The market is still pricing blockchains as if execution speed is the scarce resource. That is the wrong bottleneck. Today’s tape, with ETH down 5.51% at $1,648.93 and SOL off 6.64% at $68.81, reinforces a familiar pattern: investors sell Layer-1 beta first and ask architecture questions later. But the more durable question is not whether a chain can process another thousand swaps per second. It is whether the data behind those swaps can be published cheaply, sampled by ordinary nodes, and recovered under stress.
The modular blockchain thesis is usually marketed as a clean separation of duties: execution on rollups, settlement on Ethereum or another base layer, and data availability on a specialized network. That summary is technically correct and strategically incomplete. Separating execution from data availability is not just an engineering optimization; it is a reallocation of power away from monolithic validators and toward whichever layer controls the cheapest credible guarantee that transaction data was actually made public.
Execution is where users feel performance. Data availability is where security quietly lives.
Monolithic Chains Hid the Real Cost Center
Bitcoin and early Ethereum bundled execution, consensus, settlement, and data availability into one replicated system. Every full node downloaded the same data, verified the same state transition rules, and accepted the same canonical history. That design is robust, but it is economically blunt: if every node must process every transaction, throughput is capped by the weakest acceptable machine in the validator set.
Solana pushed the opposite design philosophy: integrate aggressively, optimize hardware assumptions, and keep execution and data propagation inside one high-performance domain. That architecture has produced impressive user experience, especially for low-value high-frequency activity, but it also makes the validator network the shared bottleneck for computation, bandwidth, and state growth. In a monolithic chain, scaling one dimension tends to raise requirements across all dimensions.
Ethereum’s rollup-centric roadmap is the admission that global execution replication does not scale. Optimistic rollups such as Arbitrum and Optimism, and ZK rollups such as zkSync, Starknet, Scroll, and Polygon zkEVM, move execution off the Layer-1 while posting compressed transaction data or proofs back to Ethereum. The problem is that rollups still need their data somewhere. If users cannot reconstruct the rollup state from published data, fraud proofs and validity proofs become governance theater rather than security mechanisms.
Data Availability Is Not Storage, and That Distinction Matters
Data availability is often mistaken for decentralized storage. It is not Filecoin with shorter retention. The DA problem asks a narrower but more security-critical question: was the transaction data for a block published to enough independent participants at the time the block was accepted? Long-term storage is useful, but the immediate guarantee of public availability is what lets light clients, challengers, provers, and independent indexers reconstruct state and detect invalid transitions.
Ethereum’s EIP-4844, activated in the Dencun upgrade, made this distinction explicit through blobs. Each blob carries roughly 125 kB of data, with a target of three blobs per block and a maximum of six, giving Ethereum a target DA throughput near 375 kB per 12-second slot and a ceiling near 750 kB. Blobs are not permanent calldata; they are temporary data objects priced in a separate fee market and retained long enough for rollup security assumptions to work.
This was a material architectural shift. Before blobs, rollups competed with ordinary Ethereum transactions for calldata, making L2 fees hypersensitive to mainnet congestion. After EIP-4844, rollups received a dedicated DA lane. That is why average transaction fees on major Ethereum L2s collapsed after Dencun, in many cases from tens of cents to sub-cent levels for simple transfers, even while Ethereum L1 execution remained expensive during bursts.
The contrarian point is that EIP-4844 did not finish Ethereum scaling. It put a price tag on the scarce resource. Blob space is now the commodity rollups actually consume, and Ethereum’s ability to expand blob capacity without compromising decentralization will determine whether it remains the default settlement and DA hub or becomes one expensive security option among several.
The New DA Market: Ethereum, Celestia, EigenDA, Avail, and Near
Celestia made the purest bet on modularity by launching a blockchain focused on data availability and consensus, not general-purpose execution. Its core mechanism is data availability sampling: light nodes sample small random pieces of erasure-coded block data to gain statistical confidence that the full block was published. This allows the network to increase block size without forcing every participant to download every byte.
Avail, originally incubated inside Polygon, is pursuing a similar DA-specialist lane using validity commitments and sampling. EigenDA takes a different route by building DA on top of Ethereum restaking through EigenLayer, outsourcing cryptoeconomic weight to restaked ETH and liquid staking tokens. Near DA competes on cost and integration simplicity, positioning its sharded architecture as a cheap publication layer for rollups and appchains.
These systems are not interchangeable. Ethereum blobs inherit Ethereum’s validator set and social legitimacy but are capacity constrained and potentially expensive during demand spikes. Celestia offers purpose-built DA and scalable sampling but asks rollups to trust a separate validator set and bridge assumptions. EigenDA may offer massive throughput, but restaking introduces correlated slashing, operator concentration, and governance complexity that the market still underprices. Near DA is pragmatic and cheap, but cheap DA without a security premium can become infrastructure arbitrage rather than credible neutrality.
- Ethereum blobs: strongest settlement adjacency, limited capacity, premium security pricing.
- Celestia: clean modular architecture, independent security domain, strong sampling design.
- EigenDA: high-throughput promise, Ethereum-aligned economics, restaking contagion risk.
- Avail and Near DA: cost-competitive alternatives, still proving ecosystem gravity and trust assumptions.
Rollups Are Becoming Customers, Not Kingdoms
The most important business model shift is that rollups are turning into DA customers. A rollup’s margin is the spread between what it charges users for transactions and what it pays for proving, sequencing, settlement, and data publication. As DA markets mature, rollups will route data the way cloud companies route workloads: by cost, latency, reliability, jurisdictional exposure, and integration risk.
This weakens the simplistic bull case for every L2 token. If execution becomes easy to fork, sequencers become competitive, and DA becomes a commodity input, then many rollups are not sovereign networks; they are application servers with bridge risk. The value accrues either to distribution, liquidity, and developer mindshare, or to the hard-to-replicate security layer underneath. Most mid-tier general-purpose rollups have neither.
There is already evidence of this compression. The OP Stack, Arbitrum Orbit, Polygon CDK, zkSync’s ZK Stack, and Starknet appchain frameworks all make it easier to launch customized execution environments. That is good for experimentation and bad for scarcity. When every exchange, game, payments app, and DeFi venue can launch an app-specific rollup, the scarce input is not another execution environment. It is credible DA plus credible interoperability.
This is why the “modular stack” should not be analyzed as a technical diagram. It should be analyzed as an industrial supply chain. Execution layers compete on developer tools and latency. Proving systems compete on cost curves and hardware acceleration. Settlement layers compete on finality and asset depth. DA layers compete on bandwidth per unit of security. The winners will be the providers that make one component reliable enough to be invisible.
The Security Trap: Cheap Data Can Be Expensive Later
Separating execution from data availability creates clean interfaces, but it also creates new failure modes. If a rollup posts data to an external DA layer and settles to Ethereum, users now depend on at least two consensus domains plus a bridge or verification mechanism between them. A failure in DA sampling, validator censorship, bridge logic, or light-client verification can strand state even if the execution engine is correct.
Validium designs make this trade-off explicit. They use validity proofs for state transitions but keep transaction data off Ethereum, reducing costs sharply while weakening data recovery guarantees. That can be perfectly rational for gaming, social, and enterprise use cases where transaction values are small and user experience matters more than maximum trust minimization. It is much less appropriate for high-value DeFi collateral, perpetuals, and cross-margin systems where unavailable data can turn solvency into a black box.
The market will eventually segment rollups by DA quality. A rollup using Ethereum blobs for core DeFi should trade at a lower trust discount than one using a small external committee. A consumer appchain using Celestia or Near DA may be economically superior if its users will not pay Ethereum-grade security rent. The mistake is pretending these are equivalent because both are called Layer-2s.
Regulators and institutional allocators will care about this distinction before retail users do. If a tokenized Treasury platform, exchange collateral system, or bank settlement pilot uses a rollup, the diligence question will be simple: where is the data, who guarantees availability, and what happens if that layer halts? The answer cannot be hidden behind throughput claims.
What to Watch Next
The next phase of modular blockchain infrastructure will be measured less by transactions per second and more by data throughput, prover costs, and cross-domain failure isolation. Ethereum’s roadmap points toward more blob capacity through PeerDAS and eventually more sophisticated sampling. Celestia must prove it can scale block sizes while maintaining a broad sampling network and economic security. EigenDA must prove restaked security does not become a leverage machine that works until the first correlated failure.
For investors and builders, three metrics matter more than headline TPS. First, DA cost per megabyte under congestion, because rollup margins are most fragile during volatility. Second, the number and diversity of independent node operators sampling or attesting to data, because paper throughput without decentralization is just a database. Third, integration depth with liquidity venues, wallets, bridges, and settlement contracts, because the best DA layer still needs demand from execution environments.
The modular thesis is right, but the consensus interpretation is too optimistic. Splitting execution from data availability does not magically decentralize blockchains; it exposes which guarantees were being cross-subsidized inside monolithic systems. That exposure is healthy. It lets the market price security, latency, and bandwidth separately.
My base case is not a single modular winner. Ethereum remains the premium DA and settlement layer for high-value financial activity. Celestia, EigenDA, Avail, and Near DA compete for cost-sensitive and application-specific rollups. Solana continues to prove that integrated design can dominate certain retail and trading use cases. The real losers are generic chains with no execution advantage, no DA advantage, and no liquidity moat.
The modular blockchain era will not be won by the chain with the loudest TPS chart. It will be won by the infrastructure layer that makes data availability cheap enough for mass adoption and credible enough that sophisticated capital does not demand a discount. That is a narrower, harder, and more valuable problem than most crypto narratives admit.