The L2 Fee Countdown Restarts: Glamsterdam Hits Testnet

For most of the summer, the story of Ethereum Layer 2 fees was a story about waiting. The cost of posting data to Ethereum had already collapsed to a floor that no amount of extra blob space could push lower, and the upgrade meant to unlock the next leg down, Glamsterdam, had drifted out of its original window into a vague Q4 2026 with no date attached. On September 18 the waiting ended. Client teams set a hard software freeze for September 29 and scheduled the first public test fork on the Sepolia network for October 6, with a second on the Hoodi testnet to follow on October 27, according to reporting from The Cryptonomist. The compression clock, stalled since August, is ticking again.

That restart matters because the easy phase of L2 fee compression is over. The average fee across the three largest rollups fell from roughly $0.18 in early 2024 to about $0.0015 in early 2026, a decline of 99.16%, according to Token Terminal. Almost all of that came from cheaper data. The next round is different: it depends on execution, on how much computation Ethereum can pack into a block and how efficiently rollups can pass those savings on to users. This is a forecast of what happens next, how far fees can realistically fall, when, and why the answer is messier than fees going to zero.

The compression clock restarts

For months, the Glamsterdam upgrade had no date. After it slipped out of its original first-half window, core developers would only say Q4 2026, and even that carried an asterisk: no activation block, no locked testnet schedule, just an intention. That changed on September 18, when client teams committed to a code freeze on September 29 and set the first public fork on Sepolia for October 6. A second rehearsal on Hoodi is tentatively slated for October 27. The Ethereum Foundation’s own Glamsterdam roadmap page now lists the Sepolia fork as the next milestone on the path to mainnet.

Two dated testnet forks are not the same as a mainnet launch, and the developers have been careful not to promise one. Mainnet remains an unscheduled Q4 2026 event, and if the two testnets surface consensus bugs, it can slide into 2027. But a fork with a real date is the first hard commitment the upgrade has had since the summer, and it re-anchors every fee forecast that had gone stale during the plateau. When Sepolia forks on October 6, the market gets its first live look at how much cheaper execution actually becomes.

Why does a testnet date move a fee forecast at all? Because the entire thesis of further compression rests on Glamsterdam shipping. Data costs are already at their floor. Nothing in the current roadmap between now and Glamsterdam changes what a rollup pays to post its data. So the timing of the next fee cut is, quite literally, the timing of this upgrade, and for the first time since August that timing has numbers attached to it.

From Dencun to the floor: how fees fell 99%

To see why execution is now the only lever left, it helps to trace how the last 99% of compression happened, because none of it came from execution.

The turning point was Dencun in March 2024, which introduced EIP-4844 and the blob: a dedicated, temporary data lane that let rollups post their transaction batches far more cheaply than stuffing them into regular calldata. Blob space cut L2 data costs by more than 90% overnight. Pectra doubled the blob allocation in May 2025. Then Fusaka arrived on December 3, 2025, carrying EIP-7594, or PeerDAS, the data-availability sampling scheme that lets the network safely scale blob throughput without every node downloading every blob, per the Ethereum Foundation’s Fusaka announcement.

Fusaka shipped conservatively at the same 6-target, 9-maximum blob count it inherited, then raised the ceiling through two Blob Parameter Only forks: BPO1 on December 9 lifted the target and max to 10 and 15, and BPO2 on January 7, 2026 took them to 14 and 21, roughly 2.3 times the pre-Fusaka blob space. The result, tracked by Token Terminal, was an average fee across Arbitrum, Base, and Optimism near a tenth of a cent by early 2026. On growthepie, a simple transfer in 2026 typically clears for around two cents on Base, three on OP Mainnet, and four on Arbitrum One, with sub-cent averages when the network is quiet and higher spikes when L1 is congested.

Why blobs cannot push fees any lower

Here is the uncomfortable fact for anyone forecasting more of the same: blobs have hit a wall, and it was put there on purpose.

Fusaka’s most consequential clause for fees is not PeerDAS but EIP-7918, which sets a floor under the blob base fee. Before Fusaka, when blob demand was light, the blob base fee could collapse toward a billionth of a gwei, effectively free data. EIP-7918 ties the blob fee to execution cost instead, pinning it to roughly the L1 execution base fee divided by 16, so it can no longer fall to nothing, as the EIP-7918 specification lays out.

The effect was dramatic. Kydo, a researcher who posts as 0xkydo, noted that the blob base fee jumped “up 15,000,000x since Fusaka” once the floor kicked in, because it had been sitting at a rounding error for long stretches, in a widely cited post on X. That is not a bug; it is the point. The floor exists so that data availability generates a consistent revenue stream for ETH holders rather than being given away, a design choice with real consequences for value accrual that we return to below.

The takeaway for a fee forecast is blunt. Adding blob capacity through future BPO forks will help absorb demand spikes and keep fees from surging, but it cannot drag the baseline lower, because the baseline is now anchored to L1 execution cost by design. BPO3 and BPO4, which developers have discussed pushing toward an aspirational 128 blobs per block, remain unscheduled while the teams study telemetry from the first two forks. More blobs are a pressure-relief valve, not a discount.

The next lever is execution, not data

If data is floored, the only place left to find savings is execution: the actual computation Ethereum performs to process a transaction. This is where Glamsterdam concentrates its firepower, and it is why the upgrade matters more for fees than another round of blob increases would.

A rollup’s cost has two big components. The first is data availability, what it pays to post its compressed batch to Ethereum, which post-Fusaka is cheap and floored. The second is the cost of execution, both on the L2 itself and, for settlement, on L1. As data has fallen away, execution and the rollup’s own operating margin have become the larger share of what a user actually pays. Squeeze execution, and you squeeze the part of the fee that data can no longer touch.

Glamsterdam attacks execution from two directions at once: it raises the amount of computation a block can hold, and it changes how that computation is priced and ordered. Neither is a simple discount. Some operations will get cheaper while others get more expensive, and the headline throughput number depends on validators voting the gas limit up over time rather than a switch being flipped. But the direction is unambiguous. For the first time since Dencun, the protocol is targeting the execution layer as the primary source of fee relief, and the changes reach far enough that they ripple all the way up to how wallets and exchanges handle deposits, a shift we examined in our look at how EIP-7702 is rewriting exchange custody.

Inside Glamsterdam: the EIPs that move fees

Glamsterdam bundles about ten EIPs, and a handful of them bear directly on fees. The two headliners are architectural. EIP-7732, enshrined proposer-builder separation or ePBS, moves block building into the protocol itself and removes the third-party relays that currently sit between proposers and builders. It also widens the window for propagating a block from about two seconds to roughly nine, which gives the network more room to handle larger blocks safely, per the Ethereum Foundation’s roadmap. EIP-7928, block-level access lists, ships a map of which pieces of state each transaction touches, so validators can execute independent transactions in parallel rather than one after another. Parallel execution is the throughput unlock that makes a much larger block practical.

The repricing EIPs are where the nuance lives, and where the popular framing of a clean fee cut breaks down. EIP-2780 reduces the intrinsic gas cost of a transaction, the flat baseline every transaction pays, which makes simple transfers cheaper. But EIP-8037 raises the gas cost of creating new state, and EIP-8038 updates the cost of accessing state. In plain terms, sending ETH to a brand-new address will cost more after Glamsterdam, not less, and the familiar assumption that a basic transfer always costs 21,000 gas no longer holds. This is a repricing to reflect real computational load, not an across-the-board markdown.

That distinction matters for the forecast. Secondary coverage has floated a headline L1 fee reduction of around 78% and an MEV cut of up to 70%, but those figures, reported alongside the 200 million gas-limit target, are projections built on the larger block and smoother ordering, not guarantees, and they will not apply evenly across transaction types. The honest read is that Glamsterdam makes throughput cheaper in aggregate while making some specific operations dearer.

The 200 million gas question

The single number that matters most for how far fees fall is the block gas limit. Ethereum blocks today target roughly 60 million gas. Glamsterdam’s client teams are testing at a 150 million reference limit and working toward a 200 million gas floor, more than triple current capacity, according to the roadmap and to The Defiant’s reporting on the final devnet.

More gas per block means more transactions competing for the same space, which relaxes the fee auction that sets prices during busy periods. On L1, that translates into cheaper settlement; on L2s, it means rollups can post and prove more activity for the same overhead, and competition should push those savings toward users. Combined with parallel execution from block-level access lists, the theoretical throughput ceiling rises sharply, with some estimates pointing to five figures of transactions per second at the base layer.

The catch is that the gas limit is not set by the upgrade. It is set by validators, who vote it up gradually as they gain confidence that larger blocks propagate reliably and do not centralize block production toward operators with the fastest hardware and connections. Glamsterdam raises the ceiling and, through ePBS and block-level access lists, makes larger blocks safer to reach; it does not force the limit there on day one. History suggests validators move deliberately. So the 200 million figure is best read as a destination for 2027, not a launch-day number, and the fee relief it implies arrives in steps rather than a single cliff.

ePBS, MEV, and the risk Potuz is warning about

Not all of a transaction’s cost shows up on the fee estimator. A meaningful slice is extracted through ordering: maximal extractable value, or MEV, the profit that block builders capture by arranging, inserting, or front-running transactions. On rollups, where a single sequencer controls ordering, this invisible cost is real even when the visible fee is a fraction of a cent, and it is one reason block-level primitives such as flash loans remain so potent.

ePBS is Glamsterdam’s answer to part of this problem on L1. By enshrining proposer-builder separation and letting a proposer commit to a builder’s bid without seeing its contents, the design aims to reduce the manipulation that relays currently enable. Secondary coverage has attached figures as high as a 70% reduction in certain MEV, though those remain projections.

But ePBS also introduces a new attack surface, and one of Ethereum’s own consensus developers has been blunt about it. Potuz, an Ethereum consensus developer, warned that the testnet is exposed to payload-withholding attacks, where an adversary floods the network with builders that promise blocks and then refuse to deliver them. “I can just spin up a thousand builders, rotate them, offer very high bids, and not produce payloads. Any teenager can do this,” he said of the Sepolia test setup, as quoted by The Cryptonomist. The attack threatens test infrastructure rather than mainnet funds, but it is exactly the kind of edge case the October forks exist to shake out, and it is a reminder that ePBS is among the most complex changes Ethereum has ever attempted. If it needs another revision, the fee timeline moves with it.

The forecast: how low, and when

Put the pieces together and a forecast takes shape. Data is floored; execution is the lever; the lever gets pulled in stages as the gas limit rises; and the whole schedule hinges on Glamsterdam clearing its October testnets and reaching mainnet in Q4 2026 or slipping into 2027.

The academic literature points the same direction. A 2026 fee-forecasting study by Ambrosia and Mizrach found that L2 median fees had already fallen more than 95%, from about $0.05 to roughly $0.0015, and projected that the median L2 fee would drop below Solana’s by October 2026, with L2 throughput surpassing Solana’s around 2029, in a paper posted to arXiv. That trajectory does not require Glamsterdam; it is the momentum already in the system. Glamsterdam is what determines whether the curve keeps bending down or flattens into a longer plateau.

The three scenarios below translate the mechanics into rough fee ranges for a simple transfer by the end of 2027. They are scenarios, not predictions of a specific price, and they move with two variables above all: whether Glamsterdam ships on time, and whether user demand grows fast enough to eat the new capacity. That second variable is the one forecasters routinely underestimate. Cheaper blockspace invites more activity, and more activity competes for space, which can hold fees flat even as the per-unit cost of computation falls. Compression is a race between supply and demand, and Glamsterdam only adds to supply.

The value-accrual reckoning: the burn broke

Every cent shaved off an L2 fee is a cent that does not get burned on L1, and that arithmetic has finally caught up with Ethereum’s monetary story. The ultrasound money thesis, that fee burning under EIP-1559 would make ETH deflationary, has quietly broken. After Dencun pushed activity onto rollups and off the base layer, the burn shrank and net issuance turned modestly positive; ETH is no longer reliably deflationary, as CoinLedger’s review of the ultrasound-money narrative spells out.

This is the counterweight to cheap fees, and Fusaka’s EIP-7918 floor was partly a response to it. Fidelity Digital Assets calculated that if the blob-fee floor had been active since the Deneb-Cancun upgrade, Ethereum would have captured an extra $78.6 million, about 24,641 ETH, in cumulative blob-fee revenue for holders, in its Fusaka value-accrual analysis. The same analysis put the added cost to rollups at roughly $6.02 per blob, with Base alone facing about $30.6 million more in fees over a year. In other words, the floor deliberately transfers a little value from rollups back to ETH holders, and Glamsterdam’s EIP-7708, which makes ETH transfers and burns emit a log, is a small piece of plumbing that makes that value flow easier to track.

The tension is structural and it is not going away. Coinbase, which operates Base, reports the network’s sequencer revenue as a new revenue stream in its filings with the U.S. Securities and Exchange Commission (SEC), and independent estimates put that revenue in the range of $60 to $70 million for the first half of 2026, against roughly $35 to $45 million for Arbitrum, per VaaSBlock. The more successful rollups become, the more value accrues to sequencer operators and the less flows to the base layer, which is why a fee forecast is inseparable from a value-accrual forecast. For ETH holders living in a low-issuance, roughly 4% staking world, the question of where rollup value lands is not academic, as we discussed in our guide to solo staking.

Who wins when fees hit the floor

Sub-cent fees are not an end in themselves; they unlock use cases that were previously uneconomic. As compression grinds toward the floor, the winners are the applications for which a few cents was the difference between working and not.

Payments and remittances are the obvious one, though with a caveat: once the on-chain fee is negligible, the binding constraint moves off-chain to currency conversion and cash-out, not gas. High-frequency and algorithmic strategies benefit directly, because every basis point of cost matters when you are rebalancing thousands of times. On-chain credit is another, since cheap execution makes small loans and frequent liquidations viable; the mechanics of that market are covered in our explainer on how on-chain lending works. Perhaps the fastest-growing category is autonomous agents: software that trades, rebalances, and pays for services without a human clicking approve. When each action costs a fraction of a cent, an agent can afford to act constantly, which is why the rise of agent-driven DeFi tracks the fall in fees so closely.

Gaming, the other half of this publication’s beat, sits squarely in this camp. On-chain items, micro-rewards, and in-game economies only make sense when a transaction costs less than the reward it delivers, and the last leg of compression is what finally puts routine game actions below that threshold. The losers, if there are any, are the rollups themselves, whose sequencer margins compress alongside user fees, and any token whose value was premised on those fees staying high.

The data-availability wildcard

One factor could push L2 fees below even the Glamsterdam floor, and it does not live on Ethereum at all. Rollups do not have to buy their data availability from Ethereum blobs; they can post to dedicated DA layers such as Celestia or EigenDA, which price data far more cheaply.

The gap is large. Analysis by Blockworks pegged Ethereum blob data at around $3.83 per megabyte against roughly $0.07 on Celestia, a difference of more than fifty times, though the exact ratio swings with demand. Celestia’s Matcha upgrade in late 2025 expanded its block capacity sharply and pushed its data prices lower still, and EigenDA has attracted rollups looking for cheap, high-throughput data. A rollup that moves its DA off Ethereum can undercut the blob floor entirely.

That freedom cuts against Ethereum’s value accrual, which is the catch. Every batch posted to Celestia instead of an Ethereum blob is revenue Ethereum does not earn and ETH that is not burned, deepening the issuance problem from the previous section. So the DA market is a genuine wildcard for the forecast: it can drive user fees below the Glamsterdam floor, but only by routing value away from the chain the whole ecosystem settles on. How rollups navigate that tradeoff, security and value alignment on Ethereum versus raw cost off it, will shape the true fee floor as much as any single upgrade.

What could still go wrong

Forecasts fail on their assumptions, so it is worth naming the ones that could break this one.

The schedule is the first risk. Glamsterdam has already slipped once, from the first half of 2026 into Q4, and the October testnets exist precisely because complex changes like ePBS tend to surface problems late. If Sepolia or Hoodi throws a consensus bug, mainnet moves into 2027 and the next fee cut moves with it. The second risk is demand. Compression assumes supply grows faster than usage, but cheaper blockspace reliably invites more of it; a surge in on-chain activity, a popular new application, an agent boom, or a market rally can keep fees flat even as per-transaction costs fall.

The third risk is complexity and security. ePBS is among the most ambitious changes Ethereum has attempted, and the payload-withholding concern Potuz raised is a live example of the kind of edge case that can force a redesign. The fourth is the gas limit itself: validators, not the upgrade, decide how high it goes, and a cautious validator set could leave much of the 200 million target unrealized for a long time. Finally, the DA wildcard cuts both ways; a security incident on an external DA layer could send rollups scrambling back to more expensive Ethereum blobs. None of these is likely to reverse compression, but each can flatten it into another plateau.

The longer arc: Vitalik’s Lean Ethereum

Zoom out past Glamsterdam and the ambition grows. Vitalik Buterin’s Lean Ethereum roadmap, outlined in mid-2026, frames the next several years as a ground-up overhaul rather than a series of tweaks. He described it as “the third major iteration of Ethereum,” placing it alongside the original launch and the Merge, and said that “almost every major piece of the protocol will be replaced” over roughly the next five years, according to The Defiant.

For fees specifically, the most relevant thread is a proposed redesign of how ERC-20 tokens are represented, moving toward a UTXO-style model that Buterin claimed could cut transaction fees by more than tenfold. The roadmap also revives the idea of L1 reclaiming execution that had migrated to rollups, which would reshape the entire L1-versus-L2 fee relationship. None of this is close; it is a multi-year program, and the specific figures in secondary coverage should be treated as aspirations rather than commitments.

The point for a 2026 forecast is that Glamsterdam is not the end of the road. It is the next concrete step in a much longer compression program, and the fact that Ethereum’s leadership is already sketching what comes after execution scaling tells you the direction of travel is settled even when the timing is not.

What to watch before mainnet

The forecast will resolve in public over the coming months, and a handful of checkpoints will tell you which scenario is playing out.

• September 29: the client software freeze. A slip here delays everything downstream.
• October 6: the Sepolia fork, the first live test of Glamsterdam’s execution changes.
• October 27: the tentative Hoodi fork, the second rehearsal.
• An All Core Developers call setting a mainnet activation block, the signal that Q4 2026 is real.
• A dated BPO3, which would confirm developers are comfortable scaling blobs further.
• The ETH burn and net-issuance trend, the value-accrual counterweight to cheaper fees.
• DA-market moves, especially any large rollup shifting its data availability on or off Ethereum.

Watch those in order. If the September freeze holds and Sepolia forks cleanly on October 6, the base case tightens toward a Q4 mainnet and the next real fee cut lands around the turn of the year. If October slips, the plateau extends, and the story in three months looks a lot like the story in August: waiting.

Frequently Asked Questions

When will the Glamsterdam upgrade go live on Ethereum mainnet?

Glamsterdam has firm testnet dates but no mainnet date. Client teams set a code freeze for September 29, 2026, a Sepolia test fork for October 6, and a tentative Hoodi fork for October 27. Mainnet is expected in Q4 2026, but no activation block has been locked, and a bug in either testnet could push it into 2027.

How much cheaper will Ethereum L2 fees get after Glamsterdam?

Glamsterdam targets execution, raising the block gas limit from about 60 million toward a 200 million floor and enabling parallel execution. Secondary estimates put the L1 fee reduction near 78%, but that is a projection, not a guarantee, and it will not apply evenly; some operations, like creating new accounts, actually get more expensive. Expect simple L2 transfers to trend from a few cents toward sub-cent levels in stages through 2027.

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

Fusaka’s EIP-7918 sets a floor under the blob base fee, tying it to roughly the L1 execution base fee divided by 16, so data can no longer be given away for free. Beyond that floor, execution cost and each rollup’s operating margin set a practical minimum, and rising demand can keep fees flat even as per-transaction costs fall.

Do cheaper L2 fees hurt the value of ETH?

They create tension. Cheaper L2s and fewer L1 transactions mean less ETH is burned, and after the Dencun upgrade net issuance turned modestly positive, so ETH is no longer reliably deflationary. Fusaka’s blob-fee floor was designed partly to route more data revenue back to ETH holders; Fidelity estimated the floor would have added about $78.6 million to holders had it been active since the Cancun upgrade.

What is ePBS in the Glamsterdam upgrade?

ePBS, or enshrined proposer-builder separation (EIP-7732), moves block building into the Ethereum protocol and removes the third-party relays that sit between proposers and builders today. It aims to reduce MEV manipulation and widen the block-propagation window, but it is one of the most complex changes Ethereum has attempted and carries new risks, such as the payload-withholding attack that developers are testing for on Sepolia.

By Priya Reddy, senior markets writer at HOGE Wire, covering Ethereum scaling, Layer 2 economics, and the predictions desk.