Ethereum L2s Process 100K TPS But Vitalik Says Model 'No Longer Makes Sense'—Did We Scale Wrong?

Back in February, Vitalik dropped a statement that’s been rattling around in my head ever since: the rollup-centric roadmap that positioned L2s as the way Ethereum would scale “no longer makes sense.”

As someone who’s been building L2 infrastructure at Polygon Labs and Optimism for the past 6 years, this felt like whiplash. We spent half a decade optimizing rollups, and now the narrative is shifting? But after digging into what’s actually happening on-chain and what’s coming with Glamsterdam, I think Vitalik’s reframing is spot-on—though the implications are more nuanced than “L2s failed.”

What Actually Changed?

Let’s start with the facts. According to L2BEAT data, Ethereum L2s are collectively processing around 100,000 TPS—a 6,500x improvement over mainnet capacity. Base alone handles 60%+ of L2 activity (142 TPS with 38.7% growth), Arbitrum sits on $16-19B TVL (41% of the L2 market), and the total L2 ecosystem holds $38-44B in value locked.

By traditional metrics, we absolutely nailed scaling. L2 fees dropped 90%+ post-EIP-4844 blobs (now $0.001-$0.05 vs mainnet), we’re processing ~2M daily transactions across L2s vs ~1M on L1, and proof generation improved 60x in 2026.

But here’s what Vitalik actually said: scaling Ethereum should mean creating “large quantities of block space that is backed by the full faith and credit of Ethereum.” If your connection to L1 is “mediated by a multisig bridge,” then you’re not scaling Ethereum—you’re building a separate chain that uses Ethereum for marketing.

That stung, but it’s technically correct. And it gets to the heart of the problem: fragmentation.

The Fragmentation Problem

We now have Base with 70% of L2 active addresses, 34 OP Stack chains live on the Superchain, and dozens more L2s competing for liquidity. But cross-L2 communication remains siloed—you need bridges, which means UX friction + security risk + MEV extraction on every hop.

From a technical standpoint, we created dozens of isolated execution environments that happen to settle to the same L1. Is that really “Ethereum scaling” or is it “Ethereum brand fragmentation”?

Meanwhile, Solana’s sitting over there with a monolithic approach: single execution layer, unified liquidity, no bridges, simpler mental model. Developers building on Solana don’t think about “which L2?” because there’s only one chain. And honestly? That’s a better developer experience, even if it sacrifices some of Ethereum’s modularity benefits.

Glamsterdam Changes the Equation

The Glamsterdam upgrade (targeting H1 2026) is a game-changer:

  • ePBS (EIP-7732): Enshrined Proposer-Builder Separation could reduce MEV extraction by up to 70%
  • BALs (EIP-7928): Block-Level Access Lists for better state management
  • Performance target: 10,000 TPS on L1 with 78% lower gas fees

If L1 can deliver 10K TPS with $0.01 transactions and built-in MEV protection, what’s the L2 value proposition beyond experimentation?

Vitalik’s answer: L2s should focus on providing value beyond basic scaling. Privacy features, application-specific design, ultra-fast transaction confirmation (<100ms instead of 12s blocks), non-financial use cases. Think of L2s as “specialized service providers” rather than “Ethereum’s only scaling path.”

That reframing makes sense. Optimistic rollups were always a pragmatic compromise (7-day withdrawal delays, fraud proof complexity). ZK rollups are elegant but computationally expensive. If L1 can scale directly, rollups become an optimization for specific use cases—not the only path forward.

So… Did We Scale Wrong?

I don’t think we scaled “wrong”—I think we’re maturing. The rollup-centric roadmap was the right bet in 2020 when L1 fees hit $200+ and scaling was existential. L2s bought us time and proved that modular architectures can work.

But now that EIP-4844 dropped blob costs by 90%+, proof generation improved 60x, and Glamsterdam promises L1 scaling + MEV protection, the calculation changes. If you’re building a new DeFi protocol today, do you:

  • Launch on L1 (expensive but maximum composability)?
  • Pick a specific L2 (cheaper but siloed liquidity)?
  • Go multi-L2 from day one (complex but maximum reach)?

Increasingly, the answer might be “just launch on L1” if Glamsterdam delivers on its promises.

The Real Question

For those of us building L2 infrastructure, Vitalik’s statement isn’t “you failed”—it’s “your job changed.” Instead of pure scaling, L2s need differentiated value:

  • Privacy-focused L2s (Aztec, Railgun)
  • Gaming chains with custom execution (IMX, Ronin)
  • Ultra-low-latency L2s for DeFi (<150ms finality)
  • Application-specific rollups with custom precompiles

The question isn’t “did we scale wrong?”—it’s “what should L2s optimize for in a world where L1 can scale?”

Curious what other L2 builders think. Are we pivoting toward specialization? Should Ethereum Foundation prioritize L1 scaling over L2 interoperability? And if we’re honest, was cross-L2 fragmentation always the wrong trade-off?

cc: other L2 engineers who’ve been through this journey—what’s your read on Vitalik’s reframing?


Sources: CoinDesk coverage of Vitalik’s L2 statement, L2BEAT activity data, Ethereum Foundation Glamsterdam upgrade documentation, 21Shares L2 research, QuickNode analysis

This hits so hard from a frontend dev perspective. I’ve been building DeFi interfaces for 3 years now, and the L2 fragmentation problem is real for users.

When someone lands on our dApp, the first question isn’t “what do you want to do?” it’s “which L2 are you on?” And if they’re on the wrong one? Bridge to another L2, pay fees, wait for confirmation, deal with slippage, hope the bridge doesn’t get exploited… it’s a terrible UX.

We literally lost 40% of users during L2 onboarding last quarter. Not because our product was bad—because choosing between Base, Arbitrum, Optimism, Polygon, zkSync felt overwhelming. One user told me “I just wanted to try yield farming, not get a PhD in Ethereum architecture.”

And here’s the thing: my normie friends who use Solana don’t have this problem. They download Phantom, connect to any Solana dApp, and it just works. One chain, one mental model, unified liquidity. Yeah, it sacrifices some of Ethereum’s modularity benefits, but honestly? Most users don’t care about modularity—they care about “does this work without a tutorial?”

Reading your post about Glamsterdam potentially delivering 10K TPS with 78% lower fees… if that happens, should we just tell projects to launch on L1 and skip the L2 complexity entirely? Because from a UX standpoint, having 34 OP Stack chains and Base with 60% market share feels more like brand dilution than scaling success.

I’m curious: if you’re an L2 builder who’s been optimizing for pure scaling for 6 years, how do you pivot to “specialized service providers”? Like, do you rebuild your entire value prop around privacy features or gaming-specific optimizations? That’s a massive strategic shift, not just a narrative reframe.

Maybe Vitalik’s right that L2s need to specialize, but the market already decided which L2s win (Base/Arbitrum/Optimism), and everyone else is fighting for scraps. If you’re building a new L2 in 2026 without a clear differentiation story beyond “we’re 5% faster,” you’re probably going to get wrecked.


(Still learning here—would love to hear from other builders who’ve dealt with cross-L2 UX nightmares)

I actually disagree that we “scaled wrong”—I think we’re seeing the natural evolution of a modular architecture, and Vitalik’s reframing is validation, not condemnation.

Let me push back on the narrative here: L1 scaling and L2 specialization aren’t mutually exclusive. They’re complementary.

Why L1 Scaling is Good News for L2s

Glamsterdam delivering 10K TPS with 78% lower fees on L1? That’s fantastic for L2s. Here’s why:

  1. Cheaper settlement costs: L2s batch transactions and post to L1 for security. If L1 fees drop 78%, L2 operating costs plummet, which means even lower fees for end users or higher margins for L2 operators.

  2. Better composability: If L1 is cheap enough for more activity, you get the best of both worlds—high-value transactions on L1 (maximum composability), specialized use cases on L2s (optimized execution).

  3. Reduced pressure to compromise: When L1 was $200/transaction, L2s had to make security trade-offs (multisig bridges, centralized sequencers) to get fees down. If L1 is $0.01, L2s can be more conservative on security and still be cheaper.

The “Specialized Service Provider” Model Makes Sense

Emma’s point about UX fragmentation is real, but that’s a coordination problem, not an architectural one. The solution isn’t “kill all L2s”—it’s chain abstraction and intent-based architectures.

Think about it: Aztec provides privacy via ZK proofs. IMX and Ronin optimize for gaming (low latency, custom game logic). Starknet experiments with Cairo for provable computation. These aren’t “Ethereum knockoffs”—they’re specialized execution environments that couldn’t exist on a monolithic L1.

Solana’s unified UX is nice for simple cases, but you can’t do private transactions on Solana (everything’s transparent). You can’t run custom precompiles. You can’t experiment with novel consensus mechanisms without forking the entire chain. Ethereum’s modularity lets teams try radical ideas without risking the base layer.

ePBS is the Real Game-Changer

The Glamsterdam upgrade isn’t just about TPS—it’s about MEV protection at the protocol level. ePBS (EIP-7732) could reduce MEV extraction by 70% on L1. That’s a huge deal because it means:

  • Validators can’t front-run users
  • Censorship resistance improves (Inclusion Lists coming in next fork)
  • L1 becomes viable for DeFi again (not just settlement layer)

But L2s can still innovate on MEV. Application-specific rollups can implement custom MEV auction mechanisms. Gaming chains can eliminate MEV entirely by making transactions private until executed. Privacy L2s can use ZK proofs to prevent MEV by design.

Market Consolidation ≠ Failure

Yes, Base/Arbitrum/Optimism dominate. But that’s not a bug—it’s the market working. Most L2s should die because they didn’t have differentiated value props. The survivors will be the ones that offer something L1 can’t:

  • Privacy (Aztec)
  • Gaming optimization (IMX, Ronin)
  • Ultra-low latency (sub-100ms finality)
  • Custom execution environments (Starknet’s Cairo, Fuel’s FuelVM)

Where I Agree with Vitalik

We absolutely over-indexed on “L2s as scaling” and under-indexed on “L2s as experimentation labs.” If every L2 is just “Ethereum but cheaper,” then yeah, Glamsterdam makes them obsolete.

But if L2s lean into specialization—privacy, custom VMs, app-specific optimizations—then we get the modular architecture Ethereum was always supposed to be: secure base layer + diverse execution environments.

TL;DR: This isn’t “we scaled wrong”—it’s “we finished Phase 1 (scaling) and now we’re entering Phase 2 (specialization).” L1 scaling is good news, not a crisis. L2s that only competed on price will die. L2s with real differentiation will thrive.


(Biased take from someone working on zkEVM, but I stand by it—modularity beats monolithic long-term)