The L2 Fee War Is Over: Who Keeps the Sequencer Margin?

On 22 October 2026, Ethereum developers pulled the Glamsterdam upgrade off the Hoodi test network just four days before it was due to go live. The security firm ChainSecurity had flagged 13 real contracts that a pair of gas-repricing changes could quietly break, and core developers chose to stop and investigate rather than ship on schedule, as CryptoSlate reported. For anyone forecasting Layer 2 fees, the message was blunt: the next cut to what you pay for a rollup transaction just slipped again, and it may not land in 2026 at all.

That matters less than it once did. Ethereum rollup fees have already fallen by more than 99% since early 2024, from roughly eighteen cents on the top three rollups to a fraction of a cent, according to Token Terminal. A simple transfer on Base now costs about two cents. The race to zero on the number a user sees is, for practical purposes, already won. So the interesting forecast is no longer how low the fee can go. It is who keeps the difference between what you pay and what it costs to serve you. That difference is the sequencer margin, and in 2026 it quietly became the most valuable real estate in the rollup business.

ETH trades near $2,295, with a market value around $280 billion, per CoinGecko. The token has been dragged into this debate because the same compression that made L2s cheap also cut what they pay Ethereum, and therefore what Ethereum burns. This piece maps the margin: how an L2 fee breaks down, who earns what today, how Base, Arbitrum and the shared-sequencer newcomers are fighting over the spread, and three scenarios for where it lands by 2028.

The L2 fee has two halves, and only one is falling fast

A rollup transaction fee is not one thing. It is the sum of three parts: the cost the rollup pays Ethereum to post its data and settle (data availability through blobs, plus some L1 verification), the cost of executing the transaction on the L2 itself, and a markup the sequencer charges on top, which is its gross margin. On optimistic rollups, the data-availability leg has historically been the bulk of operating cost, something close to 90% by one widely cited Blockworks analysis. Dencun in March 2024, and then Fusaka in December 2025, attacked that leg directly with blobs and data-availability sampling. That is why the user-facing number collapsed.

The three legs do not move together. Posting cost fell toward a floor; execution cost is the next target, and that is what the Glamsterdam upgrade is for; and the margin is a business decision, not a protocol constant. When people say L2 fees are going to zero, they mean the first two legs. The third does not have to follow, and the operators with pricing power are making sure it does not.

The 99% is real, and most of it is already spent

The headline numbers are not in dispute. Token Terminal puts the average fee across the top three L2s at $0.180219 in the first quarter of 2024 and $0.001512 in the first quarter of 2026, a drop of 99.16%. Current per-transfer costs, as tracked by L2BEAT, sit around two cents on Base, three on Optimism and four on Arbitrum One. A 2026 academic study by Ambrosia and Mizrach measured a comparable 95% decline (from about $0.05 to $0.0015) and forecast that the leading rollups would slip below Solana’s median fee by October 2026, a window that has now arrived (arXiv).

The reason the easy gains are spent is a single line of code shipped with Fusaka. EIP-7918 put a hard floor under the blob base fee so it can no longer sit at effectively nothing; the floor is pegged at one sixteenth of the L1 execution base fee (eips.ethereum.org). For most of 2024 and 2025 the blob base fee was quoted in billionths of a gwei; after Fusaka it jumped by many orders of magnitude once the floor bound. Cheaper blobs are no longer the lever. From here, further cuts to the user fee have to come from execution efficiency, better data compression, or thinner margins, which is a slower and harder fight.

Why the next cut keeps slipping: the Glamsterdam gas fight

Glamsterdam is the upgrade built to attack the execution leg. It reprices gas so that standard transfers between existing accounts get cheaper (EIP-2780 cuts a basic transfer by up to 71%, according to Ethereum’s own roadmap), while shifting cost onto operations that create new state. It also lifts the gas limit, with clients testing at a 150 million reference and a 200 million target against mainnet’s roughly 60 million, and it reshapes block building with enshrined proposer-builder separation.

The trouble is in the repricing. Four days before Glamsterdam was due on the Hoodi testnet on 26 October, ChainSecurity reported 13 real-world contracts that two of the gas changes, EIP-8037 (a separate gas reserve for creating state) and EIP-8038 (new costs for touching existing state), could silently break. The flaw is subtle: today, running out of gas reverts a transaction cleanly, but the changes could make gas available again mid-execution, letting a previously failing step complete and then be recorded as permanently failed, which opens a fresh attack surface.

Christine Kim, founder of Protocol Watch and a longtime chronicler of Ethereum’s core developer calls, first disclosed the cancellation. The Ethereum Foundation said participants had “agreed to investigate how a proposed change to Ethereum’s gas-accounting system could affect applications already operating on the blockchain.” Glamsterdam did activate cleanly on the Sepolia testnet on 6 October, running blocks near the 200 million gas limit, but the Hoodi slip has no replacement date, and the three mainnet dates developers had floated (1, 8 and 9 December) now look tight. For a fee forecaster, the lesson is that the execution cut is real but hard, contested and not guaranteed this year. Which is exactly why the margin, not the fee level, is the live question.

The margin is where the money is

Base is the clearest case. Coinbase runs the Base sequencer, collects the fees, and discloses the revenue to the U.S. Securities and Exchange Commission as part of its reporting. Independent estimates from VaaSBlock put Base sequencer revenue at roughly $60 million to $70 million in the first half of 2026. Over a comparable prior year, Fidelity Digital Assets noted Base paid Ethereum only about $5.2 million in blob fees against roughly $94 million in user-fee revenue. That gap, two cents a transaction multiplied across hundreds of millions of transactions, minus a near-floored data bill, is the margin.

Arbitrum earns an estimated $35 million to $45 million over the same half, routed to its DAO treasury rather than to a company income statement. Optimism’s take is more modest, and zkSync and other validity rollups run thinner because generating proofs costs money. The pattern is the same everywhere: the cost side is near a floor, so whatever the chain can charge above it is almost pure margin.

The unit economics are easy to picture. A rollup might charge a user a cent or two for a transaction while its own marginal cost, the slice of a blob plus a settlement fee amortized across thousands of transactions in the same batch, is a fraction of that. Multiply the spread by the volume, hundreds of millions of transactions a quarter on the busiest chains, and a business that looks trivial per transaction becomes a line worth tens of millions a year. The batching math improves as a chain gets busier: more transactions share the same fixed posting cost, so the margin per transaction widens with scale. That is why the leaders defend their user counts so fiercely; volume is not just growth, it is gross margin.

The key fact for token holders: almost none of this reaches them as cash. Arbitrum’s revenue funds a treasury that holders govern but do not receive; Base has no token at all. The margin is real, but the claim on it is contested, which is the same real-yield-versus-buyback argument now running through the rest of crypto, from perp DEXs to app-chains, as we traced in our look at Hyperliquid’s real yield debate.

Arbitrum’s pivot: from Timeboost to Priority Gas Auctions

On 24 September 2026, Arbitrum phased out Timeboost, its sealed-bid express lane, and switched to Priority Gas Auctions, a per-transaction priority-fee market, alongside a paid data product called Fast Feed. Both route 97% of proceeds to the DAO treasury and 3% to the Arbitrum Developer Guild, a split approved through a constitutional on-chain vote in June 2026, Crypto Briefing reported. The change turns transaction ordering into a metered product: pay a higher tip, get sequenced first.

Timeboost showed both the promise and the problem. It produced real revenue, but an academic study of 48.5 million express-lane transactions and 494,000 auctions between January and April 2026 found that three entities won 99.74% of auctions and that roughly 30% of boosted transactions reverted, evidence of concentration and spam rather than healthy competition (arXiv). Ordering is a margin source, but a messy one. Arbitrum’s answer was to price it per transaction and bank the proceeds for the DAO, making the margin more legible even if it stays concentrated.

Enterprise flow is the other half of the thesis. Robinhood Chain, live since 1 July 2026 on Arbitrum’s Orbit stack and serving more than 120 countries, routes 10% of its net revenue back to Arbitrum (8% to the DAO, 2% to the developer guild). Offchain Labs co-founder Steven Goldfeder put the strategy plainly to crypto.news: “as enterprise adoption accelerates, Arbitrum is ready to capture revenue.” The forecast implied there is not cheaper fees; it is more fee-paying flow, with the chain keeping a slice.

Base’s model: no token, all margin to Coinbase

Base is the purest expression of the margin thesis, because there is no token to share with and no DAO to fund. Coinbase operates the sequencer, collects the fees and books the result; the company has described Base as a growing contributor to its results in disclosures to the SEC. For a U.S.-listed company, that is an unusually clean story: a blockchain as a product line with a gross margin, reported quarterly.

It also imports the risks of a listed company. Base revenue is cyclical, rising and falling with on-chain activity, and critics note that scale alone has not solved the cyclicality problem. Because Base has no token, its users have no governance lever and no revenue claim; they get cheap transactions, and Coinbase keeps the margin. That is a defensible trade for a consumer chain, and it is also why Base can afford to keep its fees among the lowest in the market: the margin does not have to feed a token economy, only a business. For most people, the friction that remains is not the fee but the on-ramp and the wallet, a problem we have argued is sometimes worth keeping.

The invisible half of the margin: MEV and ordering

The visible fee is only part of what a transaction costs. The other part is the value that can be extracted by controlling the order in which transactions are included, known as maximal extractable value, or MEV. On Ethereum mainnet, cumulative MEV has run into the billions of dollars, with millions extracted on a typical day and the large majority of blocks built through the MEV-Boost market.

The textbook case is the sandwich: a bot spots your pending swap, buys the same asset just ahead of you, lets your trade push the price up, then sells into it and pockets the difference. On mainnet, a competitive market of builders and searchers bids for the right to arrange those trades, and some of that value flows back to validators and, increasingly, to users through rebates. On a rollup with one sequencer, there is no such auction; the operator can internalize the ordering value or sell access to it, as Arbitrum now does explicitly. For the user, the posted fee understates the true cost of transacting, and no blob upgrade touches that gap.

On an L2, a single sequencer usually decides order, so it captures that value directly, without the competitive builder auction mainnet developed. As the visible fee approaches zero, ordering revenue becomes the more durable half of the margin, because it does not fall with blob prices; it scales with how much value flows through the chain. Arbitrum’s Priority Gas Auctions are simply the explicit, on-chain version of this. The forecast consequence matters: even if the user-facing fee drops to a tenth of a cent, the all-in cost of transacting, counting what is quietly extracted around your trade, does not compress at the same rate. Glamsterdam’s enshrined proposer-builder separation aims to make ordering fairer at L1, but on an L2 the sequencer still holds the pen.

Three forces squeezing the margin

The margin is not safe forever. Three forces are pushing against it.

• The data-availability price war. Rollups can buy data space from Ethereum blobs or from external layers such as Celestia and EigenDA, where posted cost per megabyte has run far below Ethereum’s. Cheaper data cuts the operator’s cost, which can widen the margin, but competition for users tends to pass the saving on. It cuts both ways.
• Commoditized sequencing. Shared sequencers like Espresso, backed by venture funding that has included a16z, and based rollups that borrow Ethereum’s own validators, aim to turn sequencing into a neutral utility. Taiko already runs based preconfirmations on its mainnet. If ordering becomes a commodity, the margin on it shrinks.
• Competition for users. With dozens of rollups live and a handful holding most of the value, keeping users means keeping fees low, which caps how much margin any one chain can defend.

Ethereum Foundation researcher Justin Drake frames the based-sequencing design as one where “a L1 proposer may become a preconfer by opting in to two preconf slashing conditions,” a model laid out on ethresear.ch. The counter-example is cautionary: Astria, a shared-sequencer startup, shut down in 2025, a reminder that commoditizing someone else’s margin is easier to pitch than to fund.

The value-accrual bill: who pays Ethereum?

The same margin that makes L2s good businesses is what makes ETH holders nervous. When a rollup posts to cheap blobs, it pays Ethereum very little, and Ethereum burns very little, so the ultrasound-money story weakens. Fidelity Digital Assets estimated that had EIP-7918’s floor been in force since the Dencun era, it would have added roughly $78.6 million (about 24,641 ETH) to what flowed to ETH holders, with around $30.6 million a year coming from Base alone. ETH is no longer reliably deflationary.

There is a second-order twist. If rollups migrate their data to cheaper external layers such as Celestia or EigenDA, they pay Ethereum even less, and the burn shrinks further; if the Glamsterdam execution cut eventually lands and draws more activity onto cheaper L2 blockspace, volumes could rise enough to offset a lower per-unit burn. The direction of ETH value accrual therefore depends less on the headline fee and more on where rollups buy their data and how much total flow they generate. It is a reminder that the fee forecast and the ETH forecast are related, but not the same question.

This is the same who-pays-the-network problem Bitcoin faces from the other direction, as its block subsidy halves toward zero and fee revenue has to fill the gap, a tension we traced through Runes and Bitcoin’s security budget. For Ethereum the bill is paid in burned ETH, and shrinking L2 margins on the data side are the reason the bill got smaller. It is also why ETH’s investment case has leaned harder on spot ETFs and staking than on fee burn this cycle, a shift visible across the altcoin ETF wave. The framing increasingly common among analysts is that data-availability fees are now one of the few channels left through which L2 activity accrues value to ETH, and that EIP-7918 exists precisely to keep that channel from drying up.

Sequencing is turning into a commodity business

Strip away the jargon and a sequencer is a commodity processor: it sells blockspace and ordering, pays for data and settlement, and keeps the difference, much like a Bitcoin miner selling hashes against the cost of power. We have written about how that kind of margin behaves when the commodity is abundant and the price of the output falls, in the context of Bitcoin mining margins. Commodity businesses converge on the cost of capital over time, once the underlying technology stops being scarce.

Three things are pushing rollup sequencing toward that fate: rollup-as-a-service providers such as Caldera, Conduit and AltLayer that let anyone launch a chain with an off-the-shelf sequencer; shared sequencing and based preconfirmations that unbundle ordering from the rollup; and a proliferation of app-specific chains that each capture only a sliver of flow. The operators that keep a fat margin will be the ones with something scarce attached: a consumer brand and distribution (Base, through Coinbase), a deep DeFi economy and enterprise deals (Arbitrum), or eventually a token with a credible claim on revenue, which remains largely aspirational. The chains selling undifferentiated blockspace will watch their margin compress toward the cost of posting data, which is itself near a floor. That is the real fee-compression story of 2027 and 2028: not the price on the screen, but the spread behind it.

The forecast: three scenarios to 2028

Here is how the margin could play out over the next two years. These are directional scenarios, not price targets, and the drivers are the Glamsterdam execution repricing, the EIP-7918 blob floor, the data-availability price war and the pace of sequencer decentralization.

My base case is that the visible fee keeps drifting down slowly while the margin holds, because the two biggest operators have moats, distribution and a deep DeFi economy, that undifferentiated competitors do not. A user-facing number below a cent is likely; a collapse in the margin is not, at least not by 2028. The academic forecasts that have L2 fees beating Solana and approaching a tenth of a cent are plausible on the fee line, but they say nothing about who keeps the spread, which is the part that decides whether rollups end up as businesses or as utilities.

What it means for users, token holders and builders

• Users: the fee you see will keep shrinking, but the cost you do not see, the value extracted by ordering around your trade, will not shrink as fast. Choosing chains with fairer ordering or private order flow matters more than chasing the lowest posted fee.
• Token holders: a rollup token is not a claim on the margin unless the protocol explicitly routes revenue to it, and most do not. ARB funds a treasury, OP funds a collective, Base has no token. Price the token on governance and ecosystem, not on fee revenue, until a credible revenue switch exists.
• Builders: the choice of data-availability layer and sequencer now sets both your cost base and your neutrality. A cheaper external DA layer widens your margin but adds a trust assumption; a shared sequencer trades away some revenue for faster finality and credibility.
• ETH holders: watch the blob floor and the burn, not the headline fee. The value that reaches ETH is the data bill rollups pay, and by design that bill is now near a floor.

What could break this forecast

Several things could move the margin faster than the base case allows.

• Glamsterdam slips again or ships broken. The 22 October pull showed the execution repricing is genuinely hard; another slip pushes the next user-fee cut into late 2027 and freezes margins where they are.
• A data-availability shock. If an external DA layer fails or the blob market spikes, the cost side of the margin moves sharply, in either direction.
• Sequencer centralization or outage. Single-sequencer rollups remain a single point of failure; a prolonged Base or Arbitrum outage would reprice the whole assumption that the margin is safe.
• Regulation. U.S. treatment of sequencer revenue and token revenue-shares is unsettled, and the SEC’s posture on whether a revenue-routing token looks like a security could reshape Arbitrum-style models. Coinbase’s choice to run Base without a token looks, in that light, like regulatory prudence as much as product design.
• A credible revenue switch. If a major rollup token starts paying holders out of the margin, the no-claim-on-revenue assumption flips, and the token reprices with it.

Frequently Asked Questions

Will Ethereum L2 fees keep falling in 2026 and 2027?

The easy part is done. Fees on the top rollups fell more than 99% since 2024 and now sit near a cent or below. Further cuts depend on the Glamsterdam upgrade repricing execution gas, which was delayed again on 22 October 2026 after security researchers found 13 contracts it could break, so the next leg down may not arrive until 2027.

What is a sequencer margin?

It is the difference between what a rollup charges users for a transaction and what it costs the rollup to post data to Ethereum and settle. With data costs near a floor and user fees near zero, that spread, plus the value captured by ordering transactions, is where rollup revenue now lives.

Which Ethereum L2 makes the most money?

By public estimates, Base earns the most sequencer revenue, roughly $60 million to $70 million in the first half of 2026, which Coinbase discloses in its SEC filings. Arbitrum is next at an estimated $35 million to $45 million, routed to its DAO treasury rather than to token holders.

Do ARB or OP token holders earn the fees?

Generally no. Arbitrum’s sequencer revenue funds a DAO treasury that holders govern but do not directly receive; Optimism funds a collective; Base has no token at all. A rollup token is a claim on governance and ecosystem, not on the fee margin, unless a protocol explicitly routes revenue to holders.

Why do cheap L2 fees worry ETH holders?

Because the cheaper rollups post data, the less ETH Ethereum burns, which weakens the deflation story. Fidelity estimated the EIP-7918 blob-fee floor, had it existed since 2024, would have added about $78.6 million (around 24,641 ETH) to value flowing to holders. ETH is no longer reliably deflationary.

By Priya Reddy, senior markets writer at HOGE Wire, covering Ethereum scaling, Layer 2 economics and on-chain markets.