to-L2 pairs (driven by native bridge availability), 35.67% L2-to-L2 pairs (relying on third-party bridges or pre-positioned inventory). Most trades — approximately 66.96% — use pre-positioned inventory and settle in approximately 9 seconds; bridge-based arbitrages take approximately 242 seconds (about 27× longer). Concentration: the top five addresses execute more than 50% of all trades; one address captures approximately 40% of daily post-Dencun volume.

This is the chapter's "L2 fragmentation creates a new extraction surface" empirical anchor and is structurally a cross-chain analogue of the L1 builder concentration Chapter 7 documented — same shape, one architectural level out. The five-firm concentration at the top of the cross-chain arbitrage table is comparable to the three-firm L1 builder oligopoly and to the two-firm Timeboost express-lane oligopoly.

Gogol et al. (arXiv 2406.02172, October 2024) quantify non-atomic MEV between Arbitrum, Base, Optimism, and zkSync Era: more than 500,000 unexplored arbitrage opportunities identified; arbitrage opportunities run approximately 0.03%–0.05% of trading volume on Arbitrum, Base, and Optimism; approximately 0.25% on zkSync Era; opportunities persist 10–20 blocks on average.[19] Atomicity is the structural difference between L1 MEV (atomic, single-block, deterministic) and cross-chain MEV (non-atomic, multi-second to multi-minute, probabilistic). The multi-chain world is not very atomic, which creates risk surfaces searchers must price into bids — and which produces capital-intensive pre-positioning patterns where the top-five searchers maintain large inventories on every major L2 to capture the cross-chain arbitrage opportunities as they emerge. Shared sequencers (Espresso, Astria) and intent-based architectures (UniswapX cross-chain, CoW Swap) attempt to re-introduce atomicity at the sequencer or solver layer; neither has replaced bridge-based execution at scale as of May 2026.

Restaking — EigenLayer as the third extraction surface ​

EigenLayer's restaking TVL sat at approximately $15.3–19 billion across approximately 4.36–4.6 million ETH at approximately 1,900 active operators in early-to-mid 2026; the chapter uses approximately $18 billion / 4.6 million ETH as the consolidated snapshot.[20] EigenLayer holds approximately 94% market share of the restaking ecosystem. All-time high TVL: approximately $19.7 billion (some sources cite up to ~$25 billion aggregate including Eigen Cloud, the firm's broader infrastructure offering). Slashing officially went live on 17 April 2025, making EigenLayer "complete" in the founders' published framing. ELIP-006 Redistributable Slashing (mainnet, 2026) introduces Redistributable Operator Sets — slashed funds either burned (the standard slashing path) or redistributed (the redistributable path) under operator-set-specific governance.

Named active validated services in May 2026: EigenDA (data availability, the largest single AVS by restaked stake); AltLayer MACH (fast finality for OP Mainnet and Arbitrum One, bundled with Caldera/Conduit/Gelato/AltLayer RaaS); Hyperlane (cross-chain messaging); Witness Chain (DePIN proof-of-location); plus a growing "Vertical AVS" category specializing in narrow verification problems (the published trend is toward AI-model-evaluation AVSs across the 280+ crypto-AI projects EigenLayer's published research tracks). The structural finding the subsection lands: restaking is the cross-chain extraction surface's third architectural layer. An Ethereum-staked validator now secures arbitrary AVSs, some of which operate cross-chain — Hyperlane is the named example; AltLayer MACH provides fast finality across Arbitrum + OP Mainnet from a single Ethereum-secured operator set. The same restaked ETH that earns base PoS rewards now also earns AVS rewards, and is exposed to slashing across multiple operator sets simultaneously. The economic compounding — one ETH worth of stake earning revenue from N AVSs while also serving as Ethereum L1 collateral — is the structural innovation; the operator-concentration risk it creates is the structural cost.

The chapter's argument arrives here: the L2 fee economy + the restaking surface + PBS exclusive flow form the trifecta that defines Ethereum's structural shape in 2026. The next section develops the trifecta as the chapter's verdict frame.

How this plays out on each chain — and what Ethereum's structure isn't ​

Ethereum is the chain whose architecture deliberately disperses value capture across L1 PBS + L2 sequencers + restaking, producing the trifecta the chapter has documented. The dispersion is real along architectural lines; it is not real along operator lines. Coinbase Cloud is a top Ethereum staker (~5–12% depending on the metric used), the Base sequencer operator, and an OFAC-filtering-optional validator (Chapter 5). Consensys operates the Linea sequencer, MetaMask, Infura, and (post-January 2026) MEV-Blocker. Flashbots co-operates BuilderNet, runs Flashbots Protect / MEV-Boost / Flashbots Relay, and operates SUAVE / Flashnet R&D. Lido is the named L1 staking concentration (approximately 24% of all staked ETH), the largest single restaking-via-EigenLayer participant, and the named entity in the only meaningful staking-protocol governance vote that has limited validator concentration at the protocol layer. The same five-to-ten firms operate at multiple architectural layers; the dispersion produces architectural legibility without operator competition.

Solana (Chapter 8) is the chain whose vertical integration compressed the extraction surface but did not eliminate it. Five firms (Phantom, Helius, Jupiter, Jito, the leader validator) touch one retail trade; the access-to-operational ratio is approximately one order of magnitude on per-block priority fees. The vertical integration is what produces the access gap Chapter 7 documented.

Hyperliquid (Chapter 9) is the chain whose architecture eliminated the extraction surface — no mempool, no PBS, no relay layer — but re-concentrated value capture into four named loci (HLP, validator set, HIP-3 deployers, HYPE holders via buyback). The names are different; the concentration is comparable.

The three-chain Part III pattern lands here. Architectural choices produce different value-capture shapes, but the operator concentration is remarkably similar across all three chains. Solana concentrates value capture vertically into the validator-builder-searcher stack. Hyperliquid concentrates it horizontally into four named loci of comparable order-of-magnitude scale. Ethereum disperses it along three architectural lines — PBS, L2 sequencers, restaking — but the same handful of firms appear at multiple layers, producing operator concentration that mirrors the other two chains even where the architectural dispersion does not.

Who wins, who loses, why ​

Four-layer stratification reflecting Ethereum's trifecta.

The L1 builder oligopoly wins. Titan ~52%, BuilderNet ~25%, Quasar ~15% — three firms in 91% of mainnet block construction. The Wu et al. 75 EOF arrangements accounting for approximately 71% of trading-related Ethereum builder revenue (Chapter 7) is the most concentrated extractive surface in the book. Eureka Labs' new ~2% share is the only meaningful entry since 2024, and its growth has been gradual rather than disruptive.

The L2 sequencer monopolies win, layer by layer. Base (Coinbase) captures approximately $67 million per year of fees as the only profitable L2; Arbitrum (Offchain Labs) monetizes via Timeboost at approximately $3 million per year, with two firms winning more than 90% of express-lane auctions; OP Mainnet (OP Labs) operates the Superchain alongside Coinbase's Base and Uniswap's Unichain; Linea (Consensys) runs sequencer revenue through Consensys's broader infrastructure stack. The single-sequencer architecture concentrates fee revenue and MEV at the operator with no proposer-side competition — structurally identical to a block builder without a relay layer between it and the proposer.

The cross-chain searchers win, and they are concentrated. Top five addresses execute more than 50% of all cross-chain arbitrages identified by Maire et al.; one address captures approximately 40% of daily post-Dencun volume; the cross-chain MEV surface is structurally a few-firm market with high pre-positioned-inventory capital requirements. The L2-fragmentation equivalent of the L1 builder-direct contract Chapter 7 documented operates with similar concentration shape — same few-firms-with-capital-moat pattern, one architectural level out.

Restaking operators capture a new revenue layer. EigenLayer's approximately 1,900 operators across approximately $18 billion of restaked ETH provide security to AVSs in exchange for AVS reward streams; the largest operators (the named roster includes Lido, P2P.org, RockX, Coinbase Cloud, and others) capture the lion's share of operator-level fees. The structural innovation is real — Ethereum-staked validators can now secure arbitrary cryptoeconomic surfaces — and the operator concentration mirrors the staking-operator concentration the chapter has documented elsewhere.

Retail traders on L1 lose to the exclusive-flow surface Chapter 7 documented; on L2s they pay less per trade (Base approximately 2–5 bps vs L1 approximately 15–25 bps for the same swap on the same asset), but the fee flow routes to a single sequencer operator rather than dissipating across competing builders. The L2 retail experience is structurally better than L1 retail in pure cost terms; whether higher visibility plus single-operator concentration is preferable to lower visibility plus multi-operator concentration is the chapter's clinical comparison rather than a moral judgement.

The structural closing observation. Ethereum's PBS + L2 fragmentation + restaking trifecta produces a value-capture pattern that is architecturally more legible than Solana's vertical integration or Hyperliquid's eliminated-surface design — every layer of extraction is named, measured, and (in many cases) auditable on-chain. The Wu et al. paper exists. The Pahari and Canidio paper exists. The Maire et al. paper exists. The Castro et al. paper exists. relayscan.io and L2BEAT and DefiLlama report current numbers per layer in real time. The amount of public research on Ethereum's extraction surface exceeds the equivalent body of work on Solana or Hyperliquid by an order of magnitude. But the operator concentration that emerges from that legible architecture is comparable to the operator concentration the other two chains produce. The architecture does not produce operator competition; the architecture produces a legible map of where the operator concentration sits. Whether that legibility is itself a structural improvement — whether knowing exactly which firm extracts which dollar at which layer is meaningfully different from not knowing — is the chapter's open question and the question Part III closes on.

What changes when… ​

Part III is complete. The three chain-architecture chapters have documented the three structural patterns: Solana's vertical integration (compressed surface, five-firm retail-trade stack); Hyperliquid's in-consensus matching (eliminated surface, four-locus re-concentration); Ethereum's deliberately-dispersed PBS + L2 + restaking trifecta (legible architecture, comparable operator concentration). Part IV opens with the honest reckoning. Chapter 11 picks up the four chronic-loser categories the book has been building toward — retail traders unaware of toxic flow, passive LPs in informed-flow pools, slow market makers, validators without infrastructure relationships — and develops them as the systematic cost of the structures Part III has documented. Chapter 12 closes the book with the forward-looking trends: ePBS arrival; the decentralised-sequencer roadmap (Espresso, Astria, Superchain); shared-state cross-chain interop; the appchain thesis; the intent-based architecture trend (Anoma, CoW, UniswapX). The forward look is constrained by the structural finding Part III has assembled: any new architecture must reckon with the operator-concentration pattern that has emerged across three radically different chain designs.

Footnotes and sources ​

  1. Coinpedia on X, Hyperliquid #1 Crypto Application Revenue Q1 2026, https://x.com/CoinpediaNews/status/2040351547629154534; PANews, Solana Q1 2026 Network Activity Report, https://www.panewslab.com/en/articles/019da9c6-e9d5-76c0-904a-878418fd7fbc; Odaily, L1 Value Capture Shrinks, https://www.odaily.news/en/post/5209445. Hyperliquid Q1 2026 application revenue ~$144.8M (#1); Solana Q1 2026 chain REV ~$89.5M (#2, down 68% YoY); Ethereum Q1 2026 chain fees ~$82M (#3). Already cited Chapters 8, 9. Accessed 2026-05-14. ↩︎

  2. relayscan.io 24-hour snapshots (May 2026), https://www.relayscan.io/. Builder shares: Titan ~52.16%, BuilderNet ~24.63%, Quasar ~15.06%, Eureka ~2.41%, Beaverbuild ~1.81%, Bombora ~1.31%. Already cited Chapters 5, 6, 7. Accessed 2026-05-14. ↩︎

  3. Yellow Research, Ethereum L2 Fee Revenue Competition 2026, https://yellow.com/research/ethereum-l2-fee-revenue-competition-2026; CoinLaw, Gas Fee Markets on Layer 2 Statistics, https://coinlaw.io/gas-fee-markets-on-layer-2-statistics/; Pine Analytics, The Compression of L1 Value Capture, https://pineanalytics.substack.com/p/the-compression-of-l1-value-capture. Base captures ~62% of all L2 fee revenue, ~$185,291/day = ~$67M/year, ~$870M cumulative since launch, ~$55M net profit in 2025 (only profitable L2). Priority fees ~86.1% of daily Base sequencer revenue; 64.9% of priority fees from just 250 addresses. Accessed 2026-05-14. ↩︎

  4. BlockEden, EigenLayer Crosses $18B TVL, https://blockeden.xyz/blog/2026/03/20/eigenlayer-18b-tvl-vertical-avs-specialization-restaking-evolution/; Fensory, EigenLayer TVL and Restaking Market Analysis 2026, https://fensory.com/intelligence/defi/eigenlayer-tvl-restaking-market-analysis-2026; Mitosis University, EigenLayer's Restaking Economy Hits $25B TVL, https://university.mitosis.org/eigenlayers-restaking-economy-hits-25b-tvl-too-big-to-fail/; EigenLayer App, AVS Roster, https://app.eigenlayer.xyz/avs; Blockdaemon, Mainnet Slashing Now Live, https://www.blockdaemon.com/blog/eigenlayer-mainnet-slashing-now-live. EigenLayer TVL ~$15.3–19B across 4.36–4.6M ETH at ~1,900 active operators; ~94% market share; mainnet slashing live 17 April 2025; AVS roster includes EigenDA, AltLayer MACH, Hyperlane, Witness Chain. Accessed 2026-05-14. ↩︎ ↩︎ ↩︎ ↩︎

  5. Pine Analytics, The Compression of L1 Value Capture, op. cit.; DataWallet, EIP-4844 Explained, https://www.datawallet.com/crypto/eip-4844-explained. L2-to-L1 blob fee payment collapsed from ~$113M (2024) to ~$10M (2025), >90% decline. Accessed 2026-05-14. ↩︎

  6. CoinLaw, Ethereum Gas Fees Statistics, https://coinlaw.io/ethereum-gas-fees-statistics/; Bitcoin Ethereum News, Ethereum's Fusaka Upgrade Raises Blob Fees, https://bitcoinethereumnews.com/ethereum/ethereums-fusaka-upgrade-raises-blob-fees-impacts-network-economics/. Daily L1 gas revenue fell from peak ~$23M/day to ~$6.3M/day in 2025; ETH ~0.23%/year inflationary post-Pectra; daily ETH burn fell to ~3.26 ETH/day (-71%); industry estimates blob fees could be 30–50% of total ETH burn by 2026. Accessed 2026-05-14. ↩︎

  7. Maire, Sviridov, Capponi, Wattenhofer, Detecting Cross-Chain Arbitrages in the Wild, arXiv 2501.17335 (ACM SIGMETRICS 2024), https://arxiv.org/html/2501.17335v2; Flashbots Collective, Detecting Cross-Chain Arbitrages in the Wild, https://collective.flashbots.net/t/detecting-cross-chain-arbitrages-in-the-wild/5107. 242,535 cross-chain arbitrages over Sep 2023 – Aug 2024 across 9 chains; ~$868.64M volume; ~$10.05M searcher revenue / $8.65M net profit; 5.5× growth over the study; 58.35% L1-to-L2, 35.67% L2-to-L2; 66.96% use pre-positioned inventory (~9 seconds) vs 242 seconds for bridges; top 5 addresses >50% of trades; one address ~40% of daily post-Dencun volume. Accessed 2026-05-14. ↩︎ ↩︎

  8. CoinLaw, ETH Staking Statistics 2026, https://coinlaw.io/eth-staking-statistics/; Datawallet, Ethereum Staking Statistics 2026, https://www.datawallet.com/crypto/ethereum-staking-statistics-and-trends; beaconcha.in, https://beaconcha.in/charts/validators. Validator count May 2026 ~1.06–1.1M; total staked ETH ~35.86M (~28.91%); compressing post-Pectra. Accessed 2026-05-14. ↩︎

  9. Pine Analytics, The Compression of L1 Value Capture, op. cit.; DataWallet, EIP-4844 Explained, op. cit.; Zeeve, Ethereum Fusaka Upgrade, https://www.zeeve.io/blog/ethereum-fusaka-upgrade-how-does-it-affect-ethereum-and-the-ecosystem-of-rollups/. Fusaka (3 December 2025) introduced PeerDAS (EIP-7594) and EIP-7918, reduced per-node blob bandwidth ~8×; two BPO forks pushed blob target 3 → ~14 and max 6 → 21 across Dec 2025 – Jan 2026; long-term Danksharding goal ~128 blobs per block. Accessed 2026-05-14. ↩︎

  10. relayscan.io, op. cit.; Wahrstätter on X, MEV-Boost Update, https://x.com/nero_eth/status/2004972045516292339. Relay shares: Ultrasound ~35.71%, Titan Relay ~26.09%, BloXroute Max-Profit ~13.38%, BloXroute Regulated ~12.30%, Aestus ~7.32%, Flashbots ~2.09%. Already cited Chapter 5. Accessed 2026-05-14. ↩︎

  11. The Block, Ethereum Block Builder Eureka Labs Funding "Programmable Blocks", https://www.theblock.co/post/394923/ethereum-block-builder-eureka-labs-funding-programmable-blocks; Phemex News, Eureka Labs Secures $6.7 Million Seed Funding, https://phemex.com/news/article/eureka-labs-secures-67-million-in-seed-funding-led-by-spark-capital-and-collider-ventures-68729. Eureka Labs founded December 2024; $6.7M seed Q2 2025 (Spark Capital + Collider Ventures); ~1.5–2.4% builder share; "programmable blocks" product positioning. Accessed 2026-05-14. ↩︎

  12. Ethereum Foundation, Checkpoint #9, 10 April 2026, https://blog.ethereum.org/2026/04/10/checkpoint-9; Phemex, Ethereum Foundation Delays Glamsterdam Upgrade, https://phemex.com/news/article/ethereum-foundation-delays-glamsterdam-upgrade-due-to-epbs-challenges-72665; Blockonomi, Justin Drake's Strawmap, https://blockonomi.com/ethereum-foundations-justin-drake-unveils-strawmap-roadmap-with-seven-forks-planned-through-2029/. ePBS / Glamsterdam slipping past Q2 2026; "Q2 seems to me to be unlikely" per Checkpoint #9; realistic Q3/Q4 2026 with risk of 2027 slip if FOCIL added. Justin Drake's February 2026 Strawmap outlines seven forks through 2029 with one-consensus + one-execution headliner per fork. Already cited Chapter 5. Accessed 2026-05-14. ↩︎

  13. L2BEAT, Total Value Secured, https://l2beat.com/scaling/tvs; Spoted Crypto, L2 Comparison 2026, https://www.spotedcrypto.com/defi-layer-2-comparison-2026-arbitrum-base-optimism-zksync/; L2BEAT, Stages, https://l2beat.com/stages. Total L2 TVL ~$48B across ~73 active rollups; top 3 (Arb/Base/OP) ~90% of transaction volume; Stage 1 achieved by Arbitrum One, Base, OP Mainnet, Starknet, Scroll, Ink; no L2 yet Stage 2. Accessed 2026-05-14. ↩︎

  14. Eco Support, Ethereum L2 Sequencers: Centralized Today, Decentralized Tomorrow, https://eco.com/support/en/articles/14798711-ethereum-l2-sequencers-centralized-today-decentralized-tomorrow; Unchained, What Are Sequencers in Layer 2 Protocols, https://unchainedcrypto.com/what-are-sequencers-in-layer-2-protocols-such-as-optimism-arbitrum-and-base/. Every major L2 in production in May 2026 operates with a single centralised sequencer. Accessed 2026-05-14. ↩︎

  15. Yellow Research and Pine Analytics, op. cit.; The Block, Arbitrum DAO Revenue from Timeboost, https://www.theblock.co/post/396841/arbitrum-dao-2m; Messari, L2 Revenue Comparisons Q1 2026, https://messari.io/research. Combined L2 sequencer revenue ~$150–250M annually; Base ~62% of L2 fee revenue / ~$67M annualised; Arbitrum DAO gross income ~$26M annualised Q4 2025; Timeboost ~$3M/year run-rate / ~26% of Arbitrum DAO income. Accessed 2026-05-14. ↩︎

  16. Arbitrum Docs, Timeboost: Gentle Introduction, https://docs.arbitrum.io/how-arbitrum-works/timeboost/gentle-introduction; DL News, Arbitrum Scooped Up $3M from Timeboost, https://www.dlnews.com/articles/defi/arbitrum-gets-3m-revenue-bump-from-timeboost/; Castro, Capponi, Sviridov et al., Timeboost Auction Analysis, arXiv 2509.22143, September 2025, https://arxiv.org/abs/2509.22143; Blockworks, Timeboost Goes Live, https://blockworks.co/news/arbitrum-timeboost-live-dao-revenue. Timeboost: sealed-bid second-price auction; 200ms express lane priority for 60-second slot; ~$3M generated first 3 months; 20–30% of daily DEX volume on Arbitrum; two entities win >90% of auctions per Castro et al.; revenue trended downward as concentration tightened. Accessed 2026-05-14. ↩︎

  17. Espresso Systems, Espresso and Caldera: Building Infrastructure for Cross-Chain Composability, https://medium.com/@espressosys/espresso-and-caldera-building-infrastructure-for-cross-chain-composability-fb68f73ed21a; Astria, Shared Sequencer Network, https://www.astria.org/blog/astria-the-shared-sequencer-network; Optimism, Welcoming Unichain to the Superchain, https://www.optimism.io/blog/welcoming-unichain-to-the-superchain; DiaData, RaaS Map / Conduit, https://www.diadata.org/rollup-as-a-service-raas-map/conduit/. Espresso Sequencer mainnet 12 February 2026; HotShot consensus; sub-second finality; throughput target 25 MB/s from current 5+ MB/s; first integrations with Caldera + Arbitrum Orbit. Optimism's Superchain shared sequencer (with Espresso + OP Labs) targeted for 2026. No major standalone L2 has decentralised its sequencer at scale as of May 2026. Accessed 2026-05-14. ↩︎

  18. Maire et al., arXiv 2501.17335, op. cit. Detail in [7:1]. Accessed 2026-05-14. ↩︎

  19. Gogol, Köppelmann, Estensoro et al., Quantifying Cross-Rollup MEV, arXiv 2406.02172, October 2024, https://arxiv.org/abs/2406.02172; Blocknative, Fundamentals of Cross-Chain MEV, https://www.blocknative.com/blog/fundamentals-of-cross-chain-mev. >500,000 unexplored arbitrage opportunities identified; 0.03%–0.05% of trading volume on Arbitrum/Base/Optimism; ~0.25% on zkSync Era; opportunities persist 10–20 blocks on average. Accessed 2026-05-14. ↩︎

  20. EigenLayer Docs, Restaking Overview, https://docs.eigencloud.xyz/eigenlayer/restakers/concepts/overview; BlockEden source per [4:1]; Fensory source per [4:2]; Mitosis University source per [4:3]. EigenLayer TVL ~$15.3–19B (consolidated snapshot ~$18B); 4.36–4.6M ETH (~4.6M); ~1,900 active operators; ~94% restaking market share; all-time high ~$19.7B (some sources cite ~$25B aggregate including Eigen Cloud); slashing live 17 April 2025; ELIP-006 Redistributable Slashing on mainnet 2026; AVS roster includes EigenDA, AltLayer MACH, Hyperlane, Witness Chain. Accessed 2026-05-14. ↩︎