If data availability makes up 95% of a rollup's cost, does that mean the execution speed and settlement efficiency everyone's historically emphasized matter relatively less?
The key here is that "share of cost" and "importance" are two different things, and they shouldn't be directly equated. Execution speed and settlement efficiency determine user experience — how long you wait for confirmation, whether an app can support complex logic — and these dimensions affect what users actually feel very directly; it just so happens that, in absolute dollar terms, these components make up a far smaller share of the cost than data availability does.
The more precise way to understand it: data availability accounts for such a large share precisely because it deals with the resource-consumption nature of publishing a large volume of transaction data itself — that data volume grows roughly linearly with transaction count, whereas execution computation costs are relatively easier to compress through various optimization techniques. That makes data availability the key factor determining the floor price of a transaction's fee, while execution speed determines how good the experience is on top of that floor. The two are actually complementary rather than mutually exclusive — a Rollup that only optimizes execution speed while ignoring data availability costs could still leave users paying stubbornly high fees.
EigenDA markets itself on "inheriting EigenLayer's existing security" — how does that security inheritance actually work, and how does it differ from building a brand-new security mechanism from scratch?
EigenDA's security model rests on EigenLayer's Restaking mechanism: a group of operators reuse ETH (or liquid Staking tokens) already staked on Ethereum, through EigenLayer, to provide security for EigenDA — and if these operators misbehave or make mistakes in the data availability service, they face the risk of having that staked capital slashed. That means EigenDA's credibility comes directly from this pool of operators who've already put real capital on the line, rather than needing to recruit an entirely new set of validators and re-accumulate economic security scale from nothing.
If a rollup instead chose to build an entirely new data availability verification mechanism from scratch, it would essentially need to convince the market from zero that this new mechanism is safe enough — requiring time to accumulate sufficient validator participation and staked capital before reaching a trust level comparable to an already-mature system. EigenDA's value proposition is precisely bypassing that long "build trust from zero" process, directly borrowing a security foundation that's already operating in the market and already backed by real capital. The cost is that this Rollup's data availability security is now tied to the overall health of the EigenLayer ecosystem — if EigenLayer itself runs into a serious problem, rollups depending on EigenDA get pulled into that as well.
Avail chose to build an interoperability layer and a security network layer simultaneously instead of focusing on making data availability alone as good as possible — that sounds like a higher-risk strategy. Why would a team choose that path?
The logic behind this strategy rests on a judgment about where modular architecture is ultimately headed: if data availability eventually becomes a relatively standardized, easily substitutable commodity service (similar to storage space in cloud computing, where different vendors' core functionality increasingly converges), then simply making data availability as good as possible might, over the long run, struggle to build a truly hard-to-replace competitive advantage, since a competitor could catch up with similar technology at any time.
The vertical-integration bet is that bundling multiple modular components together can create overall value that's harder to replicate than any single component alone. If a rollup uses Avail's data availability, Nexus's interoperability, and Fusion's security network together, and these three are specifically designed to work tightly with each other, the difficulty of swapping out any single component increases — that "lock-in effect" is exactly the moat vertical integration is trying to build. The risk in this strategy is that if any one module (Nexus or Fusion, for instance) develops slower than expected, it could actually drag down the advantage already established in the core data availability business. This is a strategic bet between "go deep on one thing" and "bundle multiple things together," and neither approach has been definitively settled yet.
As a developer deciding which data availability layer to choose for my own rollup, what factors should I prioritize in practice?
The first thing to consider is which existing ecosystem your Rollup is most tightly tied to. If your application is already deeply integrated into the EigenLayer ecosystem, choosing EigenDA lets you directly reuse that existing trust foundation, saving the time cost of building trust from scratch. If you're using a general-purpose rollup framework (Arbitrum Orbit, OP Stack, Polygon CDK), Celestia currently has the most mature and broad integration across these frameworks, and choosing it generally means lower integration friction.
The second thing to consider is your own preference between "do one thing as well as possible" versus "one-stop modular service." If you only need a straightforward data availability service and don't want to be locked into additional architectural choices, a single-function provider like Celestia or EigenDA might fit better. If you'd rather eventually adopt interoperability, security, and other modular services all at once, and can accept the higher migration cost that comes with these services being tightly bundled together, Avail's vertical-integration path is worth evaluating. Third, it's worth keeping in mind that this entire category is still evolving rapidly, and current market-share rankings don't represent the final landscape — when choosing a data availability layer, it's also worth factoring in how difficult it would be to migrate to a different provider if the one you chose doesn't develop as expected.
When most people discuss rollup performance, they focus on how fast ordering is or how reliable settlement is — few pay attention to a hidden layer that makes up the overwhelming majority of a Rollup's cost: data availability (DA). According to figures disclosed by Celestia, data availability makes up roughly 95% of a rollup's total cost — meaning which DA layer a rollup chooses to publish its transaction data to actually determines how much users ultimately pay in fees more directly than which execution environment it chooses. In 2026, this infrastructure layer, previously hidden behind the scenes, has become the most fiercely contested battleground among Celestia, EigenDA, and Avail — a fight that determines not just market share, but which rollups ultimately survive and how low transaction costs can go.
Data availability answers a question that sounds simple but is actually the fundamental scaling bottleneck of blockchains: has the transaction data behind a Block genuinely been published, and can anyone download it? If a Node receives a new block but can't retrieve the corresponding transaction data, the entire chain can effectively stall — which is also why a rollup can't simply discard its transaction data, even though that data doesn't need to be executed, only stored and made public so anyone can later verify the rollup's state is genuinely correct. The traditional approach was publishing this data directly onto Ethereum mainnet, but that cost scales roughly linearly with transaction volume, making data availability the first ceiling a rollup hits as it tries to scale.
Celestia was the first production-deployed modular data availability chain, currently holding roughly 50% market share and having processed over 160GB of rollup data cumulatively. Its core technology is Data Availability Sampling (DAS): light nodes don't need to download an entire block — they only need to randomly sample a small portion of a block's data to probabilistically confirm the entire block's data is genuinely available. This design lets Celestia keep scaling throughput as more light nodes join, rather than depending on a small number of high-spec full nodes. Paired with Namespaced Merkle Trees, Celestia also lets each rollup download only its own data without needing to process other rollups' data volume — which is also why it's called a "modular" data availability layer: execution, settlement, and data availability are split into independent layers that can each scale on their own. Celestia's planned Matcha upgrade in Q1 2026 will double the block-size ceiling to 128MB, and its experimental Fibre Blockspace protocol has stated a long-term throughput target of 1 terabit per second.
EigenDA, built by EigenLabs on top of EigenLayer, follows a "Data Availability Committee" model: a group of operators staked on EigenLayer collectively provide the data availability guarantee, currently reaching roughly 100MB/s throughput. EigenDA's biggest differentiator is that it can directly inherit EigenLayer's already-established Staking security — for a rollup already embedded in the EigenLayer ecosystem, choosing EigenDA means not having to establish an entirely new trust assumption from scratch, and instead directly reusing economic security that already exists. That's also why EigenDA is especially appealing to rollups already tightly bound to the Ethereum ecosystem. Avail has taken a longer-term vertical integration strategy instead — its current data availability layer is only the first piece of a three-part architecture; the team is simultaneously building an interoperability layer called Nexus and a security network layer called Fusion, aiming to offer a complete full-stack modular infrastructure rather than a single-function DA service alone, and has already established integrations with mainstream rollup frameworks including Arbitrum, Optimism, and Polygon.
This competition currently shows a "winner-takes-most, but with room left over" pattern: Celestia maintains its market-share lead thanks to first-mover advantage and broad integration with general-purpose rollup frameworks (Arbitrum Orbit, OP Stack, Polygon CDK), while EigenDA has carved out a stronghold through its natural tie to the EigenLayer ecosystem, and Avail through its long-term vertical-integration bet — each holding their own niche. This war is still some distance from being genuinely settled.
If you're deciding which chain to build on or hold assets on long-term, checking which DA layer a rollup relies on tells you more about its actual cost structure and risk profile than its advertised TPS does — since DA makes up the vast majority of a rollup's cost, a rollup switching DA providers, or its chosen provider running into trouble, can directly and materially shift your transaction fees. If a rollup you're using ties its security to a specific DA layer's economic model — EigenDA's shared staking security, for instance — that dependency is worth understanding the same way you'd evaluate a Sequencer's centralization: it's another trust assumption sitting quietly underneath the chain, one that generally doesn't show up in the numbers a project chooses to advertise.