The L2 Fee Floor Is a Price War: Blobs, Celestia, EigenDA

Ask most people how Ethereum rollups got cheap and you hear one word: blobs. The March 2024 Dencun upgrade gave rollups a dedicated data lane, and the average fee across the three largest Layer 2 networks fell by more than 99% over the next two years, from about $0.18 in the first quarter of 2024 to roughly $0.0015 in the first quarter of 2026, according to Token Terminal. That part of the story is real, and it is almost finished. Ethereum has now put a hard floor under its own blob prices, which means the next leg of fee compression is not Ethereum’s to hand out. It will be set in a market Ethereum does not control: the fight for rollup data between Ethereum blobs, Celestia and EigenDA.

The next fee cut is not Ethereum’s to give

For two years the cheapest way to make a rollup cheaper was to give it more blob space. Dencun did it in 2024, the Pectra upgrade widened the lane in 2025, and Fusaka added data-availability sampling in December 2025 so nodes no longer had to download every blob. Each step cut the price of posting data to Ethereum, and each step showed up almost immediately in L2 fees.

That lever is close to spent, for a reason most fee forecasts miss. Ethereum deliberately stopped its blob prices from falling to zero, and the mechanism is covered below. Further compression now has to come from one of two places. The first is execution, meaning cheaper computation on Ethereum itself, which is what the Glamsterdam upgrade is built to do and which we tracked separately as its testnet clock restarted. The second is data availability, or DA: the cost of publishing a rollup’s data so anyone can verify it. That second lever is not controlled by Ethereum’s roadmap at all. It is controlled by a price war, and that price war is where the 2026 to 2028 fee floor actually gets set.

Where an L2 fee actually goes

To see why DA matters so much, take an L2 transaction apart. A user pays for three things: the tiny cost of executing the transaction on the rollup, a small margin the sequencer keeps, and the cost of posting the transaction’s data back to a base layer so the rollup can be audited and reconstructed. Since Dencun moved that data into blobs, the DA component has dominated. Various rollup-economics writeups put it near 90% of an optimistic rollup’s operating cost, per Blockworks.

It was not always this cheap. Before Dencun, rollups posted their data as ordinary calldata on Ethereum, competing directly with every other transaction for block space, which is why an L2 swap could cost a dollar or more whenever mainnet got busy. Blobs gave rollups a separate, temporary lane priced by its own supply and demand, and the fee collapse followed within weeks. That is the mechanism worth holding onto: L2 fees are downstream of a data market, so they move when that market’s supply, demand, or price floor moves, not when Ethereum simply wills them lower.

The result is fees that now round to pennies. A simple transfer costs roughly two cents on Base, three cents on Optimism and four cents on Arbitrum One, based on the live cost trackers at L2BEAT. When the biggest line item in that bill is data availability, the question of how low L2 fees can go becomes a narrower one: how cheaply can a rollup buy data availability, and from whom.

Why Ethereum blobs stopped racing to zero

Before Fusaka, blob space was so abundant that the blob base fee sat at its one-wei minimum for long stretches. Posting data to Ethereum was, in practice, close to free. That was great for rollup users and terrible for ETH holders, because free blobs burn almost no ETH.

Fusaka changed the incentive with EIP-7918, which sets a floor under the blob base fee. The floor is pegged to Ethereum’s own execution cost: the blob base fee cannot fall below one sixteenth of the L1 execution base fee, per the EIP text. The effect was immediate and dramatic. The Ethereum researcher known as Kydo noted that the blob base fee jumped roughly 15,000,000 times from its old floor after the change took hold. Ethereum, in other words, chose value accrual over maximum cheapness. That decision is the single most important input to any honest fee forecast, because it means Ethereum has voluntarily left room underneath its own prices for someone else to undercut it.

The design was deliberate. Ethereum’s researchers wanted blob space to stay cheap enough to keep rollups home, but not so cheap that it gave away the network’s data business for nothing, so they tied the blob floor to the cost of ordinary Ethereum computation. When L1 is busy and expensive, the blob floor rises with it; when L1 is quiet, the floor falls. Blob fees can still be tiny, but they can no longer be nothing.

The DA challengers line up

Three networks are built to fill that room, each with a different pitch. Celestia is a standalone proof-of-stake chain that sells nothing but data space. EigenDA reuses restaked ETH plus a committee to disperse data at high throughput. Avail, spun out of Polygon, sells DA plus a unification layer meant to stitch rollups together. Ethereum blobs remain the incumbent and the safest option, but no longer the cheapest.

The table hides as much as it shows, because these options are not interchangeable. The price differences are real, but so are the security differences, and a rollup choosing a DA layer is trading one against the other. The next three sections take the two most important challengers in turn, then price the trade-off.

Celestia: blockspace sold at cost

Celestia’s entire design is a bet that data availability should be a commodity sold as cheaply as possible. It carries none of the overhead of a chain that also runs smart contracts and competes for its own blockspace, which is the structural reason it can undercut Ethereum, per Blockworks.

The network’s Matcha upgrade, which activated in late November 2025, pushed that bet further. Matcha raised the maximum block size to 128 MB from 8 MB and targets about 21.3 MB/s of throughput, a sixteenfold increase, on the way to a roadmap goal of 1 GB/s, per Celestia. It also cut the network’s inflation rate from 5% to 2.5%, reported CoinDesk. Celestia now commands roughly half of the external DA market and has processed more than 160 GB of rollup data, again per Blockworks.

Matcha also pointed Celestia outward. The upgrade removed a filter that had kept the network focused on its own token and set up cross-chain data flows, part of a roadmap toward what Celestia calls lazy bridging, so rollups settling on other chains can still buy Celestia data space. The strategic logic is scale: the more chains that can pay Celestia for data, the more the network can price each megabyte near its marginal cost and still cover its bills. That is good news for anyone forecasting cheaper fees and awkward news for anyone hoping TIA behaves like a claim on that growth.

Celestia’s co-founder Mustafa Al-Bassam has been blunt about the pricing philosophy. He has described the network’s blockspace as “basically free”, priced at the minimum needed to deter spam, and said the network is optimizing for growth rather than revenue. That is a gift to rollups and a problem for the token. TIA trades near $0.43, down about 98% from its February 2024 peak of $20.85, with a market value around $416 million, per CoinGecko. The cheaper Celestia makes data, the harder it is for TIA to capture the value it creates, a tension every DA-token holder now lives with.

EigenDA: restaked security at committee speed

EigenDA answers the security objection differently. Instead of a new validator set and a new trust base, it reuses Ethereum’s economic security by having operators post restaked ETH through EigenLayer, then disperses data through a committee rather than to every node. The pitch is ETH-aligned security at throughput Ethereum blobs cannot match.

The numbers are aggressive. EigenDA’s V2 release runs on mainnet at 100 MB/s, with more than 200 operators and over 40,000 restakers backing it, per EigenCloud. Its flagship customer is Mantle, which moved from its own MantleDA (ten operators) to EigenDA (more than 200 operators) and reported up to an 80% cut in data costs while lifting throughput to about 15 MB/s, per Mantle. For a rollup that wants to feel Ethereum-aligned without paying Ethereum blob prices, that is the whole sales pitch in one migration.

There is a catch that mirrors Celestia’s. Restaked ETH is not free security; it is borrowed security, and it carries its own failure modes. Operators can be slashed for misbehavior, but a Data Availability Committee still concentrates the job of actually holding and serving data in a limited set of hands, and the deeper restaking sits underneath a rollup, the more that rollup’s safety depends on EigenLayer’s slashing and incentive design working as intended under stress. EigenDA is now marketed as one product inside a broader EigenCloud stack of verifiable services, which widens the pitch but also means a rollup buying DA is increasingly buying into an ecosystem, not just a data pipe.

The 55x gap, and why it is not the whole story

The headline that drives this entire debate is a single comparison. When Ethereum L2s were paying $3.83 per megabyte to post data in blobs, the rollup Eclipse was paying Celestia about $0.07 for the same megabyte, roughly 55 times cheaper, which let Eclipse post more than 83 GB of data without draining its treasury, per Blockworks.

Two things keep that number from settling the argument. First, blob prices move. The gap widens when blobs are congested and narrows when they are quiet, and Ethereum keeps adding blob capacity through its Blob Parameter Only forks, which raised the per-block target and maximum in stages after Fusaka. More blob supply pushes Ethereum’s price back down toward the challengers. Second, the comparison quietly mixes security models, which is the part most fee charts leave out. Rollups already vote with their wallets on price: Blockworks notes that operators like Scroll and Taiko have simply delayed posting data when blob markets spiked. The open question is whether they will move permanently to a cheaper layer, or only visit one when Ethereum gets expensive.

The blob-supply side of that fight is already scheduled in pieces. The first two Blob Parameter Only forks after Fusaka raised the per-block target and maximum in steps, from six and nine blobs toward the low teens and low twenties, and each increment is a lever Ethereum can pull to push its own DA price back down without a full upgrade. Whether that is enough to hold rollups depends on how fast demand for blob space grows to fill the new supply, which is the same race every capacity expansion has run since Dencun.

The security tax nobody prices

Cheaper data availability is not free money; it is a different risk profile. Ethereum blobs inherit Ethereum’s full validator set and its data-availability sampling, so a rollup posting to blobs gets the strongest guarantee available that its data was actually published and can be recovered. Celestia offers its own proof-of-stake validator set with sampling, smaller than Ethereum’s but still a live, decentralized network. EigenDA leans on a Data Availability Committee backed by restaked ETH and slashing.

A committee is a trust assumption. If enough of it withholds data, a rollup can be stalled or, in the worst case, censored, and users may be unable to prove what they own. That is the same category of counterparty risk over who holds the data that has produced some of crypto’s most expensive failures, a pattern we traced through the repeat bridge hacks of the past two years. None of this makes external DA a bad deal. It makes it a priced deal, and the price is a security discount that does not show up in the per-megabyte number. A fee forecast that ignores it is really forecasting the fee of a slightly different, slightly weaker product.

How a rollup actually chooses its DA layer

For a rollup team, picking a DA layer is not a spreadsheet decision about price per megabyte. It is a bet on four things at once: cost, security, sovereignty, and switching risk.

Cost is the obvious axis, and the one the 55x headline captures. Security is the discount hiding behind that number, as the previous section laid out. Sovereignty is subtler: a rollup that posts to Ethereum blobs inherits Ethereum’s neutrality and its social consensus, while a rollup that leans on a committee is trusting a smaller group not to collude or disappear. Some teams want that Ethereum alignment as a trust anchor even when it costs more, which is exactly the demand EigenDA is built to capture with restaked ETH rather than a fresh token.

Then there is switching risk, which fee charts never show. Moving DA layers is not a config change. It touches the rollup’s proof system, its bridge contracts, and its assumptions about how data is recovered if the DA layer misbehaves. A validity-proof rollup and an optimistic rollup face different migration costs, and a chain that has already built its entire security story around Ethereum blobs cannot casually walk away without reopening questions its users thought were settled. That friction is why the Mantle move mattered so much: it was a large, established rollup showing that the migration is survivable, which lowers the perceived risk for everyone watching.

The cleanest way to read the market is as a spectrum rather than a binary. At one end sits a rollup posting full data to Ethereum blobs, the most expensive and most secure option. In the middle sit rollups using Celestia or EigenDA, cheaper with a different trust base. At the far end sit validiums, which keep data entirely off Ethereum and accept the strongest cost savings and the weakest guarantees. Every rollup is a point on that line, and the fee a user pays is mostly a function of where its team decided to stand.

What cheaper DA does to Ethereum’s value accrual

Here is the uncomfortable other side of the trade. Every megabyte a rollup posts to Celestia or EigenDA instead of Ethereum is a megabyte that does not pay Ethereum and does not burn ETH. The compression that helps users directly erodes the fee revenue that was supposed to make ETH a productive asset.

The scale is easy to underestimate. Fidelity Digital Assets calculated that Base paid only about $5.2 million in blob fees over a year against roughly $94 million in user revenue, and that EIP-7918, had it been live since the Deneb and Cancun upgrades, would have returned about $78.6 million (around 24,641 ETH) to holders, per Fidelity. Blockworks has called DA fees ETH’s last real source of value accrual. Yet the deflation story has already cracked: net ETH issuance turned modestly positive after Dencun, so ETH is no longer reliably deflationary, per CoinLedger. ETH trades near $2,718, up about 5% on the day, per CoinGecko, but the burn engine underneath that price is weaker than the 2021 narrative promised.

This is the crux of the ETH investment debate in 2026. One camp argues that Ethereum should not care where rollups buy their data, because a thriving rollup ecosystem drives demand for ETH as collateral, as gas, and as the asset staked to secure the whole system. The other camp argues that if Ethereum becomes a settlement layer rollups touch as rarely as possible, buying their data elsewhere, then the fee-and-burn story that justified ETH’s premium quietly hollows out. EIP-7918 was the protocol picking a side: it chose to defend some fee revenue at the cost of some cheapness. The DA price war is the market testing whether that defense holds.

Ethereum’s answer is to reclaim the activity it exported. Vitalik Buterin’s Lean Ethereum roadmap describes what he calls the “third major iteration” of the protocol, after the original launch and the Merge, including a redesign of ERC-20 balances that he argues could cut base-layer transaction fees by more than 10x. If L1 gets cheap enough, some activity comes home, and some DA demand comes with it. That is the strategic reason the value-accrual debate and the DA price war are the same story told from two ends. It also matters for anyone staking ETH for yield rather than for burn, a shift we covered in our solo-staking guide for a 4% world.

The MEV footnote: compression does not erase ordering cost

Even if data availability fell to zero tomorrow, users would still pay an invisible fee. Sequencers decide the order of transactions, and ordering has value: it can be sold, auctioned, or quietly extracted through sandwiching and other maximal-extractable-value strategies. Cheaper DA does nothing about that. It lowers the visible sticker price while leaving the hidden ordering cost intact, and on some high-volume rollups that hidden cost is now the larger of the two. We have written before about how smart accounts are becoming both a target and a shield for MEV, and the same logic caps how low a user’s true, all-in cost can go regardless of what happens to blob or Celestia prices.

Glamsterdam moves the L1, not the DA war

It is tempting to fold Ethereum’s next upgrade into the same forecast, but Glamsterdam runs on a separate track. Its headline changes are execution-side: enshrined proposer-builder separation (EIP-7732), block-level access lists for parallel execution (EIP-7928), and a set of gas repricings that includes cheaper simple transfers (EIP-2780) but also a higher cost to create new accounts (EIP-8037), so it is not a uniform discount. The upgrade is scheduled to fork the Sepolia testnet on 6 October 2026 and the Hoodi testnet on 27 October, with mainnet still pencilled into the fourth quarter and no activation block locked, per the Ethereum roadmap and reporting from The Cryptonomist.

Glamsterdam matters for fees. It lowers L1 execution cost and raises the gas limit, which helps rollups indirectly and helps Ethereum compete for activity directly. But it does not change the data-availability economics that set the L2 floor. A rollup that buys its DA from Celestia will see roughly the same DA bill the day after Glamsterdam as the day before. That is why the DA price war deserves its own forecast rather than being treated as a footnote to Ethereum’s roadmap.

Forecast: three roads for the L2 fee floor to 2028

Start from the trend. A 2026 study by Ambrosia and Mizrach found L2 median fees had already fallen about 95%, from roughly $0.05 to $0.0015 between early 2024 and early 2026, and projected that L2 median fees would drop below Solana’s by October 2026, with L2 throughput surpassing Solana’s around 2029, per its preprint. The direction is not in doubt. The destination depends on who wins the DA market. Three scenarios bracket the range.

In the first, Ethereum holds the DA market. Blob Parameter Only forks keep adding capacity, PeerDAS keeps the cost of that capacity low, and most rollups decide the security premium of Ethereum blobs is worth a fraction of a cent. Fees keep grinding toward the tenth-of-a-cent range, and Ethereum’s burn survives.

In the second, external DA wins meaningful share. Celestia’s move toward 1 GB/s and EigenDA’s 100 MB/s pull large, data-heavy rollups off blobs the way Mantle already left. Fees fall into hundredths of a cent, but Ethereum’s burn erodes and ETH leans harder on staking yield for its value.

In the third, and most likely, the market fragments. Rollups route by workload: cheap external DA for gaming and high-throughput consumer apps, Ethereum blobs for high-value financial settlement where the security premium is worth paying. The fee a user sees then depends less on Ethereum’s roadmap than on which rollup they happen to be using and what that rollup decided about data.

None of the three is a clean win. The fragmented outcome is the most likely precisely because rollups are not one market: a high-value settlement chain and a free-to-play game have genuinely different tolerances for the security discount, so they will rationally make different DA choices. For a user, that means the honest answer to how low fees go is a distribution, not a single number, and the shape of that distribution is set less by Ethereum’s roadmap than by how the DA price war resolves over the next two years.

What to watch through 2027

The forecast resolves through a handful of concrete signals rather than through price predictions:

• Whether a top-five rollup by value migrates its DA off Ethereum, following the Mantle-to-EigenDA precedent. One large defection would reprice the whole debate.
• The scheduling of the next Blob Parameter Only forks. BPO3 and BPO4 remain unscheduled, with an aspirational target near 128 blobs per block; more blob supply is Ethereum’s cheapest way to fight back on price.
• Celestia’s progress toward 1 GB/s, and any sign that TIA can capture value rather than only enabling it.
• EigenDA’s operator and restaker growth, and whether ETH-aligned DA becomes the default for new rollups.
• A DA-layer incident: a committee that withholds data, a reorg, or a recovery failure would instantly reprice the security tax that the per-megabyte charts ignore.
• Glamsterdam’s mainnet activation, which will lower L1 costs but leave the DA question open.

For the applications that live or die on the fee floor, from consumer payments to the on-chain credit markets that only work when moving collateral is nearly free, the answer to how low L2 fees can go is now mostly a question about data availability. Ethereum handed the next round of that fight to a market it does not run. Watching who wins it is the most useful thing a fee forecaster can do between now and 2028.

Frequently Asked Questions

What is data availability, and why does it set L2 fees?

Data availability is the cost of publishing a rollup’s transaction data so anyone can reconstruct and verify its state. After the EIP-4844 blob upgrade, data availability became the largest component of an L2 transaction fee, often estimated near 90% of an optimistic rollup’s operating cost, so the price a rollup pays for data availability largely sets the fee its users pay.

Is Celestia cheaper than Ethereum blobs?

Historically yes, by a wide margin. In one widely cited comparison, Ethereum L2s paid about $3.83 per megabyte for blobs while the rollup Eclipse paid Celestia about $0.07 for the same megabyte, roughly 55 times cheaper. That gap moves with blob congestion and does not account for the different security guarantees behind each option.

Will L2 fees ever reach zero?

No. EIP-7918 puts a floor under Ethereum blob fees at one sixteenth of the L1 execution base fee, and users still pay a small sequencer margin plus hidden MEV and ordering costs. Independent research projects L2 median fees falling to fractions of a cent by 2028, not to literally zero.

Does cheaper data availability hurt ETH?

It can. Data posted to Celestia or EigenDA does not pay Ethereum and does not burn ETH. Fidelity Digital Assets estimated that Base paid only about $5.2 million in blob fees against roughly $94 million in user revenue, and net ETH issuance turned modestly positive after Dencun, which weakened the deflation narrative.

What is the difference between EigenDA and Celestia?

Celestia is a standalone proof-of-stake network with its own validator set and token. EigenDA reuses Ethereum’s economic security through restaked ETH plus a Data Availability Committee, so it markets ETH-aligned security rather than a new trust base, and it reports 100 MB/s throughput on mainnet with more than 200 operators.

By Priya Reddy, HOGE Wire senior markets writer.