L2 Fee Compression Forecast: How Low Can Rollup Fees Go?

The story of Ethereum layer-2 fees over the last two years is a story of collapse. A simple transfer that cost a couple of dollars on a busy rollup in early 2024 now settles for a fraction of a cent. According to Token Terminal, the average transaction fee across the three largest rollups fell from about $0.18 in the first quarter of 2024 to roughly $0.0015 in the first quarter of 2026, a decline of 99.16%. That is not a rounding error; it is a repricing of blockspace itself.

The interesting question for 2026 and 2027 is no longer whether fees fall, but how much further they can go, where the floor sits, and who pays the bill when blockspace is nearly free. Two events frame the forecast. Behind us is Fusaka, the December 2025 upgrade that switched on data-availability sampling and handed developers a dial to crank blob capacity higher every few weeks. Ahead of us is Glamsterdam, the next hard fork, now in final testing, which rewrites how blocks are built and reprices gas across the board.

This piece lays out where L2 fees stand in mid-2026, the mechanics pushing them lower, the one mechanism that stops them reaching zero, and three scenarios for where the median rollup fee lands by the end of 2027. The short version: the compression is real, but so is the floor, and the more important fight is shifting from the cost of a transaction to who captures the value when transactions cost almost nothing.

Where L2 Fees Actually Sit in Mid-2026

By the middle of 2026, sending a plain token transfer on a major Ethereum rollup costs somewhere between a fraction of a cent and a few cents, depending on the network, the moment, and, crucially, how you measure it. Live trackers such as L2BEAT publish per-operation cost estimates that move with L1 gas and blob demand; comparison dashboards consistently show Base as the cheapest of the large optimistic rollups, with Arbitrum One and OP Mainnet a hair higher. A swap or a multi-step DeFi interaction costs more than a transfer, sometimes several times more, because it burns more execution gas.

To feel the scale of the change, rewind to 2021, when a single token swap on Ethereum mainnet during a busy stretch could cost tens of dollars in gas, and a complex DeFi position could run into three figures. Rollups existed, but they were a niche escape hatch rather than the default. The move to sub-cent fees is the difference between blockspace as a scarce luxury and blockspace as a commodity, and it is why the center of gravity for everyday activity has shifted almost entirely onto L2s.

Two numbers are worth separating: the median simple transfer and the average transaction. The gap between them is large because averages are dragged upward by heavy DeFi calls. That is why one source will quote sub-cent medians while another quotes five-to-nine-cent averages for the same network in the same month; both can be right. The table below shows approximate mid-2026 costs for a routine transaction on the leading rollups, with the firm caveat that these figures shift daily and vary by tracker.

The headline is not any single number but the shape of the curve. On L2BEAT’s cost tracker and Token Terminal’s fee series, the trend line for every leading rollup points in one direction, and it has for two years. To understand why, and where it goes next, you have to look at how Ethereum engineered the drop.

How the Floor Fell: Dencun, Pectra, and Fusaka

The compression did not happen by accident. It was built, upgrade by upgrade. The pivot was Dencun in March 2024, which introduced EIP-4844 (proto-danksharding) and with it blobs, a dedicated, cheap data lane for rollups to post their compressed transaction data. Blobs launched with a target of 3 and a maximum of 6 per block, and they cut L2 data-posting costs by more than 90% almost overnight. Before Dencun, rollups competed for the same expensive calldata as everyone else; after it, they had their own aisle.

Pectra followed in May 2025, raising the blob target and maximum to 6 and 9. Then came Fusaka on December 3, 2025, the upgrade that changed the trajectory. Fusaka’s headline was EIP-7594, PeerDAS (peer data-availability sampling), which lets nodes verify that blob data is available by sampling small pieces of it instead of downloading every blob in full. That is the unlock: once nodes no longer have to hold all the data, the network can safely carry far more of it. The Ethereum Foundation confirmed the December 3 activation at slot 13,164,544.

Here is the subtle part that most coverage missed: Fusaka did not raise the blob count on day one. It shipped the machinery for higher throughput and then scaled up through a brand-new mechanism, one that turned blob capacity from a number set once per hard fork into a dial developers can turn on a schedule.

The BPO Machine: Turning Blob Capacity Into a Dial

Fusaka’s quiet revolution was the Blob Parameter Only fork, or BPO. Instead of bundling every blob increase into a full, multi-month hard fork, developers can now schedule a lightweight fork that changes only the blob target and maximum. It is a throttle, not a rebuild, and it lets the network add capacity in measured steps while it watches how clients and nodes cope.

The throttle moved fast. Per the Ethereum Foundation, BPO1 activated on December 9, 2025, lifting the target to 10 and the maximum to 15. BPO2 followed on January 7, 2026, taking the target to 14 and the maximum to 21. In roughly five weeks, per-block blob capacity more than doubled from where Fusaka started. That is the supply side of the fee equation expanding on a cadence the network had never had before.

Blob fees follow a supply-and-demand curve much like ordinary EIP-1559 gas: when supply outruns demand, the blob base fee sits at its minimum and rollups pay almost nothing to post their data. Through much of early 2026, that is exactly what happened, because capacity ran well ahead of demand. Developers have since made clear they will not rush BPO3; core contributors have said further increases wait on real blob usage catching up to the capacity already added, with a longer-term path toward roughly 128 blobs per block discussed but not scheduled. That patience is itself a forecast input: the supply dial is paused until demand justifies the next turn. And it points at the mechanism that stops the price from ever hitting zero.

Why L2 Fees Cannot Reach Zero: The EIP-7918 Floor

Here is the single most important fact for any honest fee forecast: Ethereum has deliberately built a floor under blob prices. Fusaka shipped EIP-7918, which bounds the blob base fee by execution cost. Before it, when blob demand was thin, the blob base fee could decay toward a negligible number, and for long stretches it did exactly that.

EIP-7918 introduced a reserve price that ties the blob base fee to a fraction (about one-sixteenth) of the prevailing execution-layer base fee per blob. In plain terms, posting a blob can never be much cheaper than a comparable amount of ordinary L1 gas. When the network is quiet the floor is low, but it is a floor, and it rises and falls with L1 rather than collapsing to zero.

The effect was immediate and easy to misread. Analyst Kydo (0xkydo) pointed out that the blob base fee jumped roughly 15,000,000 times after the change took hold, precisely because it had been pinned at 0.000000001 Gwei for long stretches before the floor existed (via X). The number sounds absurd until you notice the starting point was effectively nothing; the floor did not make blobs expensive, it made them non-free. That distinction is the whole ballgame for a 2026-2027 forecast.

The takeaway is blunt: any prediction of free L2 transactions is wrong by construction. Further compression from here cannot come from cheaper blobs alone; it has to come from execution efficiency, better data compression, thinner sequencer margins, and lower L1 base fees. That reframes the entire outlook around a different upgrade, the one now in testing.

Glamsterdam: The Next Lever on the Fee Curve

If Fusaka was about data, Glamsterdam is about execution. The upgrade, bundled under Meta EIP-7773, entered its final devnet phase in mid-2026, with public testnet activations rolling out in August 2026 and a mainnet target in the second half of the year; core developers have floated September, following Ethereum’s usual fork cadence, per The Defiant. Three changes matter for fees.

First, EIP-7732 enshrines proposer-builder separation (ePBS) in the protocol, formalizing the split between the validator that proposes a block and the specialized builder that fills it. Second, EIP-7928 introduces block-level access lists (BALs), which let clients execute a block’s transactions in parallel rather than strictly one after another; proponents argue that lifts throughput substantially. Third, and most directly, a gas-repricing package centered on EIP-7904 rebalances the cost of individual operations to reflect what they actually cost modern hardware, cutting the price of overpriced operations while raising the price of underpriced ones that have been abused as attack vectors.

The repricing is the headline for fee watchers. Backers of the package project it could cut execution-layer fees by roughly 78%, with a common action like a Uniswap swap that costs a few dollars in gas potentially dropping under a dollar on L1, and proportionally more on rollups that inherit the cheaper execution. Those are projections, not guarantees, and the final figure will depend on the parameter set that ships; but the direction is clear. Glamsterdam also raises the block gas limit toward 200 million, roughly triple today’s level, expanding raw L1 capacity alongside the price cut.

Why does an L1 upgrade compress L2 fees? Because rollups pay for two things: data (blobs, addressed by Fusaka) and, indirectly, the L1 execution cost that anchors the EIP-7918 floor and prices the settlement transactions they still post to Ethereum. Cheaper, repriced L1 execution lowers that floor and shrinks the settlement bill. Glamsterdam pushes on the exact lever Fusaka could not reach.

The Forecast: Three Scenarios Through 2027

Put the pieces together and a forecast takes shape. The base case assumes BPO3 lands sometime across 2026 and 2027 as blob demand catches up, Glamsterdam ships close to schedule, and no sustained demand shock saturates blobs. Under those conditions, routine L2 transfers drift from low-single-digit cents toward tenths of a cent, and complex DeFi interactions that cost a dime today settle for a few cents. The table below sketches the range.

The bear case is the one most forecasts ignore. Because the EIP-7918 floor is pegged to L1 execution, a busy, expensive Ethereum mainnet drags rollup floors up with it. A cheap L2 depends on a cheap, uncongested L1, which is one more reason Glamsterdam’s gas repricing matters far beyond L1 users. And if a genuine demand wave, a viral app, an AI-agent explosion, or a payments boom, fills blobs faster than BPOs add them, blob fees rise off the floor and the compression pauses, or reverses, until the next parameter bump. Blockspace has a long habit of expanding to fill whatever room it is given.

The base case is the center of gravity for a reason. It requires nothing heroic: BPO forks continue on the cautious cadence developers have already signaled, Glamsterdam ships within its stated second-half-of-2026 window, and blob demand keeps rising but stays a step behind the capacity each fork adds. That is roughly the pattern of the last two years extended forward. The bull case needs everything to break right at once, including a smooth BPO3 and BPO4 and a persistently quiet L1; the bear case needs a genuine demand shock or a stalled roadmap. Absent a catalyst in either direction, orderly compression toward tenths of a cent is the path of least resistance.

What the Models Say

Analysts have started to formalize the trend. A 2026 study modeling transaction costs across Ethereum mainnet and its L2s, posted to arXiv, fits fee trends to gas-limit and blob-capacity changes and projects continued decline. The paper reports that L2 median fees had already fallen more than 95%, from roughly $0.05 to about $0.0015, and forecasts that leading rollups will undercut Solana-level median fees around late 2026 on the current trajectory.

The methodology deserves respect even if the point estimates do not deserve full faith: it treats each gas-limit and blob increase as a measurable input and extrapolates from the observed response. The weakness in any such model is that it assumes demand stays polite. That is the single biggest reason to hold these forecasts loosely, because the history of cheap blockspace is a history of new demand rushing in to consume it. A model that nails the supply curve can still miss badly on price if it mis-reads the demand curve, and demand is the harder half to predict.

Who Wins When a Transaction Costs Nothing

Sub-cent fees are not an end in themselves; they change what is economically possible on-chain. Three categories of activity benefit most, and one of them could saturate all the new capacity.

Payments and remittances. When a transfer costs a hundredth of a cent, on-chain settlement competes with card networks on price for the first time. Stablecoin payments on Base and Arbitrum, already the cheapest large rollups, become viable for sub-dollar transactions that would have been laughable at 2021 gas prices. Cheap fees do not just save money on existing activity; they unlock use cases that were never worth doing before.

DeFi and trading. Cheaper execution lowers the size at which a swap or a rebalance makes sense, which matters for automated market makers whose economics depend on fee-sensitive flow; our explainer on how automated market makers work shows why gas cost sits underneath every AMM design decision. Cheaper blockspace also means arbitrage and liquidations fire more often and closer to the margin, tightening on-chain prices.

Machine-to-machine activity. The category that could actually saturate this new capacity is autonomous software. As we explored in our look at AI agents on-chain, agents that pay per action need fees measured in fractions of a cent to transact thousands of times a day. If that thesis plays out, agents are precisely the demand wave that could push blob usage back into the BPO throttle’s teeth and slow the compression a model would otherwise project forever.

There is a user-experience dimension too. Smart accounts, enabled by EIP-7702 on Ethereum and its rollups, make it possible to sponsor gas, batch actions, and pay fees in stablecoins; we covered the rollout in our piece on how smart accounts remade wallets. Combine near-zero fees with gas abstraction and the fee stops being something most users ever see, which is arguably the point of the whole exercise.

The Value-Accrual Problem: Cheap Fees, Thin Burn

Every dollar a rollup does not pay to Ethereum is a dollar that does not get burned. Under EIP-1559, base fees, including the blob base fee, are burned, removing ETH from supply. When blobs are nearly free and rollups post barely anything to L1, the burn thins, and the argument that L2s free-ride on Ethereum grows louder. Cheap fees for users and healthy revenue for Ethereum are not automatically the same goal.

The numbers are stark. Fidelity Digital Assets, analyzing Fusaka’s value-accrual effects, noted that Base paid a total of about $5.2 million in blob fees over the prior year while earning roughly $94 million in revenue from user transaction fees. That gap, cheap inputs from Ethereum against healthy sequencer revenue for the rollup operator, is the core of the tension. Fidelity estimated that EIP-7918’s floor would have added meaningful revenue back to ETH holders, on the order of an extra $30.6 million from Base alone over a year at its blob rate, at roughly $6.02 in additional cost per blob. That is part of why the floor was designed the way it was: it is a value-capture mechanism as much as a stability mechanism.

Rollup operators, for their part, are increasingly explicit that sequencing is a business. When Offchain Labs and Robinhood unveiled a Robinhood-branded Arbitrum chain, co-founder Steven Goldfeder framed it directly, telling crypto.news that as enterprise adoption accelerates, Arbitrum is ready to capture revenue. The subtext of the entire fee-compression story is that lower user fees do not necessarily mean lower operator margins; they can mean far higher volume against a very thin per-transaction take. The winners in a near-zero-fee world are the operators who own the flow.

The Demand-Side Lever: Data-Availability Competition

Fees are set by supply and demand, and Ethereum no longer controls the demand side alone. Rollups can post their data to alternative data-availability layers, Celestia and EigenDA among them, that undercut Ethereum blobs on raw price. Blockworks, examining Ethereum’s blob economics, put the gap at roughly 55x, with Ethereum blob data around $3.83 per megabyte against Celestia near $0.07.

This is a double-edged lever. If rollups migrate data off Ethereum to chase the cheapest byte, Ethereum’s blob demand and its burn fall further, but users may see even lower fees. If Ethereum’s expanding blob capacity and the EIP-7918 floor keep its data competitive on security-adjusted terms, demand stays home and the burn holds up. Either way, the price a rollup pays for data is now a contested market rather than a monopoly, and that competition is itself a force pushing fees down. The fight over who pays for blockspace is not unique to Ethereum, either; Bitcoin is having its own version, as we covered in Bitcoin’s blockspace war over Taproot and BIP-110. Contested blockspace is becoming the norm across every major chain.

The Wildcard: When Ethereum L1 Reclaims Execution

The tidiest forecasts assume the L1-versus-L2 division of labor holds: rollups do execution, Ethereum does data and settlement. Vitalik Buterin’s Lean Ethereum roadmap, outlined in a July 4, 2026 post, complicates that assumption. Buterin described it, as reported by The Defiant, as “the third major iteration of Ethereum in the same way that the Merge was the second,” a three-to-four-year overhaul in which “almost every major piece of the protocol will be replaced,” with gas-limit, blob, and slot-time changes stacking up over “roughly the next five years.”

One detail bears directly on fees. Buterin noted that rewriting an ERC-20 token to use a new UTXO-based storage design currently under exploration could cut its transaction fees by more than 10x. If L1 execution becomes dramatically cheaper and faster, some activity that fled to rollups purely for cost reasons could come back, which changes the demand picture for both L1 gas and L2 blobs. Layer on proposals like exponential gas-limit growth (EIP-7938 has floated raising the limit an order of magnitude over a couple of years), and the assumption that L1 is the permanently expensive settlement layer starts to look less permanent. For a fee forecast, that is a genuine wildcard: it could accelerate compression on L1 while muddying who captures the underlying demand.

What Could Break This Forecast

Forecasts of ever-cheaper fees share one blind spot: they assume demand behaves. It rarely does. The clearest risks to the compression thesis are concrete and worth watching.

• Demand outruns capacity. A viral consumer app, an AI-agent surge, or a payments boom could fill blobs faster than BPO forks add them, lifting blob fees off the floor until the next bump.
• BPO3 stalls. Core developers have said they will not push blob counts higher until real demand justifies it; if usage stays soft, the throttle stays put, and so do fees.
• An expensive L1. Because the EIP-7918 floor tracks L1 execution, a congested, high-base-fee Ethereum drags rollup floors up. Macro risk-on episodes that jam mainnet, the kind of volatility that can follow a surprise from Chair Warsh’s Fed (see our FOMC reaction coverage), can raise L2 floors indirectly.
• Sequencer centralization and MEV. Most rollups still run a single sequencer; margins, censorship risk, and MEV extraction all sit with the operator, and none of that is guaranteed to pass through to users as lower fees.
• Data-availability fragmentation. If rollups scatter across cheaper DA layers, Ethereum’s fee and burn dynamics shift in ways that make any single-chain forecast fragile.

The Bottom Line and What to Watch

The compression is real, it is engineered, and it is not finished, but it is bounded. L2 fees will very likely keep falling through 2027, with routine transfers heading toward tenths of a cent in the base case and complex interactions settling into the low cents. They will not hit zero, because Ethereum built the EIP-7918 floor precisely to stop that, and because a cheap L2 ultimately depends on a cheap L1.

The signals that will tell you which scenario is playing out are easy to track:

• A locked BPO3 date. Still unannounced as of mid-2026; when it lands, and at what blob target, it sets the next leg of supply.
• Glamsterdam mainnet activation and the realized effect of the EIP-7904 repricing versus the projected ~78% cut.
• Blob utilization. If average blobs-per-block starts pressing the target, demand is catching capacity and the floor stops being the binding constraint.
• The burn and the value-accrual debate. Watch whether ETH holders push for more of the floor’s revenue, and whether rollups keep their data on Ethereum or defect to cheaper DA.

The first two years of the rollup era were about proving that fees could fall. The next two are about discovering where the floor sits, and arguing over who keeps the difference.

Frequently Asked Questions

How low will Ethereum L2 fees go in 2026 and 2027?

In the base case, routine transfers on major rollups drift from low-single-digit cents toward tenths of a cent through 2027, while complex DeFi interactions fall into the low cents. The exact path depends on whether BPO3 ships, whether Glamsterdam lands on schedule, and whether demand stays below blob capacity. Fees will not reach zero, because the EIP-7918 floor ties blob prices to L1 execution costs.

Why can’t L2 transaction fees reach zero?

Fusaka’s EIP-7918 sets a reserve price that pegs the blob base fee to a fraction (about one-sixteenth) of the L1 execution base fee, so posting data can never be much cheaper than ordinary L1 gas. Before this floor existed, the blob base fee spent long stretches at effectively zero; the floor deliberately prevents that, which is why truly free L2 transactions are not on the table.

What is a BPO fork and how does it lower L2 fees?

A Blob Parameter Only (BPO) fork is a lightweight network upgrade that changes only the blob target and maximum per block, without a full hard fork. BPO1 in December 2025 raised the target and max to 10 and 15, and BPO2 in January 2026 took them to 14 and 21. More blob capacity means rollups compete less for data space, so blob fees, and the L2 fees built on top of them, fall.

How does the Glamsterdam upgrade affect L2 fees?

Glamsterdam, in final testing during 2026, reprices gas (EIP-7904, which its backers project could cut execution fees by roughly 78%), adds parallel execution via block-level access lists (EIP-7928), enshrines proposer-builder separation (EIP-7732), and lifts the gas limit toward 200 million. Cheaper L1 execution lowers the cost rollups pay to settle and reduces the EIP-7918 floor that anchors L2 fees.

Which Ethereum L2 has the lowest fees in 2026?

Among the large rollups in mid-2026, Base is typically the cheapest for routine transactions, with Arbitrum One and OP Mainnet slightly higher; validity rollups like zkSync Era, Linea, and Scroll are competitive. Exact rankings shift daily with L1 gas and blob demand, so check a live tracker such as L2BEAT before relying on any single number.

Written by Marcus Okafor, senior markets editor at HOGE Wire, covering Ethereum scaling, rollups, and market structure.