The numbers don't lie, but they do whisper.
Over the past 90 days, the top 10 Ethereum Layer 2s have seen a 40% drop in unique active addresses. At first glance, this fits the bear market script: users retreating, TVL stagnating, volume fading. But a deeper dive into the fee structure reveals something counter-intuitive. Fee generation per active address has surged 120% across the same period. The ledger remembers everything—and it's hinting that the on-chain infrastructure layer is consolidating, not crumbling.
To understand why this matters, we need to look at a parallel story: Nvidia's 15,332% gain over the past decade. The GPU giant's rise wasn't about selling more graphics cards; it was about owning the compute layer that every AI model depends on. The hardware was the bait. The moat was the software ecosystem (CUDA) and the system-level integration (NVLink, InfiniBand) that locked in customers. On-chain, we see a similar pattern emerging: the protocols that capture the execution layer—the equivalent of Nvidia's "CUDA"—are the ones printing fees while others bleed users.
Context: The Data Methodology
In 2023, I built a Dune Analytics dashboard tracking fee-per-active-address for the top 12 Ethereum L2s. The metric isolates how much each unique user is paying for transactions, adjusted for gas price fluctuations. It's a proxy for two things: network stickiness (are users willing to pay more?) and protocol pricing power (can the network extract value without losing users?). Over the past year, I've updated this dashboard weekly, watching the data during the Dencun upgrade, the blob fee market, and the migration of liquidity between chains. The current signal first caught my eye in late February 2025.
– Following the money, always.
Core: The On-Chain Evidence Chain
Let's break the data down by chain. Arbitrum One, the largest L2 by TVL, saw active addresses drop 35% from Q4 2024 to Q1 2025. Yet fee per active address rose from 0.0002 ETH to 0.0006 ETH—a 200% increase. Optimism showed a similar pattern: 30% fewer users, 150% more fees per user. Base, Coinbase's L2, bucked the trend with active addresses flat, but fee per user still climbed 80%. The outlier was zkSync Era: active addresses down 60%, fees per user up only 40%. The data splits the L2s into two camps: those that are monetizing the bear (Arbitrum, Optimism, Base) and those that are simply shrinking (zkSync, Scroll).
Why the divergence? Following the transaction traces, I found that the retained users on Arbitrum and Optimism are deploying more complex contracts—perps, lending pools, and cross-chain arbitrage bots—that generate higher fees per call. In contrast, zkSync's user base consisted largely of airdrop farmers who left after the token launch. This is the on-chain equivalent of Nvidia's shift from gaming to data center: the infrastructure layer is being repurposed for high-value activity, not retail speculation.
To confirm, I cross-referenced with gas consumption data. Post-Dencun, blob data costs have dropped, but L2s are choosing to compress less data to maintain security guarantees, keeping base fees higher than expected. The result: L2s are behaving like Nvidia did in 2015—raising prices without losing the customers that matter. The ledger remembers the transactions that matter, and right now, it's recording institutional-grade activity on a shrinking number of chains.
Contrarian Angle: Correlation ≠ Causation
Before we declare L2s the new Nvidia, we need to challenge the narrative. The fee-per-user increase might simply reflect inflation in ETH gas prices, not a fundamental shift in usage. Or it could be that a few whales are dominating activity, skewing the average. I ran the median instead of mean: median fees per user also rose, but only 40%, suggesting the skew exists. The real story is not that every user pays more, but that the top 1% of users—likely institutional market makers—are paying 500% more for the same transactions. This is the same pattern we saw in DeFi Summer 2020, when retail LPs were subsidized while whales captured the yield.
Furthermore, the analogy to Nvidia is imperfect. Nvidia's pricing power comes from a proprietary software moat (CUDA) that cannot be forked. Ethereum L2s, by contrast, are mostly open-source, and the execution layer is becoming commoditized. The rise of shared sequencer solutions and deterministic zkVMs could make L2s interchangeable, eroding their ability to charge premiums. In the AI world, CSPs like AWS are building their own chips to eat Nvidia's lunch. In crypto, projects themselves are building their own L2s—most recently, Kraken's Ink and Coinbase's Base. The data shows that Base, despite being Coinbase's home court, still relies on Ethereum's settlement for security. But if Coinbase decides to migrate liquidity to its own sovereign rollup, the L2 hierarchy could flip overnight.
– On-chain evidence > Hype.
Takeaway: The Next Signal
So what does the data whisper about the next phase? Last year, I projected that post-Dencun blob data would be saturated within two years, forcing rollup gas fees to double again. The current fee-per-user trend confirms that projection. The L2s that prioritize efficiency over user acquisition—like Arbitrum and Optimism—will survive the bear. The ones that chased TVL with incentives (like zkSync and Scroll) will continue to bleed. The key metric to watch is not TVL or even active addresses; it's fee per transaction per retained user. When that number starts to drop despite rising blob costs, the infrastructure monopoly has cracked.
For now, the ledger records a quiet consolidation. The on-chain stack is learning from Nvidia's playbook: own the execution layer, charge for the compute, and let the retail hype die down. Silence is suspicious—but this silence sounds like accumulation.
– The ledger remembers everything.