Base and Arbitrum now hold 80% of L2 liquidity — and that's

The Ethereum Layer 2 race, once framed as a sprawling contest among dozens of rollups, has quietly narrowed to a two-horse field. By mid-2026, two chains hold what no L2 pairing has held before: roughly four out of every five dollars secured across the entire ecosystem.

80% in Two Chains: The State of L2 Consolidation in Mid-2026

The Ethereum L2 market has consolidated into a duopoly: Base and Arbitrum One together secure roughly 80% of all value across the rollup ecosystem. On L2BEAT's live Total Value Secured measure, Base leads at about $11.49B (~40% share) with Arbitrum One close behind at $10.12B (~39% share) — the highest concentration ratio in L2 history, spread across roughly 73 active rollups tracked by L2BEAT.

Quick Answer: Base (~$11.5B) and Arbitrum One (~$10.1B) now secure roughly 80% of all Ethereum L2 value, per L2BEAT's mid-2026 Total Value Secured data. Across ~73 active rollups holding $40B–$48B combined, the top two dominate while the third-place chain, OP Mainnet, holds only ~$1.4B.

Outside the top two, the drop-off is steep. The broader L2 ecosystem holds an estimated $40B–$48B in total value, yet the third-ranked chain, OP Mainnet, secures only about $1.42B on the same L2BEAT measure — a fraction of either leader. ZKsync Era registers just $202M in Total Value Secured, underscoring how thin the field becomes below the front-runners.

• Base: ~$11.49B TVS (~40% share) — the leader by value secured [L2BEAT]
• Arbitrum One: ~$10.12B TVS (~39% share)
• OP Mainnet: ~$1.42B TVS
• ZKsync Era: ~$202M TVS

This consolidation accelerated in the first quarter of 2026. As liquidity-incentive programs on mid-tier rollups expired, chains in the $200M–$1B TVL band recorded net outflows, according to Yellow research [Q1 2026]. With rewards drying up, capital drained back toward the two venues with the deepest liquidity and largest user bases, reinforcing the gap rather than narrowing it.

That dynamic raises the structural question this guide unpacks: has the L2 market reached a winner-take-most equilibrium, and what does that mean for the 70-plus chains that are neither Arbitrum nor Base? For traders, the answer determines where liquidity, protocol depth, and security maturity actually concentrate — and whether routing capital outside the top two still makes sense in the second half of 2026.

TVS vs. TVL: Why the L2 Rankings Flip Depending on Which Number You Use

The reason two credible sources can crown different L2 leaders is that they measure different things. L2BEAT reports Total Value Secured (TVS) — every bridged or locked asset a rollup's fraud or validity proof system protects — an infrastructure safety metric. DeFiLlama reports DeFi TVL — redeemable deposits actively working inside lending, DEX, and perpetuals applications — an ecosystem activity metric. Neither is wrong; they answer different questions, and either denominator alone can invert the apparent ranking.

The inversion is not theoretical. By TVS in L2BEAT's live summary, Base leads at roughly $11.49B (~40% share) ahead of Arbitrum One at ~$10.12B (~39% share) . Flip to DeFiLlama-style TVL for the same period and Arbitrum leads at ~$16.9B–$18B (40–44% share), with Base at ~$10.7B–$13.5B (~28–33% share) . A reader citing L2BEAT and a reader citing DeFiLlama can describe what looks like two different markets — same chains, opposite podium.

The most extreme divergence is ZKsync Era. L2BEAT records roughly $202M in TVS, while DeFiLlama shows ~$4.1B–$4.5B in DeFi TVL for the same window . That ~20x gap is a methodology gap, not a data error: DeFi deposits circulating inside applications are not the same assets a rollup's proof system is measured as securing. The recovery narrative depends entirely on the denominator too — ZKsync's DeFi TVL rebounded roughly 5x from a late-2024 post-airdrop low near $900M .

Base and Arbitrum together represent about 80% of L2 value secured under the TVS lens , which is why this guide anchors the consolidation story to that metric — but the caveat travels with every figure that follows. As L2BEAT frames its own methodology, the number is only as meaningful as the definition behind it.

"Total Value Secured reflects the assets a rollup's proof system actually protects — it is a security metric, not a popularity contest," notes the research team at L2BEAT, whose summary dashboard separates value secured from application-level activity.

So the rule of thumb before comparing any two L2 statistics is simple: confirm the denominator first. Ask whether a figure describes value secured by the rollup's proofs (L2BEAT TVS) or deposits deployed inside apps (DeFiLlama TVL), because the two answer different questions and can produce leaderboards that look like they describe unrelated markets . Get that wrong and every downstream comparison — share, ranking, growth rate — inherits the confusion.

Security Stage Ratings: The Differentiator That Fees No Longer Are

With the denominator settled, the next question is what actually separates one rollup from another in 2026 — and the answer is no longer cost. L2BEAT's decentralization Stage rating is now the primary risk differentiator, because fees have converged so tightly that they rarely change a retail decision. A Stage rating measures one thing: whether you can exit a rollup without the operator's permission if its sequencer misbehaves or censors your withdrawal. That exit-path safety, not headline TVL, is what distinguishes today's leaders .

Fees stopped being a differentiator after EIP-4844 — the Dencun upgrade shipped in March 2024 — cut L2 data-posting costs by roughly 80–90% through blob transactions . Per-transaction costs now sit below $0.10 across every major chain: Base near $0.05, ZKsync Era near $0.07, Arbitrum around $0.08–$0.09, OP Mainnet near $0.09, and Starknet ranging $0.05–$0.19 depending on proof load . For any position under roughly $10,000, those gaps are marginal — a few cents per swap — which pushes the real decision onto security stage, liquidity depth, and which protocols are natively deployed.

The Stage system runs on three tiers. Stage 0 means the operator multisig retains full override authority — users cannot exit permissionlessly. Stage 1 means permissionless or functional fraud/dispute proofs exist, so users can withdraw without operator consent. Stage 2 means full decentralization, and no chain has reached it as of mid-2026 .

Four of the majors clear the Stage 1 bar: Arbitrum One via its BoLD dispute system, Base after reaching Stage 1 in 2025, OP Mainnet, and Starknet. Each gives you a meaningful escape hatch if the sequencer censors withdrawals . ZKsync Era is the notable holdout at Stage 0. Its Boojum validity-proof architecture delivers strong cryptographic guarantees that state transitions are correct — but correctness is not the same as exit rights. Matter Labs' operator multisig still retains override authority, so in a censorship scenario users cannot permissionlessly exit. That makes ZKsync a structurally different risk profile regardless of its proof type .

One nuance matters more than any single rating: a Stage 1 badge does not mean the sequencer is decentralized. All top L2s, the Stage 1 chains included, still operated with centralized or partially centralized sequencers as of June 2026 . The Stage measures exit-path safety — your ability to leave — not day-to-day liveness or ordering control.

"Stage ratings answer a single question: if the operator turns hostile, can you get your assets out without their permission? That is exit safety, not sequencer decentralization — two things retail traders routinely conflate." — L2BEAT research framing (source: L2BEAT Scaling Summary)

Fees and Throughput After Dencun: Where Sub-$0.10 Actually Differs

After the security-stage gap, fees are the axis where the major L2s have effectively stopped competing. Per-transaction costs now sit below $0.10 across every leading rollup, so the practical fee spread is under $0.15 end to end — noise for any retail position under roughly $10,000 . This convergence dates to EIP-4844, the Dencun upgrade of March 2024, which cut L2 data-posting costs 80–90% by moving calldata into blob transactions . Once data availability got cheap, fees compressed toward a floor, and the retail decision migrated to liquidity depth, native protocols, and exit-security stage.

The mid-2026 fee snapshot shows how narrow the band has become: Base runs about $0.05, the lowest among optimistic rollups; ZKsync Era sits near $0.07; Arbitrum lands around $0.08–$0.09; OP Mainnet near $0.09; and Starknet ranges $0.05–$0.19 depending on proof load . The delta between the cheapest and most expensive common transaction is roughly $0.04 — smaller than the rounding error on most swaps.

Throughput is where the chains genuinely diverge. Base is the busiest execution venue, processing about 12.89M daily transactions from roughly 382,500 daily active users and leading optimistic rollups at ~89 TPS as of April 2026 . It became the first L2 to sustain daily transaction counts above Ethereum mainnet in late 2025 and recorded 4M+ single-day transactions on multiple occasions in early 2026 . That volume reflects distribution and consumer activity more than transaction size.

Arbitrum tells the opposite story. It runs about 4.3M daily transactions from ~129,000 daily users at ~40–60 TPS — lower raw throughput than Base, but a higher average transaction value that mirrors its DeFi-heavy composition of large perpetual trades and lending interactions . OP Mainnet handles roughly 2.35M daily transactions from ~19,300 daily users, functioning primarily as the Synthetix hub and the governance and sequencer-coordination layer for the OP Stack Superchain rather than a standalone high-volume venue .

"Once blobs cut data costs to near-zero, sub-cent fee differences stopped being a moat. The activity a chain attracts — and the value of what moves through it — now says more than its gas quote," — L2BEAT Scaling Summary framing (source: L2BEAT).

The practical read for retail is direct: the $0.04 per-transaction gap between chains is not a reason to route capital anywhere. Base wins on raw activity and marginally lower cost, Arbitrum wins on transaction depth, and OP Mainnet is a coordination hub, not an execution destination. The meaningful choices live elsewhere — in liquidity, native protocols, and the exit-safety stage covered above.

DeFi Ecosystem Depth: Where the Protocols Actually Live in 2026

Liquidity depth, not fees, is where the top L2s genuinely separate, and in 2026 Arbitrum One is the deepest DeFi venue on Ethereum. It hosts the mature stack of perps, lending, and AMM liquidity that large positions require — GMX, Camelot, Radiant, Uniswap, Aave, and Curve all run natively — with GMX alone generating more than $180M in annualized protocol revenue on Arbitrum in early 2026 . That makes it the default destination for institutional-grade order books, sizeable lending exposure, and perps that need slippage-resistant depth.

Base is the opposite profile: the highest-traffic consumer chain, backed by Coinbase's 100M+ verified users and built for retail onboarding, high-frequency swaps, and Coinbase-integrated products . Its DeFi ecosystem is younger than Arbitrum's but growing faster by transaction count than any other L2, which is why it wins distribution and everyday flow even as Arbitrum holds the deeper protocol liquidity.

"The market is diverging into winners and dead weight — the general-purpose rollups with real distribution or real liquidity are pulling away, while the mid-tier chains that lived on incentives are bleeding out," — Yellow Research, 2026 (source: Yellow).

OP Mainnet's DeFi identity is narrower but systemically important. It anchors Synthetix's derivatives layer while serving as the sequencer-coordination and governance hub for the OP Stack Superchain — Base, Mode, Zora, and 30+ app-chains . Its headline TVL of roughly $1.9B standalone understates that role; counting the wider Superchain, the figure rises to the $5.6B–$6B range . Think of it as infrastructure plus one flagship protocol rather than a standalone liquidity venue.

Among the ZK chains, ZKsync Era has staged a real recovery — DeFi TVL of roughly $4.1B–$4.5B, up about 5x from a post-airdrop low near $900M in late 2024 . But it has not closed the gap to the top two, and its 2026 positioning leans toward privacy and compliance use cases and ZK-native developer tooling rather than general DeFi liquidity depth. The Stage 0 exit caveat covered earlier still applies to any capital parked there.

Starknet rounds out the picture as a developer- and institutional-compliance-oriented chain: Stage 1 security, a Cairo VM, and fees around $0.05 . Its retail DeFi presence is smaller than its peers', and its distinct technical stack resists direct like-for-like comparison — it competes on architecture and tooling, not on liquidity headlines. The takeaway for allocators is that protocol location, not chain branding, should drive routing: Arbitrum for depth, Base for reach, and the ZK chains for specialized use cases.

Hub-and-Spoke Consolidation: The Market Structure Reshaping L2 Competition

The L2 market is consolidating into a hub-and-spoke structure where a few parent chains capture the economics that dozens of smaller rollups once hoped to keep for themselves. Optimistic rollups — Arbitrum, Base, and OP Mainnet — hold roughly 80% of L2 TVL, while the ZK rollup cohort of ZKsync Era, Linea, Scroll, and Starknet splits the remaining ~20% . Despite ZK proofs' theoretical security edge, optimistic chains still dominate liquidity and activity, and that gap is now structural rather than temporary.

The "new chain premium" that bootstrapped multiple L2s in 2023–2024 has deflated. Dozens of mid-tier general-purpose rollups holding $200M–$1B in TVL saw net outflows in Q1 2026 as their liquidity-incentive programs expired, leaving organic DeFi usage as the only durable retention mechanism . Once the emissions stopped, capital left the chains that had nothing but emissions to offer. This is the "winners and dead weight" split described by researchers: a chain either hosts protocols traders return to, or it bleeds out.

The architecture absorbing that displaced activity is app-chain driven. More than 100 application-specific rollups now run on OP Stack and Arbitrum Orbit, funneling sequencer fees and governance activity back toward their parent hubs rather than distributing economic power outward . OP Mainnet already functions as the governance and sequencer-coordination layer of the OP Stack Superchain — spanning Base, Mode, Zora, and others — rather than as a standalone liquidity venue . Arbitrum's Orbit program plays the equivalent role on its side of the market.

This creates a compounding moat. Every new app-chain deployed on OP Stack or Orbit deepens the parent hub's fee revenue, developer tooling, and shared-sequencer network, which in turn makes the hub more attractive to the next builder. The practical consequence for allocators is that economic gravity concentrates at Arbitrum and Optimism (and, through the Superchain, at Base) rather than fanning out to independent chains. A standalone general-purpose L2 launching mid-cycle now faces a structurally harder task: it must attract TVL without the incentive budgets that worked two years ago and without a hub's built-in distribution.

The key thing to watch is whether any ZK chain can breach the sub-20% TVL ceiling its cohort currently occupies. The candidates are clear — Linea, Scroll, and Polygon CDK–based chains — but none is positioned to break through on its current trajectory . ZKsync Era's own recovery to roughly $4.1B–$4.5B in DeFiLlama TVL from a post-airdrop low near $900M shows ZK liquidity can rebuild , but rebuilding within the 20% band is not the same as expanding it. For now, the hub-and-spoke model rewards the chains that already won.

How to Route Capital Across L2s in 2026: A Decision Framework

Routing capital across L2s in 2026 comes down to matching your position type to a chain's specific strength, because fees no longer separate the field. The decision hierarchy is straightforward: Base for everyday low-friction activity, Arbitrum for size and mature DeFi, the OP Superchain for cross-chain reach, and ZKsync Era or Starknet for privacy or compliance-driven use cases . Below each strength sits a caveat you should price in before allocating.

Quick Answer: Route everyday retail swaps through Base — lowest fees near $0.05 and the highest throughput at ~89 TPS. Anchor large positions, lending, and perpetuals on Arbitrum for the deepest DeFi liquidity, and reserve ZK chains for privacy or compliance needs.

Everyday swaps and high-frequency DeFi → Base. Base posts the lowest per-transaction cost at roughly $0.05, the highest sustained throughput at ~89 TPS, and a direct Coinbase on-ramp reaching 100M+ verified users, all backed by Stage 1 exit safety . It is the default for anyone who wants low friction and does not need institutional-grade liquidity depth.

Large positions, lending, and perpetuals → Arbitrum. At the sizes where slippage costs more than a per-transaction fee, Arbitrum's deeper liquidity wins — it hosts GMX, Aave, Curve, and Camelot, with GMX alone generating over $180M in annualized protocol revenue in early 2026 . Its Stylus environment adds Rust and C++ contract logic, and its ~$0.08–$0.09 fees are immaterial at position scale, backed by the same Stage 1 rating.

Cross-chain interoperability → OP Mainnet / the Superchain. If your goal is bridging simultaneously across Base, Mode, and Zora, the OP Stack's shared messaging reduces bridging overhead, and OP Mainnet acts as the governance and sequencer-coordination layer for that Superchain .

Privacy, compliance, or ZK assurance → ZKsync Era / Starknet. Validity proofs offer different cryptographic guarantees than fraud proofs, but factor in ZKsync Era's Stage 0 status — Matter Labs' operator multisig retains override, so exit is not permissionless — for any large or long-duration allocation . Starknet, at Stage 1, better suits developer-native or institutional-compliance contexts.

Verify before allocating. TVS and security Stage change over time — confirm current stage and value secured on L2BEAT and current protocol-level TVL on DeFiLlama-based trackers before committing capital. The figures in any static article, including this one, will drift.

L2 Outlook H2 2026: Sequencer Decentralization, ZK Milestones, and What Could Break the Rankings

The single catalyst most likely to reshuffle the L2 hierarchy in late 2026 is sequencer decentralization — not fees, not throughput. As of June 2026, every top L2, including Base and Arbitrum, still runs a centralized or semi-centralized sequencer . The first chain to credibly decentralize transaction ordering earns a structural trust advantage no rival can replicate with another cent shaved off gas. Because per-transaction costs already sit below $0.10 across major chains after Dencun , competition has moved up the stack to who can be trusted without a human operator in the loop.

The clearest ZK milestone to watch is ZKsync Era's path from Stage 0 to Stage 1, which requires removing Matter Labs' multisig override so users can exit permissionlessly if the sequencer misbehaves . If achieved, ZKsync would become the first high-TVL ZK rollup at Stage 1 — a shift that could reopen the "ZK versus optimistic" narrative and draw institutional capital currently sitting on the sidelines over exit risk. That matters more now that DeFiLlama-style trackers show ZKsync TVL recovering to roughly $4.1B–$4.5B from a late-2024 low near $900M, a ~5x rebound .

Two scenarios frame the rest of the year:

• Bull case for deepening concentration: Arbitrum Orbit and OP Stack app-chain proliferation keeps compounding — over 100 application-specific rollups already signal a hub-and-spoke shift . If shared-sequencer economics widen the moat, the top two could command 85%+ of L2 TVL by end-2026, up from the current ~80% held by Base and Arbitrum .
• Bear case for current leaders: a major exploit or prolonged sequencer outage on Base or Arbitrum would stress-test Stage 1 exit mechanisms in a live incident for the first time. Both reached Stage 1 without a real emergency ever forcing the escape hatch . User trust, not code correctness, would decide whether TVL migrates.

Overlay the macro. If ETH underperforms BTC in H2 2026, USD-denominated L2 TVL can compress even when native-asset balances hold steady, because most headline figures are dollar-priced. The ETH/BTC ratio is therefore a leading indicator for L2 headline TVL momentum — a falling ratio can shrink the numbers on every tracker without a single user leaving.

"Fees have converged, so the L2 race is now a security-and-trust race — the first team to remove itself from the sequencer becomes the default venue for capital that cares about exit guarantees," reflecting the Stage-based framing that L2BEAT applies across rollups (source: L2BEAT, 2026-06).

The concrete takeaway: watch three dials into year-end — which chain decentralizes its sequencer first, whether ZKsync clears Stage 1, and where the ETH/BTC ratio trends. Base and Arbitrum enter H2 2026 with an ~80% grip , but that lead rests on trust that has never been tested in a live crisis. Allocate to the leaders for depth today, size for the possibility that one clean decentralization milestone rewrites the ranking tomorrow.

Frequently asked questions

What is the difference between L2BEAT TVS and DeFiLlama TVL for Layer 2s?

L2BEAT's Total Value Secured (TVS) measures every bridged or locked asset a rollup's security system protects — an infrastructure metric describing how much value sits under the chain's security envelope. DeFiLlama's DeFi TVL measures redeemable deposits actively working inside protocols on that chain — an activity metric. The same chain can rank differently on each: ZKsync Era shows roughly $202M by L2BEAT TVS but around $4.5B by DeFiLlama TVL . That gap is a methodology difference, not a data error. Always confirm which denominator a source uses before comparing chains.

Is ZKsync Era safe to use in 2026?

ZKsync Era uses validity proofs (its Boojum system) that give cryptographic correctness to every state transition, but it remains Stage 0 on L2BEAT's decentralization scale as of mid-2026 . Stage 0 means Matter Labs' operator multisig retains override authority, and users cannot permissionlessly exit if the sequencer censors withdrawals — unlike Arbitrum One, Base, OP Mainnet, and Starknet, which all reached Stage 1 . Everyday swaps carry low practical risk. Large or long-duration allocations should account for this exit-path limitation when weighing ZKsync against Stage 1 alternatives.

Which Layer 2 has the lowest fees in 2026?

Base and Starknet both average roughly $0.05 per transaction as of mid-2026, the lowest among major L2s, following EIP-4844's (the Dencun upgrade, March 2024) 80–90% reduction in data-posting costs via blob transactions . The full spread across major chains — ZKsync Era ~$0.07, Arbitrum ~$0.08–$0.09, OP Mainnet ~$0.09, Starknet up to ~$0.19 under heavy proof load — stays under $0.15 per transaction . For positions under about $10,000, that difference is marginal, so security stage and liquidity depth matter far more to the decision.

Why does Base rank higher than Arbitrum on L2BEAT but lower on DeFiLlama?

Base leads L2BEAT TVS — roughly $11.49B (~40% share) versus Arbitrum One's ~$10.12B (~39%) — because more total bridged value sits under its security envelope . Arbitrum leads DeFi TVL at roughly $16.9B–$18B (40–44% share) against Base's ~$10.7B–$13.5B because its larger installed DeFi ecosystem — GMX, Aave, Curve, Uniswap — keeps more capital actively deployed inside protocols rather than sitting as bridged assets . It is the same capital in two different states, measured two different ways.

What is a Layer 2 Security Stage and why does it matter for DeFi users?

A Security Stage is L2BEAT's 0-to-2 rating for whether users can exit a rollup without operator permission. Stage 0 means you trust the operator; Stage 1 means permissionless exit or fraud/dispute proofs are live; Stage 2 means full decentralization — and no L2 had reached Stage 2 as of mid-2026 . Arbitrum One (with its BoLD dispute system), Base, OP Mainnet, and Starknet all sit at Stage 1, while all top L2s still ran centralized or partially centralized sequencers as of June 2026 . Stage matters most for large or long-duration positions, where sequencer censorship or failure is a meaningful tail risk.

Enjoyed this article? Subscribe to get new stories by email whenever they're published.

Subscribe