Ethereum’s layer-2 (L2) centric roadmap has received a number of criticisms regarding its “extractiveness” on ETH. Indeed, the Dencun upgrade and the rise of L2s have resulted in the reduction of total fees to three year lows, potentially signaling the end to ETH’s “ultrasound money” narrative. However, we posit that these criticisms are largely misplaced and are a byproduct of successful blockchain scaling – outcomes that would have occurred regardless if scaling was done on the L1 or L2. 

Specifically, we find that the reduction in average transaction cost has outpaced new demand for blockspace, resulting in a drop of total fees. This is despite the number of ETH-based transactions increasing more than six-fold since the start of 2023. Additionally, we demonstrate that onchain trading activity – the historic (and current) largest driver of onchain fees – has had a reduced impact on fees in recent years despite occupying a larger proportion of Ethereum’s blockspace. In our view, these two shifts have been the primary contributors to the drop in ETH burn rate. 

This does not preclude L2 tradeoffs entirely, however. Transaction fees on L2s do not include EIP-1559 style burns and sequencers are able to capture some profits based on the differences between L2 fees and L1 rent. Furthermore, the introduction of separate blob pricing may have caused a sudden glut of cheap blockspace that may not be saturated in the near term. Despite this, we think that the high throughput and low fee environments on L2s could encourage new suites of applications that further drive ETH utility. Moreover, we believe that these L2-specific tradeoffs are minor compared to other sources of flows like staking.

A Scaling Success Story

The majority of Ethereum’s activity now occurs on L2s, which is a testament to the ongoing technical success of its scaling roadmap in our view. At the start of 2023, 1M daily transactions occurred on L2s, accounting for 35% of all ETH-denominated transactions. By early 2024, this figure surpassed 5M daily transactions (63%) and further increased to 11M daily transactions (89%) as we go to publish.

The effects of the Dencun upgrade have been significant, with the total L2 transaction count immediately doubling from 5M to 10M in the weeks following its March 2024 release. Amid this L2 growth, however, the throughput on mainnet Ethereum has remained unchanged. This has been reflected in its relatively flat transaction count, depicted in gray in Chart 1. Ethereum’s gas limit has been constant since August 2021 while L2s have been constantly expanding.

Nevertheless, the total ETH spent on transaction fees has been decreasing despite the exponential growth in transaction counts. ETH transaction fees have reached three year lows with an average spend of Ξ1.0K per day in July 2024. In comparison, the next previous low during this timeframe (October 2023) exceeded this level two-fold with an average Ξ2.0K spent per day. Given that an average of Ξ3.4K was spent on transaction fees in 2023, we estimate that there has been at least a Ξ1K reduction in daily ETH spent on transactions due to the availability of cheap blockspace (i.e. scaling). In fact, we believe this number may skew even higher due to the exponential effects of chain congestion on pricing. (See Chart 2.)

Higher transaction counts for a lower average cost is the result of successful scaling by definition. That said, the importance of ETH “revenue” to the asset is not lost on us. As we head to publication, 92% of all ETH spent on transaction fees still occurs on mainnet Ethereum, despite having only 11% of the transaction count. This disproportion in fee structure embodies the massive cost reduction of L2s, which are more than two orders of magnitude cheaper than the L1. On August 1, 2024 the median Arbitrum One and Base transactions were 1/166th and 1/277th the cost of the median Ethereum transaction respectively, for example.

Unanticipated Side Effects

The combination of increased L2 activity and cheap blob transactions (via EIP-4844 in Dencun) have meaningfully changed ETH’s fee-based demand profile. In particular, the expansion of L2 capacity has seemingly outpaced that of demand, to the extent that the overall ETH spent on fees has dropped. Ironically, success in scaling Ethereum blockpace (by technical definition) now seems bearish for ETH since the reduction in fees could also reduce the ETH buying required for user onboarding and fees. 

That said, we importantly think this is a signal that the scaling roadmap is working as intended, which is a distinct conversation from L1 and L2 scaling tradeoffs. In our view, there is currently an excess supply of cheap blockspace, which is reflected in the reduction of average daily ETH spent on fees since 2021. (See Chart 4.) We think it’s likely that this trend will continue as onchain activity continues to shift towards low cost L2s.

We want to emphasize that the drop in total transaction fee spend does not imply that the L2-centric roadmap is the direct cause of the lowered transaction fee spend. We think that any scaling roadmap that increased blockspace faster than demand would result in a net reduction of total fees spent. Chart 5 demonstrates this by plotting the relationship between fees and transaction counts (in blue) fixed at the 2021 peak where the average transaction fee * daily transaction count = 9460 (i.e. the average daily fee). Scaling transaction counts and fees along this curve would keep fee spend constant with historic highs. 

However, it is apparent that the transaction count from 2022 onwards has not sufficiently increased to offset the reduction in average transaction cost. This is not a problem specific to L2 architecture in our view, but instead results from a lack of transaction demand to offset the increase in cheap blockspace. That is, an alternative roadmap that had similarly scaled the L1 would have likewise resulted in a reduction in overall transaction fee spend.

In fact, 2021 appears to be an anomaly when plotting lines of best fit through the 2020-24 averages. Finding a line of best fit with a similar to the x*y=k format (with a weighting bias towards more recent data) yields a curve of average transaction fee * daily transaction count = 4186 that is denoted in green in Chart 5. The computed value of Ξ4186 daily spend is more than 40% higher than the YTD average daily fee of Ξ2897, however, which could indicate that this may still be an overestimate. In other words, total transaction fees may continue to drop further if Ethereum blockspace continues expanding faster than transaction counts keep up. 

An alternative power law fit (shown in red) may provide a more accurate depiction of the scaling fee structures in our view. This assumes a nonlinear relationship between transaction costs and transaction activity. It has a lower mean squared error (MSE) than the constant product function fit on the dataset, and also scales down transaction costs faster than transaction counts, which is in line with observed behavior as well as plans for further L2 scaling. Based on this fit, reaching 20M average daily transactions would yield an average total daily fee spend of Ξ2443. This phenomenon is a byproduct of accelerated scaling – not one that is unique to an L2 roadmap.

The Primary Driver of Fees

An increase in cheap blockspace does not fully explain the reduction in total transaction fee spend, however. The transformation in the relationship between onchain trading activity and transaction costs on Ethereum has morphed in recent years. The initial dramatic increases of Ethereum fees in the summer of 2020 occurred in lockstep with the rise of DEX transactions. That is, the rise in fees were primarily the result of an increase in opportunistic trading activity. Chart 6 depicts the relationship between the percentage of total Ethereum transactions arising from decentralized exchange (DEX) activity in blue versus the total transaction fees paid on the network in gray.

Each time there was a surge in the DEX transaction ratio (the percentage of total transactions that were DEX trades), the total network fees also increased. That said, the impact of the DEX transaction ratio on network fees has been tapering off since 2022. We capture this in the dashed red line that depicts the DEX transaction ratio, but gradually scaled down by a factor of 0.6 over the time series. The closer correlation between the downscaled ratio and total fees indicate that an increase in the DEX transaction ratio still impacts total fees, but less so than previously.

Similarly, it’s important to note that the DEX transaction ratio is currently near all time highs, above even the peaks of 2020 and 2021. This creates an interesting phenomenon where the proportion of DEX-based activity on Ethereum is near all time highs, but transaction fees remain near historic lows. In fact, Uniswap remains the single largest driver for Ethereum transaction fees, with the Uniswap Universal Router responsible for driving nearly 12% of all Ethereum fees in 2Q24. (See Table 1.) 

Sources: Dune, Etherscan and Coinbase. Period measured is from 2024-04-01 to 2024-06-30. Excludes simple ETH transfers and contract creation.

We believe that a key (but often overlooked) factor for the reduced impact of trading activity on transaction fee variability is that the trading environment on Ethereum has significantly matured. This has reduced the opportunity space for certain “simple” MEV bidding wars that would historically drive up onchain fees for priority transaction ordering. (Some of this has shifted to block proposer payouts instead.) In addition, highly volatile memecoin activity has migrated to chains like Solana or Base, further reducing the arbitrage and MEV opportunity space. Low default slippage settings in many DEXs and a focus on MEV-protected swaps such as CoW Swap have also reduced the baseline importance of priority fees relative to the early days of Ethereum trading.

As a result, we think that the comparisons often made between total fees spent on Ethereum and Solana overly focus on scaling strategy and tend to miss the critical question of fee sustainability, which we think is unclear (particularly for the latter network). For example, over 95% of the non-vote transaction fees spent on Solana are a result of priority fees, which are likely linked to time-sensitive trading transactions. Furthermore, of all non-vote transaction fees, between 25% and 45% are spent on failed transactions. How this proportion of transaction fees evolves over time will depend heavily on Solana’s roadmap, which may diverge significantly from Ethereum with regards to MEV and priority fee calculation. That is, Solana may be moving toward its local maxima of fee spend like 2020-21 Ethereum. Or alternatively, the full implementation of its local fee markets could support, and even compound, its current dynamic of high priority but low base fees. Regardless, we again emphasize our view that the reduction in total transaction fees may be inevitable in any successful blockchain scaling roadmap, regardless of L1 or L2 focused designs.

Ultrasound Money

This thought extends to the “ultrasound money” narrative for ETH, which has been the primary victim of reduced fees. Since the Merge in September 2022, ETH has been a net deflationary asset due to high levels of the base fee transaction burn (introduced in EIP-1559 more than a year earlier). Between the Merge and Dencun, the ETH supply had reduced by more than Ξ440K from the burn mechanism. 

This trend reversed shortly after Dencun. Since the upgrade, the supply of ETH has increased by nearly Ξ200K. At the current pace of inflation, the cumulative supply change of ETH post-Merge would turn net inflationary before year end. This trend may only accelerate further if the ETH staking ratio continues to increase (and net issuance by extension). 

We don’t think this is a problem driven primarily by L2 scaling, however. While L2 transaction fees do lack the EIP-1559 burn mechanism, the total fees spent on L2s are comparatively inconsequential. That is, even if 100% of L2s fees were burned, ETH would remain inflationary. Instead, the change in inflation rate seems to be driven by lower mainnet fees more broadly (a scaling challenge not limited to L2 architecture) as well as a rise in the ETH staking ratio which has accelerated overall issuance speed. In our view, conversations around the viability and importance of “ultrasound money” is one that should be separated from that of the L2 scaling approach. Instead, it relates more to the growing staking ratio as well as more general scaling economics.

Sequencer Profitability

A last major critique of the L2 scaling roadmap is the value capture of L2 sequencer operators. In principle, sequencers are able to capture profit as the difference between the total transaction fees spent on the L2 minus the fees paid to the L1 as “rent”. This ETH-denominated profit does not undergo burn. Instead, it is likely to be sold (in part) to fund operations, resulting in additional sell pressure on the asset.

The Dencun upgrade further increased the opportunity for sequencer profitability by reducing the average costs paid by L2s to mainnet Ethereum (i.e. the “rent”) by more than 90%, or approximately Ξ200-300 per day on average. (See Chart 7.) Despite this, rollups have surprisingly seen mixed results regarding onchain profitability metrics, with certain rollups even seeing reductions in median daily sequencer profitability post-Dencun. 

The reason is that most of the “rent” cost reduction from Dencun has in fact been passed onto L2 users in the form of lower fees. The median transaction fees for L2s have compressed below the $0.01 cent range for leading rollups like Base, making them competitive in cost and speed to high performance L1s like Polygon PoS and Solana. (See Chart 8.)

That said, we think that the reduction in L1 “rent” to the tune of Ξ200-300 is the most direct measure of foregone transaction fees from the L2-specific approach to scaling (as opposed to a hypothetical L1 scaling strategy). The introduction of blob transaction types, which currently remain at the minimum price levels due to excess supply, have resulted in a total daily fee reduction of about 10%. While this number may be somewhat impactful relative to total fees, we think it holds comparatively little significance relative to other large drivers like the increasing staking ratio, net issuance, and application growth, which we discuss below. 

Growth from Utility

In the long term, we think that the average transaction fee is likely to continue trending towards zero regardless of scaling architecture. With the exception of select time sensitive transactions (e.g. trading) that command higher priority fees, the absolute reduction of fees via scaling is a desirable outcome for moving users onchain – even at the cost of reduced total fee spend. 

This, however, begs the question of fundamental drivers for native L1 assets in the long term if total fee spending tapers off. Onboarding fees are likely to remain in our view (absent the mass adoption of smart wallets). We’ve previously written about how initial wallet funding amounts tend to remain anchored to fiat terms once transaction costs drop below a certain threshold. Primarily, however, we think that application-based utility may emerge as the largest demand driver. In fact, we think that the high throughput, low fee L2 environment will enable new forms of applications and native asset utility that were not previously feasible on the L1. 

There are early signs that this shift could be happening. More than 4 million ether have been bridged to L2s via canonical bridges. This is more than the combined ETH spent on fees between January 2022 and August 2024 (at about Ξ3.7M). Much of this bridged ETH is currently deployed into various protocols, though DeFi applications currently still serve as the largest liquidity sinks. Aave on Arbitrum currently holds over Ξ200K for example.

At the same time, the utility of ETH on mainnet protocols remains strong. An additional Ξ4.7M has been added to the supply of staked ETH YTD (Ξ28.8M to Ξ33.5M), driven in part by Eigenlayer and restaking. Long standing DeFi protocols have also continued to attract new ETH inflows, with Aave V3 on Ethereum seeing an additional Ξ500K deposited to the wrappedETH vault in 2024 to reach a current total of more than Ξ2.8M in deposits across different forms of ether. (See Chart 10.) The net growth of ETH locked in Ethereum Aave V3 has been similar to the entirety of ETH spent on transaction fees in 1H24 at Ξ524K.

ETH further serves as a quote currency for many onchain trading pools, which represents an important demand originator for ETH. A single Uniswap V3 pool on Ethereum (WBTC/ETH) holds over Ξ80K for example. We think that the native L1 asset will continue to play a critical role as the primary “credibly neutral” base asset of a chain, but it’s not yet clear precisely how this may play out in new blockchain verticals due to their nascency.

Conclusion: Are L2s Extractive?

Ultimately, we think the L2 roadmap may be slightly extractive to ETH demand relative to direct L1 scaling for two reasons: (1) there is no EIP-1559 style burn for L2 transactions, and (2) there is an opportunity for L2 sequencers to capture some profit via the delta between L2 transactions and L1 rent. 

However, we think that this “extractiveness” is a red herring for analyzing long term ETH supply dynamics. First, we do not think the record 2021 level fees were sustainable with general blockspace scaling and reduced MEV opportunities. This means that the reduction in total ETH fees and burn is not primarily a problem with the L2 approach, but rather one of scaling economics in general. Similarly, we think criticisms that the L2 approach handicaps the ultrasound money narrative are misguided – the growth in net issuance and overall blockspace are the primary drivers of net inflation in our view.

Current trends lead us to expect continued strength in ETH demand from other protocol-based avenues such as collateral in money markets or trading pairs in DEXs. Beyond DeFi, we think that the low fee, high throughput blockspace environment of L2s may lay the foundation for further ETH-centric applications still in the early stages of development.