Apps Eat the Stack: Why L2 Tokens Keep Underperforming

Optimism's token went from $2.18 in early 2025 to $0.23 by October. Arbitrum trades at $0.21 despite sitting under $16.3 billion in total value locked. Layer 2 networks collectively handle 5.19 times Ethereum's mainnet transaction volume. The infrastructure is winning. The tokens are not.

Meanwhile, Hyperliquid (a perpetuals exchange that built its own chain) generated $54 million in fees in a single month and surpassed Coinbase in notional trading volume at $2.6 trillion. Its token is up 25% year-to-date. Pendle, a yield tokenisation protocol, peaked at $8.27 billion in TVL. Base, Coinbase's L2, finished 2025 as the highest-revenue Layer 2. And it does not even have a token.

The pattern is clear enough that it has a name now: the Fat App Thesis. And it is showing up in the revenue data, not just in narrative.

The evidence#

For a decade, crypto operated under the Fat Protocol Thesis: the idea that value accrues to the base layer, not the applications built on top. Invest in Ethereum, not the DEXs. Buy the L1, not the lending protocol. The logic was that protocols capture value from every application above them, while applications compete on thin margins.

That thesis made sense when base layer capacity was scarce and applications were simple. It does not describe what is happening now.

L2 tokens have a value capture problem. Arbitrum and Optimism process enormous transaction volumes, but their tokens are governance tokens, not revenue tokens. ARB and OP do not entitle holders to sequencer fees, MEV revenue, or any direct claim on network economics. They grant voting rights over treasury allocation. Useful for governance participants. Useless for price discovery.

The structural problem is worse than weak utility. Arbitrum unlocks more than 90 million ARB per month into a market that has no mechanism to absorb the supply. Without fee-sharing, buybacks, or staking yield tied to real revenue, the token is a perpetual dilution machine. Strong fundamentals and weak tokenomics produce exactly the chart you would expect: usage up, price down.

Applications are capturing the revenue directly. Hyperliquid routes 97% of its protocol fees into token buybacks. That is a live mechanism that turns usage into scarcity, not a governance proposal or a future roadmap item. The result: $645 million in cumulative buybacks, a token that tracks revenue growth, and a market cap that reflects what the protocol actually earns.

Uniswap activated its fee switch and is projecting $460 million in annual token burns. Pendle captures yield spread on $300 million in TVL tied to HyperEVM products alone. These are applications generating and distributing real revenue to token holders: the thing L2 governance tokens were supposed to do eventually but never did.

Base proved the model without a token at all. Coinbase's L2 generated $369.9 million in ecosystem revenue in 2025: the highest of any Layer 2. All sequencer fees flow to Coinbase. There is no token to dilute, no governance theatre, no unlock schedule. The value accrues to the operator via traditional corporate equity (COIN stock). Base is exploring a token, but the fact that the most successful L2 by revenue ran for two years without one is the sharpest indictment of the current L2 token model.

Why this pattern holds#

Three structural forces are driving the rotation from infrastructure tokens to application tokens. None of them are temporary.

Blockspace became abundant. When L2s launched, the pitch was that they would relieve Ethereum's capacity constraints, and the token would capture value from that scarce resource. But L2s proliferated faster than demand grew. There are now dozens of rollups competing on fees, and fees have been driven toward zero. EIP-4844 (blobs) reduced L2 data costs by over 90%. When blockspace is cheap and abundant, the infrastructure layer commoditises and the application layer differentiates.

Revenue models matured at the app layer. Early DeFi protocols had no revenue model beyond token emissions. The current generation has real fee structures: Hyperliquid charges trading fees, Pendle captures yield spreads, Aave earns interest margins, GMX takes a cut of leveraged positions. These protocols can point to income statements. L2 governance tokens cannot, because the revenue flows to the sequencer operator, not the token holder.

Users follow applications, not chains. The most successful L2 by user growth is Base, not because of its technical architecture (it is a standard OP Stack rollup) but because Coinbase funnels its 100+ million users into it. Hyperliquid's growth came from building the best perpetuals product, not from chain-level marketing. When user acquisition is driven by application quality rather than chain incentives, the application captures the relationship and the value that comes with it.

What breaks it#

The L2 token thesis could recover if two things change.

Decentralised sequencers with fee distribution. If Arbitrum or Optimism decentralise their sequencers and route a share of fees to token stakers, the governance-only token becomes a revenue-bearing asset. Optimism has signalled intent to do this. Arbitrum's governance has debated it. Neither has shipped it. Until they do, the tokens remain indirect claims on value that someone else captures.

Token-gated access to scarce resources. If L2 networks develop features that require token ownership (priority transaction ordering, dedicated blockspace, or application-specific sequencing), the token gains utility beyond voting. Some newer L2 designs are exploring this. But the established networks would need to retrofit it, which means governance votes, migration risk, and timeline uncertainty.

Both paths are plausible. Neither is imminent. And the longer the gap persists between L2 usage growth and L2 token performance, the more capital rotates toward applications that have already solved the value capture problem.

What readers should take from this#

The infrastructure-versus-application debate is not abstract; it is directly observable in the price charts. OP and ARB are down 80–90% from their peaks despite record network usage. HYPE is up while generating hundreds of millions in real fees. The market is not confused, it is repricing where value actually accrues in the stack.

For positioning:

• L2 tokens are a bet on future tokenomics reform, not current economics. If you hold ARB or OP, you are betting that governance will eventually activate fee-sharing or sequencer decentralisation. That may happen. It has not happened yet.
• Application tokens with live revenue mechanisms are the closer analogue to equities. Hyperliquid's buyback model, Uniswap's fee switch, and Pendle's yield capture are all legible to traditional investors. That matters as institutional capital enters crypto with a revenue-first framework.
• The tokenless L2 model is not an anomaly. Base demonstrated that L2 economics work fine without a token: the value just accrues to equity holders instead. If Base launches a token, watch whether it includes real fee-sharing or repeats the governance-only pattern.

The Fat Protocol Thesis was right for a specific era: when blockspace was scarce, applications were thin, and the protocol layer captured value by default. That era ended when L2s made blockspace cheap and applications learned to monetise directly. The stack did not change. Where the money settles did.