Seventy-Four Rollups, One User Base: The Liquidity Slice Nobody Priced

Hook

Seventy-four rollups. Six with real usage. The rest are renting credibility.

Between March 2024, when EIP-4844 pushed the first blob-carrying block onto Ethereum mainnet, and the Pectra upgrade roughly fourteen months later, the marginal cost of posting a rollup's data to Layer 1 fell by close to an order of magnitude. Blob capacity moved from a target of three per block toward six. Sequencer fee revenue for the largest Layer 2s collapsed in dollar terms even as transaction counts climbed to all-time highs. And the number of production rollups — counting optimiums, validiums, and Orbit-style appchains — kept climbing past eighty.

Markets don't reward the count of chains. They reward the count of users, and the users did not multiply. They shuffled.

That distinction is the entire trade. I have spent the last several months rebuilding the fragmentation model from bridge-level deposit data rather than from TVL dashboards, because TVL dashboards count the same dollar twice, three times, four times as it rotates through incentive programs. The output is uncomfortable for anyone holding a diversified basket of L2 tokens. The marginal rollup is not competing for new capital or new users. It is competing for a slice of a float that was already thin on a per-chain basis before the hundredth chain existed.

Speed is the only currency that never depreciates. Positioning ahead of that recognition is where the alpha sits.

Context: How a Scaling Thesis Became a Chain-Launch Industry

The rollup was originally a narrow engineering answer to a specific problem. Ethereum's execution layer could not scale, and sharding the execution layer had been tried and abandoned as too complex. The compromise was elegant: move execution off-chain, keep data availability and settlement on-chain, and let cryptography or economic games guarantee that the off-chain execution was honest.

Optimistic rollups assumed execution was valid and allowed anyone to challenge within a dispute window. Zero-knowledge rollups proved validity up front with a cryptographic proof. The security models differed, the trust assumptions differed, and the marketing departments promptly flattened both into one word: scaling.

For roughly three years, that word worked. Arbitrum and Optimism competed openly for developer mindshare. zkSync and Starknet launched with enormous token events. Airdrop farming became the default growth channel, and the metric that mattered was the number of wallets that had bridged in and left a balance sitting there. TVL became the scoreboard because TVL was the only number that could be printed on a slide.

The first structural break came from infrastructure, not from cryptography. Rollup-as-a-service providers — Conduit, Caldera, AltLayer, Gelato — turned chain deployment from a multi-quarter engineering project into a configuration file. When Optimism open-sourced the OP Stack and started licensing it under a shared governance umbrella, launching a chain became a business decision rather than a technical one. A gaming studio could spin up its own rollup. A consumer app could spin up its own rollup. A team with a token and a Telegram group could spin up its own rollup.

That is where the arithmetic started to break, and almost nobody noticed because the aggregate charts kept going up.

The second break was mechanical. Before EIP-4844, a rollup's dominant operating cost was calldata — the compressed transaction data it posted to Ethereum. That cost scaled with usage, which meant that as a rollup grew, its cost of growth grew alongside it. Blobs replaced calldata with a separate fee market: temporary data storage, priced independently, and priced dramatically lower.

The effect on unit economics was immediate and brutal. A rollup's cost base stopped being a function of its own success. It became a function of how congested the shared blob market was at any given moment. In quiet blocks, blob space was nearly free. In the occasional congestion spike — usually driven by inscription-style data spam rather than by genuine rollup demand — the blob base fee would spike by several orders of magnitude and then collapse again within hours.

Pectra widened the pipe. More blobs per block meant more supply of the cheapest data in crypto. It also introduced EIP-7702, which let externally owned accounts temporarily execute smart contract code, and it raised the validator staking ceiling. Both changes matter for the long arc and neither matters for the near-term trade.

What matters is this: we now operate in a market where block space is abundant, data availability is nearly free, users are scarce, and the only genuinely constrained resources are distribution and order flow.

The rest of this piece is about what that does to valuation.

Core Analysis

The User Base Is a Fixed Set, and It Is Smaller Than the Chain Count Implies

Start with the number that every ecosystem report buries in an appendix: unique active addresses.

Across the entire rollup landscape, the set of addresses that transact more than once a month on more than one chain is measured in the low hundreds of thousands. The set that transacts more than once a month at all — across every rollup, combined — is measured in the low millions, and that figure is inflated by bots, by airdrop farmers running scripted wallets, and by gas-abstraction relayers generating transactions on behalf of users who never see them.

When I ran cross-platform arbitrage between Compound and Aave in 2020, the first thing I learned had nothing to do with interest rates. It had to do with the difference between capital that was deposited and capital that was committed. Mercenary capital moves the moment the incentive ends. Committed capital stays through a drawdown. The dashboard metric of the era — total value locked — made no distinction between the two, and I watched protocols celebrate nine-figure TVL numbers that evaporated within a week of a rewards program ending.

The rollup landscape inherited that measurement disease and compounded it. Every new chain dilutes the denominator of genuine users while inflating the numerator of wallet counts. A user who bridges to eleven chains is one user and eleven addresses, and every analytics dashboard counts him eleven times.

I rebuilt the overlap model from bridge deposit events for the period covering the last three quarters, using a simple filter: an address counted as a real user only if it had at least two outbound transactions separated by more than thirty days, and at least one transaction that was not a claim, a bridge-in, or a stablecoin transfer. That filter removes the airdrop-farming cohort, which is large and loud, and it removes the pure bridge-rotation cohort, which is larger and quieter.

The output is stark. The three largest rollups by cumulative activity account for the overwhelming majority of filtered users. The long tail — and the tail is now genuinely long — accounts for a rounding error on a per-chain basis. Several chains with nine-figure cumulative bridge deposits support filtered monthly active user counts in the low thousands.

Base is the important exception, and it is important precisely because it is not a technical exception. Base's growth came from a distribution channel that no other rollup possesses: a consumer application with a KYC-completed user base of tens of millions, embedded as a wallet inside an app that people already had on their phones. That is a distribution moat, not a throughput moat. It would have worked on any rollup with adequate block space. Base happened to be the one it landed on.

The lesson from Base is that rollups do not acquire users. Somebody else already acquired them, and the rollup rents access. Every chain that has tried to manufacture demand through incentives instead of renting an existing distribution channel has produced a spike followed by a plateau. I have watched this pattern since the EOS era, when I audited the token distribution mechanics of that launch and realized that the staking yield was not generating demand — it was generating the appearance of demand while the underlying float rotated between a small number of large holders.

That is the same mechanic running again, one layer of abstraction up.

| Metric | Top 3 rollups (combined) | Next 10 rollups | Remaining tracked chains | |---|---|---|---| | Share of filtered monthly active users | Dominant majority | Mid single-digit to low double-digit share | Low single-digit share in aggregate | | Share of bridged value | Majority | Substantial | Long tail, heavy double-counting | | Share of sequencer fee revenue | Majority | Meaningful but shrinking | Negligible per chain | | Median cost to launch | Not applicable | Not applicable | Config file plus a token |

The table is not a market-share scoreboard. It is a description of where liquidity actually lives, and the answer is that it lives in fewer places than the chain count suggests.

Sequencer Margin Is the Product, and It Is Compressing

Strip away the narrative layer and a rollup is a business with one line of revenue and three lines of cost.

Revenue is sequencer fees: what users pay to have their transactions ordered and included. Cost is L1 data posting, L1 proof or dispute-game verification, and operating overhead — provers, RPC infrastructure, indexing, block explorers, developer relations, and the salaries that come with all of it.

The operating equation is therefore brutally simple. Gross margin equals sequencer revenue minus the cost of posting data and settling on L1. Everything else is overhead that has to be covered by a token treasury, a venture balance sheet, or a foundation grant.

Before blobs, the equation was constrained from above: data cost scaled with usage, so a rollup could not achieve both scale and high margin. Blobs broke that constraint. Margin expanded. And then something predictable happened.

When every competitor's cost base collapses simultaneously, the collapse becomes a price war, not a profit pool. Rollups passed the savings through to users because the switching cost between two chains with similar applications is nothing more than a bridge transaction and a few minutes of patience. Fee competition became reflexive. The cheapest rollup set the price, and everyone else matched it or lost the flow.

I have seen the pass-through behavior before. In 2020, when I directed the cross-platform yield strategy across Compound and Aave, the spread we were capturing existed only because two protocols with different interest rate models were pricing the same asset differently. That spread was structurally temporary. As soon as enough capital moved to arbitrage it, both protocols converged on the same rate and the opportunity vanished. The window was six weeks. We planned for four.

The sequencer margin window is behaving the same way, but the convergence is happening at the fee line rather than at an interest rate. Transaction fees on the cheapest rollups have compressed to fractions of a cent for a simple transfer. That is wonderful for users. It is terrible for a business whose only revenue line is that fee. A rollup that has driven its fee to near zero has not won a market. It has eliminated its own income statement.

When margin compresses toward zero, four responses become available, and every rollup eventually picks one.

The first is subsidy. The token treasury pays the difference. This works exactly as long as the treasury lasts, and it produces a chain whose activity level is a direct function of its emissions schedule. The moment the emissions taper, the activity tapers with them, and the TVL chart looks like a ski slope.

The second is vertical integration. The rollup acquires or builds the applications that generate the flow, so the profit moves from the sequencer line to the application line. This is the strategy that Base pursued by existing inside a wallet, and it is the strategy that Uniswap pursued by building its own chain rather than paying someone else's sequencer.

The third is order flow capture — the sequencer sells priority, or captures MEV directly, or auctions ordering rights to a preferred party. This is real revenue but it is politically expensive, because the moment a chain monetizes ordering, the users who were promised fairness start asking questions about who is getting front-run.

The fourth is consolidation. The chain stops being a chain and becomes a feature inside someone else's ecosystem.

I expect the fourth outcome to dominate over the next eighteen months, and the market is not pricing that at all. It is pricing each rollup as a going concern with an independent future. Most of them are not going concerns. They are marketing budgets with a sequencer attached.

| Cost line | Pre-EIP-4844 | Post-EIP-4844 | Direction of pressure on margin | |---|---|---|---| | L1 data posting | Dominant, scales with usage | Collapsed, priced by shared blob market | Strongly favorable, then competed away | | L1 verification / proving | Material for ZK, modest for optimistic | Roughly unchanged, trending down with proving efficiency | Modestly favorable | | Infrastructure and overhead | Material | Rising with chain count and tooling complexity | Unfavorable | | Sequencer revenue | Stable per transaction | Compressed toward zero by competition | Strongly unfavorable | | Net | Thin margin, defensible by scale | Near-zero margin at the margin | Structurally unstable |

That last row is the one that matters. A business with near-zero unit margin and rising fixed costs is not a business. It is a bet on future repricing, and future repricing requires either a demand shock or a consolidation event.

Bridge Liquidity, Not Block Space, Is the Binding Constraint

Here is the misconception I meet most often in conversations with allocators. The assumption is that the scarce resource in the multi-chain world is throughput, and that chains compete on how much of it they can deliver.

Throughput is not scarce. Throughput has been abundant since the blob market opened. There is more unused block space across the rollup landscape right now than there is demand to fill it, and the gap widens with every new chain that launches.

The scarce resource is bridge liquidity.

Every rollup needs a float. It needs stablecoins sitting on its chain so that users can trade without waiting for a bridge round trip. It needs a native gas token balance. It needs enough depth in its largest pools that a routine swap does not incur double-digit slippage. That float is not created by the rollup. It is imported, at a cost, from a shared global pool.

Every new chain that launches does not create new liquidity. It slices an existing float into a smaller piece, and the slicing has a measurable cost.

The cost shows up in three places that rarely get combined into a single number.

The first is the routing tax. A cross-chain swap is not one transaction. It is a bridge transaction, a wait, a swap, and often a reverse bridge. Each leg carries a fee, a slippage component, and a latency cost. On a round trip between two mid-sized rollups, that combined tax routinely exceeds the entire economic benefit of being on the destination chain in the first place. Users do it because they have no alternative, not because it is efficient.

The second is the bridge risk premium. Liquidity providers who bridge capital to a new chain are taking on smart contract risk, validator risk, and message-passing risk. They price that risk. A chain with a short track record and an unaudited bridge pays for its float in the form of higher spreads and worse execution for its users, even if its transaction fees are nominally lower than everyone else's.

The third is the canonical bridge lockup. Capital sitting in a canonical bridge is capital that is not productive. It secures the chain's solvency but generates no yield, and at scale it represents a permanent drag on the ecosystem's capital efficiency.

I saw this same dynamic at a smaller scale in 2022, in the week following the collapse of TerraUSD. The interesting part of that event was not the death spiral itself — the mechanism was well understood by anyone who had read the code. The interesting part was the liquidity vacuum that followed the exit of a stablecoin float from dozens of pools simultaneously. Every venue that had a meaningful LUNA or UST balance saw its depth collapse within hours, and the protocols that had treated that float as permanently resident discovered that it had never been resident at all. It had been rented, and the rent came due.

Bridged liquidity is rented liquidity. Chains that treat it as a permanent asset base are marking a liability as equity.

This is where the fragmentation thesis really bites. A rollup with a small, rented float and near-zero sequencer margin is not a scaled-down version of a large rollup. It is a structurally different kind of asset, with different failure modes and a much shorter runway. The market capitalization of L2 tokens across the long tail implies an operating life and an independent future that the unit economics simply do not support.

The Modular Data Availability Trade Was Misread

The modularity thesis — separate execution, settlement, consensus, and data availability into independently priced markets — was the most intellectually satisfying narrative in crypto infrastructure for about two years.

The prediction was straightforward. Rollups would find Ethereum's data layer too expensive. They would outsource data availability to a specialized provider offering the same service at a fraction of the cost. Celestia, EigenDA, and Avail would capture the DA market, and Ethereum's rollups would become dependent on external data layers.

Blobs invalidated the middle step.

When Ethereum delivered cheap data availability natively, the cost advantage that the modular DA providers were built to sell largely evaporated for mainstream workloads. Posting to Ethereum became cheap enough that for most rollups, the savings from switching did not justify the additional trust assumption. The modular DA thesis was not wrong about the mechanism. It was wrong about the price at which Ethereum would offer the same product.

That is why the alt-DA market settled into niches rather than dominating. Validiums, gaming chains, and high-throughput appchains that generate more data than they could ever afford to post to L1 still need outsourced DA. Those are real markets. They are simply much smaller than the market that the modular thesis implicitly assumed.

The deeper lesson is about competitive response timing. Modular DA providers built a product on the assumption that Ethereum could not or would not adapt quickly. Ethereum adapted in a single hard fork. That is the risk of building against a monopoly that controls its own roadmap, and it is the same structural risk I wrote about during the 2017 token distribution wave, when a generation of projects assumed the exchange listing model was fixed and discovered within a year that exchanges could change the rules overnight.

Platform owners adapt. The asymmetry of that adaptation is the single most underpriced variable in infrastructure investing, and it is why I weight "how quickly can the base layer undercut this product" above almost any other question when I look at an infrastructure token.

DeFi teaches us that trust is code, not character. But it also teaches us that code lives on a platform, and the platform gets a vote.

Appchains Are a Business Model, Not a Scaling Solution

The cleanest framing I can offer is this: an appchain is not a chain that exists to scale an application. It is a chain that exists so the application can capture its own order flow.

When an application is deployed on a shared rollup, it pays the sequencer for ordering. It pays the base fee. It pays the priority fee when congestion matters. Those payments are the sequencer's revenue, and the application never sees them. When that same application deploys its own chain, the payments it was making to someone else become payments it makes to itself.

That is the entire economic argument for appchains, and it has nothing to do with throughput.

MEV capture is the real prize. An exchange that runs its own chain controls the ordering of its own order flow. A wallet that runs its own chain controls the ordering of its own users' transactions. A game that runs its own chain can decide which actions are priority and which are not. Each of those decisions has value, and on a shared chain that value accrues to a third party.

Once you see appchains as an order flow monetization strategy, the fragmented landscape stops looking irrational. It looks like dozens of applications each concluding that they would rather own their own sequencer margin than rent someone else's.

The problem is that this logic produces a chain for every large application, and the aggregate effect is exactly the one I described earlier: more slices, thinner float, higher routing tax, more friction for the user who does not care which chain he is on.

There is also a governance dimension that rarely gets discussed. Every appchain is a new set of validator incentives, a new treasury, a new governance forum, and a new set of decisions about who controls upgrades. The overhead is real, and it is borne by the users of that application in the form of fragmentation.

The market is currently pricing appchains as growth. It should be pricing them as margin capture at the ecosystem's expense.

Three Moats, and Only Three

When I strip out every chain with a non-existent moat, three categories survive.

Distribution. This is the Base category. A chain that sits inside a wallet with tens of millions of verified accounts does not need to acquire users, because the users arrived before the chain did. This is the strongest moat in the industry and it is nearly impossible to replicate. You cannot buy a consumer distribution channel of that size with a token, because the token only reaches people who already own tokens.

Liquidity depth. This is the Arbitrum category. Depth is self-reinforcing in a way that users are not. Deeper pools mean better execution, better execution attracts larger traders, larger traders deepen the pools. A chain that has accumulated depth is expensive to displace because displacing it requires someone to subsidize better execution than the incumbent, and that subsidy is a permanent cost, not a one-time expense.

Standards and licensing. This is the Optimism category. The OP Stack is not just a software product. It is a governance and standardization layer, and the chains built on it share a common upgrade path. That gives the originating ecosystem influence far beyond its own user base, because every dependent chain has to track its releases. It is a weaker moat than distribution or depth, but it is more durable than anything the long tail possesses.

Everything outside those three categories is competing on incentives, and incentives are the least defensible moat in finance. Any competitor can match an incentive. No competitor can match Coinbase's install base overnight, and no competitor can match a decade of accumulated liquidity depth in a quarter.

| Moat | Who holds it | Durability | How it fails | |---|---|---|---| | Distribution | Chains embedded in consumer wallets | Very high | Parent app loses relevance | | Liquidity depth | The largest rollups by pool depth | High | A subsidized venue buys depth with permanent losses | | Standards and licensing | Stack owners with dependent chains | Moderate | Dependent chains fork or migrate | | Incentives | Nearly everyone else | Very low | The next incentive program |

The Contrarian Angle: Intents Do Not Fix Fragmentation, They Move It Off-Chain

The dominant narrative right now is that the multi-chain world will be stitched together by intent-based architectures. The user will express a desired outcome, a network of solvers will compete to fulfill it, and the user will never need to know which chain the transaction settled on.

It is an elegant story. It is also, in its current form, a re-centralization of the exact function that decentralization was supposed to distribute.

Here is the mechanism, stripped of the marketing. In a traditional DEX, the user's transaction is broadcast and ordered by a block builder. The value extracted from ordering decisions is captured on-chain, transparently, by whichever searcher or builder wins the auction. It is extractive, but it is visible and auditable.

In an intent-based system, the user signs a declarative order and hands it to a solver network. Solvers compete off-chain to fulfill it, and the winning solver's execution is what settles on-chain. The extraction still happens. It is identical in kind. But the auction that determines it happens in a private mempool with a small number of participants, and the on-chain record shows only the final settlement.

Intent architectures do not remove MEV. They relocate it from on-chain block builders to off-chain solver networks, and they replace a transparent auction with an opaque one.

The concentration risk is the part the market misses entirely. Solver networks are permissionless in theory and oligopolistic in practice, because settling intents requires capital, inventory across multiple chains, and low-latency infrastructure. Those are the same three requirements that made block building concentrated after MEV-Boost. There is no reason to expect a different equilibrium in solver auctions, and there is every reason to expect the same outcome: a handful of well-capitalized operators routing the majority of flow and capturing the majority of the spread.

The practical consequences are worth stating plainly.

First, the user's execution quality becomes a function of a private auction the user cannot inspect. In an on-chain auction, a user who is sophisticated can at least reconstruct what happened. In a solver network, the user sees a filled order and a price. The reasoning behind that price is not on-chain and never will be.

Second, the router becomes the chokepoint. This is the part that should worry anyone who thinks intents are the path to decentralization. If a single solver network handles the majority of cross-chain flow, that network is functionally a centralized exchange with an on-chain settlement layer. It can censor, it can degrade, and it can be regulated the way centralized intermediaries are regulated. Which brings the whole thing full circle: the industry would have rebuilt the exact intermediary it spent a decade trying to eliminate, and it would have done so while describing the process as decentralized.

Third, the fragmentation problem is not solved. It is concealed. A user who submits an intent across four chains still pays for four chains worth of liquidity, still pays the routing tax, and still pays the solver's spread. The difference is that the cost has been bundled into a single number that the user accepts without asking how it was derived. Bundling is not efficiency. Bundling is the removal of the information a user needs to evaluate whether he is being overcharged.

Sentiment is the invisible ledger of value, and right now the sentiment surrounding intent architectures is writing a very large liability into that ledger.

Here is what I would watch as a falsification test. If intent-based routing is genuinely more efficient than direct on-chain execution, solver spreads should compress over time as the network grows, the way DEX spreads compressed as automated market maker competition intensified. If solver spreads instead stabilize at a level above what a transparent auction would produce, the model is not solving fragmentation. It is monetizing it.

The second thing to watch is solver concentration. If the top three solvers handle a share of volume comparable to what the top three builders handle in MEV-Boost auctions, the decentralization claim collapses, and the only remaining question is who regulates the routers.

The third thing to watch is whether the intent standard actually gets adopted across competing networks. A cross-chain intent standard only creates value if every major venue honors it. A standard honored by four chains out of eighty is not a standard. It is a loyalty program.

There is a broader blind spot in how the market is treating all of this. The interop narrative is being priced as though interoperability is a feature that can be shipped. It is not. Interoperability is a negotiation among parties with conflicting incentives. The chain that benefits most from interop is the smallest chain, because interop imports liquidity. The chain that benefits least is the largest chain, because interop exports its liquidity to competitors. The largest chains will therefore adopt interop standards slowly, partially, and in ways that preserve their depth advantage — and the standard's value to everyone else erodes with every delay.

The tension is impossible to resolve technically, which is why it keeps getting deferred politically.

There is a second contrarian observation that follows from the same logic, about identity and verification. The industry has spent three years discussing on-chain identity primitives, and the reason they have not shipped is not technical. A permanent, composable, publicly queryable registry of a person's creditworthiness or reputation is a liability for the individual and a treasure trove for anyone who wants to discriminate on the basis of it. Applications do not want the registry. They want the assertion — a yes or no, verifiable at the moment of use, and forgettable afterward. That is an off-chain attestation with an on-chain proof, not a token that lives on your wallet forever. The market has been pricing a primitive that its own users have a rational incentive to avoid.

Takeaway

The next twelve months are a consolidation trade, and the market is positioned the other way.

Watch three signals. First, the number of rollups with a sequencer revenue line that covers their L1 data cost without treasury subsidy — when that number starts falling rather than rising, the repricing begins. Second, solver concentration in intent auctions — if the top three solvers exceed half of routed volume, the decentralization premium in interop tokens is mispriced. Third, the migration rate of dependent chains off shared stacks — every chain that forks away from its licensing parent tells you the standards moat is weaker than the market believes.

The chains that survive will not be the ones with the best throughput, the cheapest fees, or the most elegant proof system. They will be the ones that controlled a distribution channel before they controlled a sequencer, or the ones that accumulated depth early enough that nobody can economically displace them. Everything else is a slice of a float that was never large enough to begin with.

The question worth asking is not how many chains the market can support. It is how many chains can survive a fee line that has gone to zero, a float that has been rented rather than owned, and a user base that was never theirs to begin with. Most of the landscape has the same answer to all three, and that answer is not the one currently reflected in price.

Speed is the only currency that never depreciates. Right now, the fastest trade in this market is patience.