The L2 Fee Plateau: Glamsterdam Slips to Q4 2026

The story of Ethereum scaling in 2026 has been, more than anything, a story of collapsing fees. In the first quarter of 2024, the average cost of a transaction across the three largest Layer 2 rollups sat at roughly eighteen cents ($0.180219). By the first quarter of 2026, that same average had fallen to about a tenth of a cent ($0.001512), a decline of 99.16% in two years, according to Token Terminal. Sending value on Arbitrum, Base or Optimism became something most users stopped thinking about, the way almost nobody checks the cost of sending an email.

That collapse had an engine, and for most of the last two years the engine was cheap data. The problem is that the engine just ran out of road. On 17 August, Ethereum’s protocol developers launched a dedicated public testnet called Platåberget to begin trials of Glamsterdam, the next major network upgrade, and in the same stretch confirmed that Glamsterdam’s mainnet activation has slipped from a mid-2026 target to the fourth quarter of the year, with no locked date. The upgrade that was supposed to deliver the next leg of fee compression is now months away.

This piece lays out what that delay means for the price of a rollup transaction between now and the end of 2027. The short version: the two forces that pushed L2 fees toward zero are on different clocks. One of them, cheap data availability, has already done almost all the work it is going to do for now and is pinned against a hard floor written into the protocol. The other, cheaper execution, is exactly what Glamsterdam was meant to unlock, and it just got postponed. Between now and Q4, expect a plateau, not another cliff.

L2 Fees in August 2026: A Snapshot

Before forecasting where fees go, it helps to fix where they are. As of late August 2026, the major rollups cluster in a narrow band that runs from well under a cent to a few cents for a simple transfer, with swaps and heavier contract calls costing several times that. Base, Coinbase’s rollup, is generally the cheapest of the large general-purpose chains; the optimistic rollups Arbitrum One and OP Mainnet sit close behind; the zero-knowledge chains zkSync Era, Linea and Scroll tend to run slightly higher per transaction because generating validity proofs carries a cost that batching data alone does not erase.

Exact figures depend heavily on the tracker and the moment you look, since fees move with network demand and with the ETH price. The numbers below, drawn from L2BEAT and consistent with first-half 2026 medians, should be read as typical rather than precise; on a busy afternoon they can be several times higher, and on a quiet night lower. What matters for the forecast is the shape, not the third decimal place: every major L2 now settles a simple transfer for pennies or less, and has done so consistently since the blob era began.

Two structural facts sit behind this table. First, all six of these chains buy their data availability from the same place, Ethereum, by posting compressed transaction data as blobs. That shared dependency is why the whole band moves together when Ethereum changes its blob rules. Second, these rollups now process several times as many transactions as Ethereum mainnet itself, so the fees on this table, not mainnet’s, are what most users actually experience day to day.

How Fees Got This Low: Blobs Did the Heavy Lifting

The reason fees fell so far, so fast, has a precise cause. Before March 2024, rollups paid for space in Ethereum’s regular transaction area, competing directly with every swap, mint and transfer on mainnet. That was expensive and volatile. The EIP-4844 upgrade, shipped with Dencun in March 2024, introduced a separate, temporary storage lane called a blob: a chunk of data that rollups can post cheaply, that lives on the network for a couple of weeks and then expires. Because blobs are priced in their own fee market, rollup data costs decoupled from mainnet congestion and collapsed almost overnight.

From there the roadmap has been about adding more blob capacity. Pectra, in May 2025, roughly doubled the blob target. Then Fusaka, which activated on 3 December 2025, introduced PeerDAS, a data-availability sampling scheme that lets nodes verify blob data without downloading all of it, and paved the way for two rapid follow-up bumps: BPO1 on 9 December 2025 lifted the blob target and maximum to 10 and 15, and BPO2 on 7 January 2026 pushed them to 14 and 21. Net of all this, the network carries well over twice the blob space it did before Fusaka, and data availability, which accounts for the large majority of a rollup’s operating cost, got cheaper again.

That is the machine that produced the 99% decline. It is also a machine that has, for the moment, largely finished its job.

The Floor Is Real: Why L2 Fees Will Not Hit Zero

For a while, the direction of travel looked like it pointed at zero. During long stretches of weak demand in early 2025, the blob base fee, the per-unit price rollups pay for blob space, sat at the literal minimum the protocol allowed, a number with eight zeros after the decimal point. Blobs were, in effect, almost free, and every extra bump in capacity threatened to keep them that way.

Fusaka changed that with a quieter, less-discussed measure: EIP-7918, which puts a floor under the blob base fee. Rather than letting blob prices collapse toward nothing when demand is soft, the rule anchors the blob base fee to the cost of Ethereum’s ordinary execution, so it cannot fall below roughly one sixteenth of the Layer 1 execution base fee. The effect is that blob space now has a real reserve price, and that price rises and falls with the rest of the network instead of flatlining near zero.

The shift was not subtle. The pseudonymous analyst Kydo (0xkydo) pointed out that once the floor kicked in, the blob base fee had jumped by roughly 15,000,000 times its old resting level. That sounds alarming until you remember the starting point was almost nothing; in absolute terms blobs remain cheap. But the design intent is clear, and it matters enormously for any forecast: Ethereum has deliberately taken the zero-fee outcome off the table. Further compression cannot come from making data ever closer to free, because the protocol will no longer allow it. It has to come from somewhere else.

The Second Motor: Execution, Not Data

If data availability is one motor of L2 fees, execution is the other. Every rollup transaction has to be computed, ordered and eventually settled against Ethereum, and each of those steps costs something. With the data motor now pinned against its floor, the swing variable for the next round of fee cuts becomes how cheaply rollups can execute transactions and how much settlement headroom Ethereum gives them.

This is precisely the territory Glamsterdam is built to address. Where Dencun and Fusaka were about data, Glamsterdam is about computation: raising the amount of work Ethereum can do per block, letting transactions execute in parallel rather than strictly one after another, and repricing individual operations so that gas costs better reflect the real burden they place on nodes. Cheaper, higher-throughput execution on Layer 1 feeds directly into lower settlement costs on Layer 2, and more room on mainnet means rollups can batch and settle more aggressively. That is the mechanism the market was counting on for the next leg down. It is also the mechanism that just got delayed.

There is a quieter third lever, too, and it depends on no upgrade at all: sequencer margin. Every major rollup today runs a single operator, the sequencer, that orders transactions and keeps the difference between what users pay and what the chain owes Ethereum for data and settlement. During a plateau, a rollup that wants to advertise lower fees can simply surrender some of that margin, and several already do. The catch is that this lever is shallow: margins on the cheapest chains are already thin, and trimming them is a marketing decision rather than a structural one, so it can nudge headline fees a little without touching the floor that Glamsterdam is meant to lower.

Glamsterdam Just Slipped to Q4 2026

Until recently, Glamsterdam was penciled in for roughly the middle of 2026, and some observers expected activation as early as September. That is no longer the plan. On 17 August, the Ethereum Foundation’s protocol team stood up Platåberget, a short-lived public testnet whose whole purpose is to let the community stress-test Glamsterdam before it reaches the longer-lived testnets. Platåberget forked to Glamsterdam on 20 August; a non-finality devnet is due to follow within the month, after which the Sepolia and Hoodi testnets will upgrade, and only then will mainnet follow.

The upshot, confirmed across developer coverage, is that the mainnet target has slipped from the first half of 2026 to the fourth quarter, with the Sepolia and Hoodi forks now expected in September. According to Odaily, the delay is a function of scope: Glamsterdam has grown into one of the largest single restructurings Ethereum has attempted, and core developers have repeatedly stressed that the correctness of the upgrade takes priority over hitting any particular date. Crucially, there is still no locked mainnet date; that will only firm up when an all-core-developers call agrees on a specific activation block. The upgrade that comes after Glamsterdam, informally called Hegota, is already being scoped in parallel, which tells you how far out the tail of this roadmap now stretches.

Read the table and the near-term picture is stark. The blob columns stopped moving in January; the next fork that changes anything a user actually feels is Glamsterdam, and Glamsterdam is now a fourth-quarter event at the earliest, with real-world fee effects bleeding into 2027.

What Glamsterdam Actually Changes, and Why It Is Not a Uniform Cut

It is tempting to file Glamsterdam under generic fee cut, but the details are more interesting and more uneven than that. The centerpiece, EIP-8037, introduces what developers call a state gas dimension. In plain terms, it charges extra for operations that create new state on the network, such as a transfer to an address that has never held funds before. One consequence is that the long-standing rule of thumb, that a simple transfer always costs 21,000 gas, no longer holds universally; some transfers will cost more, not less. Glamsterdam is a repricing, and repricings create winners and losers among transaction types rather than shaving a flat percentage off everything.

Alongside that sit the two headline changes most people associate with the upgrade. EIP-7732, known as enshrined proposer-builder separation or ePBS, bakes the block-building pipeline directly into the protocol and removes the external relays that currently sit between block proposers and builders; as a side effect it expands the window in which a block must be delivered and validated from around two seconds to roughly nine, which gives the network more room to carry data for rollups. EIP-7928, block-level access lists, lets clients know in advance which parts of state a block will touch, unlocking parallel execution. Add EIP-8007’s broader gas repricings, larger contract-size limits under EIP-7954, and forward-compatible consensus data structures under EIP-7688, and you have a package aimed squarely at execution throughput and a substantially higher effective block capacity.

How much does this actually lower fees? The honest answer is that nobody knows yet, and the projections vary. Some repricing analyses suggest execution costs for common operations could fall by well over half, but that is a modeled estimate, not a guaranteed protocol outcome, and EIP-8037’s state charges pull in the opposite direction for a subset of transactions. To keep data growth from spiraling as throughput rises, the upgrade also caps the network’s yearly data growth at a predictable ceiling, on the order of 120 gibibytes per year. The through-line is that Glamsterdam raises execution capacity substantially, but the benefit is uneven across transaction types and, for now, entirely theoretical, because none of it is live.

The Forecast: A Plateau Through 2027

Put the pieces together and the base case writes itself. The data motor is against its floor and will not move again until a future blob-parameter bump, and none is scheduled. The execution motor is real but postponed to the fourth quarter, and upgrades of this size tend to take weeks to propagate into the fees an ordinary user sees, as rollups tune their own software to pass the savings through. So the most likely path for the next several months is not another cliff but a plateau: fees roughly flat, wobbling with demand and the ETH price, until Glamsterdam lands and delivers a step down sometime in 2027.

Academic work points the same way while reminding us the long arc still bends down. A 2026 study by Meghan Ambrosia and Bruce Mizrach, published on arXiv, measured a decline of more than 95% in L2 median fees, from about $0.05 to $0.0015, between January 2024 and March 2026, and forecast that L2 median fees would fall below Solana’s by October 2026 and that L2s would surpass Solana on raw throughput by March 2029. That forecast assumes blob capacity keeps expanding on schedule; a delayed execution upgrade does not reverse the trend, but it flattens the near-term slope, which is exactly the plateau this piece describes.

One more wrinkle deserves emphasis, because it trips up a lot of forecasts: L2 fees are paid in ETH, so the dollar cost of a transaction is a product of gas used and the ETH price. With ETH trading around $2,420 in late August, per CoinGecko, a sharp rally could nudge dollar-denominated fees up even as the underlying gas cost keeps falling. The three scenarios below are framed in dollars and should be read with that caveat: they describe the direction the protocol is pushing, not a promise about any given day’s screenshot.

The Catch: Cheaper Fees Mean Less ETH Burned

Every cent shaved off an L2 transaction is a cent that does not flow to Ethereum the settlement layer, and that has quietly become one of the most contested questions in the ecosystem. Ethereum burns the base fee of ordinary mainnet transactions, so heavy L1 activity removes ETH from circulation. Blobs, by design, are cheap and priced in a separate market, so as activity migrated from mainnet to rollups posting blobs, the amount of ETH burned fell with it. Critics summarize the worry bluntly: the rollups took Ethereum’s users and left it with a fraction of the revenue.

The numbers make the tension concrete. In an analysis of the Fusaka upgrade, Fidelity Digital Assets noted that Base paid only around $5.2 million in blob fees over the prior year while earning roughly $94 million in user transaction fees, a striking gap between what a rollup collects and what it pays Ethereum. Fidelity framed EIP-7918’s floor as a deliberate attempt to give Ethereum pricing power in the data-availability market, and estimated that had the floor been active since the Deneb-Cancun upgrade, it would have added on the order of $78.6 million, about 24,641 ETH, in cumulative revenue for token holders. The blob floor, in other words, is not only about preventing zero-fee blobs; it is about making sure the base layer captures value as it scales.

This is where fee compression stops being a pure user-experience story and becomes an economic-security story. Bitcoin faces a version of the same problem from the other direction, as block rewards shrink and the network leans harder on fees to fund the miners that secure it, a dynamic we covered in our look at hashprice and Bitcoin’s security budget. For Ethereum, the burn and staking rewards are the twin pillars of ETH’s economics, and the people most exposed to that math are the holders of staked ETH and the liquid-staking tokens built on it, a landscape we mapped in our comparison of Lido, Rocket Pool and Frax. Cheaper fees are unambiguously good for users; whether they are good for ETH the asset depends on upgrades like EIP-7918 doing their job.

The Data-Availability Battleground

There is a second reason the blob floor matters, and it explains why Ethereum bothered to install a reserve price at all: Ethereum is no longer the only place a rollup can buy data availability. A pair of dedicated data-availability networks, Celestia and EigenLayer’s EigenDA, now sell blob-like space at prices well below Ethereum’s, and every rollup that moves its data off Ethereum both cuts its single largest cost and stops contributing to ETH’s economics.

The pricing gap is not marginal. Celestia’s co-founder, Mustafa Al-Bassam, has argued that Celestia’s blockspace is basically free, priced only at the minimum needed to deter spam, because the network is optimizing for growth rather than revenue. On that philosophy, external data availability can undercut Ethereum blobs by an order of magnitude or more. EigenDA takes a different route to the same destination, borrowing economic security from restaked ETH, a mechanism we unpacked in our guide to restaking and the great unwinding. Either way, the competitive floor on what Ethereum can charge for blobs is set not only by its own roadmap but by what Celestia and EigenDA are willing to charge, which is exactly why the EIP-7918 floor is described by its designers as pricing power rather than a mere anti-spam measure.

For the fee forecast, the DA battleground cuts both ways. Cheaper external DA could pull some rollups’ costs below what Ethereum blobs allow, extending compression for those chains. But it also drains demand and burn away from Ethereum, sharpening the value-accrual fight above and giving Ethereum every incentive to keep its own blob prices from collapsing. The plateau, in short, is partly a truce in a price war that has not actually ended.

Who Wins the Plateau, and Who Waits

A plateau is not neutral; it rewards some rollups and pressures others. Base enters it from the strongest position: consistently the cheapest of the large chains, home to a leading share of L2 activity, and backed by a parent, Coinbase, that reports Base’s sequencer revenue as a line item in its filings with the U.S. Securities and Exchange Commission. Base has no token to defend and no external investors demanding that fee income be shared, which lets it run thin margins for as long as it likes. Coinbase’s broader strategy, including how it is folding on-chain infrastructure into its exchange business, is something we examined in our piece on the exchange AI-agent race.

Arbitrum and Optimism enter the plateau with large treasuries and token economies to manage, which makes the margin question more delicate: sequencer revenue that is not passed to token holders becomes a governance flashpoint the longer fees stay flat. The zero-knowledge chains, zkSync Era, Linea and Scroll, arguably have the most to gain from what comes next, because Glamsterdam’s execution repricing helps proving-heavy workloads more than it helps the optimistic chains; for them, the plateau is a waiting game with a real prize at the end. Sitting over all of it is a slow-moving regulatory question about whether and how these tokens and their fee flows are treated as securities under the SEC, an argument we followed in our coverage of the fight over the DeFi gatekeeper. The plateau favors whoever has the deepest pockets and cleanest distribution during the wait; the step down, when it comes, favors whoever is most execution-bound today.

The Longer Horizon: Lean Ethereum and the L1 Comeback

Zoom out past Glamsterdam and the picture gets more ambitious. In July, Vitalik Buterin outlined what he calls the Lean Ethereum roadmap, a multi-year overhaul he described to The Defiant as the third major iteration of Ethereum, with the Merge as the second. Over what he framed as roughly the next five years, almost every major piece of the protocol is slated to be replaced, including a redesign of how tokens such as ERC-20 assets are stored that he said could cut their transaction fees by more than ten times.

The reason this matters for an L2 fee forecast is subtle but important. Cheaper, higher-capacity execution on Layer 1 does two things at once: it lowers the settlement cost that rollups pass through to users, and it lowers the very number the EIP-7918 blob floor is pegged to, since that floor tracks L1 execution cost. It also raises an awkward possibility that the L2 narrative rarely confronts: if mainnet becomes cheap and fast enough, some of the activity that fled to rollups could drift back, reshaping the relationship between L1 and L2 fees rather than simply pushing both to zero in lockstep. Glamsterdam is the near-term event; Lean Ethereum and the fork after it are the reminder that the fee floor itself is a moving target over the multi-year horizon.

What to Watch Between Now and Q4

Forecasts age quickly in this corner of crypto, so here is the short list of signals that would confirm or break the plateau thesis over the coming months.

• An all-core-developers call that locks a specific Glamsterdam mainnet activation block. That single event converts a vague Q4 target into a real date and starts the countdown to the next step down.
• The Sepolia and Hoodi testnet forks, expected in September. Clean forks keep Q4 alive; another slip pushes the fee step into 2027.
• A scheduled BPO3. A dated blob-parameter bump would add data capacity independently of Glamsterdam and is the only near-term lever that could move fees before the execution upgrade lands.
• Any large rollup migrating its data availability to Celestia or EigenDA, or a disruption at one of those layers. Either would reprice the DA battleground and, with it, the competitive floor on fees.
• The ETH price. Because fees are ETH-denominated, a strong rally can lift dollar fees even while gas costs fall, and a sell-off can flatter them.
• Sequencer-revenue disclosures, from Coinbase’s filings to Arbitrum and Optimism governance, which reveal how much margin each chain is actually willing to give up during the wait.

None of these changes the core conclusion. The cheap-data era delivered a once-in-a-cycle collapse in rollup fees, that era is now bounded below by a deliberate protocol floor, and the next meaningful leg down is gated on an execution upgrade that just moved to the far end of the year. For users, fees are already low enough to ignore. For anyone forecasting the next move, the honest answer is patience: the price of a transaction is going to sit roughly where it is until Glamsterdam ships.

Frequently Asked Questions

When will Ethereum’s Glamsterdam upgrade go live?

Ethereum’s core developers are targeting the fourth quarter of 2026 for Glamsterdam, but no exact date is locked. The public Platåberget testnet began on 17 August 2026, with Sepolia and Hoodi testnet forks expected in September and mainnet activation to follow once an all-core-developers call sets a specific activation block.

Will Ethereum Layer 2 fees keep falling in 2026?

Only modestly. The cheap-data lever that drove most of the decline is now pinned against the EIP-7918 blob fee floor, and the execution upgrade that would deliver the next leg, Glamsterdam, has slipped to the fourth quarter of 2026. Expect L2 fees to plateau through the rest of 2026, with a larger step down more likely in 2027 once Glamsterdam is live.

Why can’t Layer 2 fees just fall to zero?

Ethereum’s EIP-7918 upgrade sets a floor on the blob base fee, pegging it to roughly one sixteenth of the Layer 1 execution base fee, so blob space can no longer be almost free during quiet periods. Rollups also still pay for execution, proving and sequencer operation, so there is always a nonzero cost to process a transaction.

Which Ethereum Layer 2 has the cheapest fees?

Among the large general-purpose rollups, Base is generally the lowest-cost for a simple transfer, with Arbitrum One and OP Mainnet close behind and the zero-knowledge chains a little higher because proving carries its own cost. All of them sit between a fraction of a cent and a few cents, and the ranking shifts with network demand and the ETH price.

What is EIP-8037 and how does it change gas costs?

EIP-8037 introduces a state gas dimension in the Glamsterdam upgrade. It charges extra gas for operations that create new state, such as transfers to brand-new accounts, so the long-standing assumption that a simple transfer always costs 21,000 gas no longer holds. It is a repricing, not a blanket cut, so some actions get cheaper while state-creating ones get more expensive.

By Marcus Okafor, DeFi and infrastructure correspondent at HOGE Wire.