Transaction Costs and Speed in the Ethereum Ecosystem: Scalability of the Mainnet and Layer 2s

Abstract

We study the evolution of transaction speed and fees from January 2024 through March 2026, comparing Ethereum Mainnet and its Layer 2 (L2) networks, as well as Solana and Polygon. Ethereum has undergone upgrades that have increased block size and blob count. These upgrades have doubled transactions per second (TPS) on both the Mainnet and the L2 networks. Mainnet median fees have fallen from over $2 to under $0.02, and L2 median fees have fallen by more than 95% from $0.05 to $0.0015. We forecast that Mainnet median fees will converge with Solana’s in August 2027, but TPS will remain below 100 until 2034. The L1 Strawmap, proposing EIP-7938, a potential exponential increase in the gas limit, brings the Mainnet to only 100 TPS in January 2028. With continued blob expansion, L2s will surpass Solana’s TPS in March 2029 and have lower median fees by October 2026.

1. Introduction

Transaction speed and cost are defining competitive dimensions of all financial markets. Equity markets have long set the standard, offering free commissions and millisecond execution for retail orders, https://www.fidelity.com/trading/execution-quality/overview, accessed on 1 September 2026. Recent Ethereum blockchain upgrades are enabling the Mainnet and its adjacent Layer 2 networks (L2s) to close the gap with traditional financial markets and the fastest blockchains, most notably Solana.

The two key improvements to the Mainnet and its L2s are increased gas limits and the introduction of blobs. A gas limit is defined as the maximum amount of computational work (gas) permitted into a block. The number of transactions that can be contained in a block on the Mainnet is constrained by the gas limit value. Binary Large Objects (blobs), introduced in the Dencun upgrade in March 2024, are the mechanism by which L2s post their compressed transaction data to the Mainnet.

We compare transaction speeds and fees across Ethereum and its L2s with Solana and Polygon, from January 2024 through March 2026. This window captures the impact of the three most recent Ethereum protocol upgrades: Dencun in March 2024, Pectra in May 2025, and Fusaka in December 2025. Two blob-parameter-only (BPO) upgrades, which increase the number of blobs permitted per block, are activated in December 2025 and January 2026. The period also corresponds with significant increases to the block gas limit, rising from a median daily gas limit of 30 million in January 2024 to 60 million by December 2025.

The daily average Mainnet transactions per second (TPS) rose from 13.67 in the first quarter of 2024 (2024 Q1) to 25.78 in the first quarter of 2026 (2026 Q1). The L2 totals rose from 78.60 to 226.92 TPS over the same time period. Solana TPS did not grow as quickly, rising from 804.72 in 2024 Q1 to 1303.46 in 2026 Q1; Polygon rose slightly faster than the Mainnet, rising from 45.72 TPS in 2024 Q1 to 91.51 TPS in 2026 Q1.

There is established literature covering Ethereum transaction speeds, and there are different methods of defining transaction time. Ref. [1] finds no relationship between the amount of gas and gas price and the delay time in confirming transactions. Studying EIP-1559 [2], which introduced the base fee, Ref. [3] finds that the EIP reduced the time between when a transaction enters a mempool and when it is mined. Additionally, they find that, when Ether’s price is volatile, wait times increase. Conversely, Ref. [4] finds that recent gas prices are the most important feature in explaining block inclusion. Our paper differs from this literature in that we measure transaction speed after block formation.

Ref. [5] examines the ability of various clients, including Nethermind, to support higher gas limits and concludes that expanded block space increases TPS. Ref. [6] finds that raising the block gas limit increases the number of transactions per block even at shorter block formation times.

Mainnet speeds are increasing with the block gas limit. We estimate that a ten-million-unit increase in the gas limit is associated with 3.52 more TPS. The Mainnet, even with the gas limit at more than four times the current 60 million, is not projected to reach 100 TPS until February 2034, less than one-tenth of Solana’s current speed.

There are ongoing proposals to increase the gas limit as outlined in the L1 Strawmap, https://strawmap.org, accessed on 1 September 2026. The Ethereum Improvement Proposal (EIP)-7938 [7] presents a hundred-fold gas limit increase for four years beginning in 2027. Applying this to our current Mainnet projection, we predict the Mainnet to reach 100 TPS by January 2028, six years earlier than our baseline trend. Ethereum Foundation researchers such as [8] have raised concerns about the growing storage requirements of the network’s history (“state bloat”) as well as the impact on validator centralization.

Related literature looks at speed on Layer 2 networks. Ref. [9] studies time to finality across the Mainnet, L2s (Arbitrum and Optimism), and Ethereum Virtual Machine (EVM) compatible blockchains (Avalanche, Binance Smart Chain, and Harmony). Ginting finds that Arbitrum has the fastest time to finality at 1.7 s for simple transactions and 2.1 s for smart contracts. Ref. [10] notes that the combination of faster block times and private mempools reduces sandwich attacks: “across all L2s, we find no evidence of sustained or economically meaningful sandwich attacks.”

There is also a considerable amount of literature that uses simulations to analyze both fees and speed. Simulations in [11] shows that when 90% of transactions are routed to L2s from the Mainnet, the throughput on the Mainnet can increase by up to 20%. Ref. [12] show that EIP-4844 can reduce gas costs for L2 rollup submissions by up to 52%. Throughput can also nearly be doubled in larger blob sizes. They do find some saturation when blob sizes exceed 96 kb.

We find that blob parameter increases on the Mainnet have increased the speed of the L2s. We estimate that each additional target blob is associated with 10.75 additional L2 total TPS. With continued blob expansion at the BPO increment rate, L2 total TPS is projected to surpass Solana’s predicted TPS by March 2029, converging at 1820 TPS. We recognize that L2 TPS has slowed in 2026, however.

Our manuscript then turns to transaction costs. Transaction fees on the Mainnet have fallen substantially, with a median transaction fee of $3.786300 in 2024 Q1, declining 99.68% to $0.012283 in 2026 Q1. The average fees of the three highest-volume L2s: Arbitrum, Base, and Optimism, which we call the L2 leaders, declined 99.16% from $0.180219 in 2024 Q1 to $0.001512 in 2026Q1. While Solana’s fees also fell, the percentage decrease was slighter. Median fees fell from $0.000687 in 2024 Q1 to $0.000496, a decline of 27.80%. Polygon fees are an exception to this trend, rising 19.62% over this period from $0.005742 to $0.006868.

Studying fees from November 2017 through January 2019, Ref. [13] finds that the effect of network congestion on Ethereum transaction fees is statistically insignificant until blocks reach 90% capacity. On the supply side, they find that increasing the block’s capacity limit is associated with lower fees. Studying the Ethereum network’s transition from Proof-of-Work (POW) to its current transaction method, Proof-of-Stake (POS), Ref. [14] finds that Ethereum fees in USD increase nonlinearly with mempool count. In addition, it was found that fees decreased with the change from POW to POS. Using data from January 2021 through December 2022, Ref. [15] confirms that a higher block gas limit is associated with lower transaction fees. Studying the period around the Dencun upgrade, Ref. [16] finds that EIP-4844 [17]. increased rollup usage on Ethereum while lowering rollup transaction fees. Ref. [18] shows that after Pectra, fees on Arbitrum and Optimism show a weaker dependence on Mainnet fees.

We also compare the Mainnet and Layer 2 networks to competing blockchains. Ref. [19] compare Solana and Ethereum using two submission patterns: sequential submission, where transactions are sent one after another, and parallel submission, where they are sent simultaneously in batches. They measure TPS, latency, and gas fees and find that Solana outperforms Ethereum across all three metrics in both modes. Ref. [20] finds that lower costs on Arbitrum and Polygon attract smaller orders away from the Mainnet.

Ref. [21] cautions that speed can have drawbacks. They study the period from November 2023 through March 2024, and find that Solana has a transaction failure rate of approximately 20%, whereas Ethereum has only 0.1%. Solana also has a 7.6% zero-value transfer rate, while Ethereum has a 0.66% rate.

As with speed, we find that gas limits are a key factor in declining transaction costs. We estimate that every five-million-unit increase in the gas limit is associated with a 59.2% decline in median transaction fees. The Mainnet is projected to converge with Solana’s predicted median transaction fee in August 2027.

For the L2 leaders, each additional target blob is associated with a 4.3% decline in median transaction fees. Assuming continued blob expansion at the BPO increment rate, we project that the L2 transaction fees will converge with Solana’s predicted fee in October 2026.

Our paper contributes to this literature in three ways. First, we document the joint effects of the Dencun, Pectra, Fusaka, BPO, and gas limit upgrades on Mainnet transaction speed and fees from January 2024 through March 2026. To our knowledge, no prior work covers all these upgrades. We extend the cross-chain comparison literature by jointly studying the Mainnet, L2s, Solana, and Polygon. We also provide quantitative forecasts of Mainnet and L2 speeds under extrapolated trends and the EIP-7938 proposal in the February 2026 L1 Strawmap, which has not yet been empirically studied.

We then examine the effects of higher block gas limits and blob targets on fees. Our analysis also predicts dates at which the Mainnet and Layer 2s will reach lower fees than Solana.

This study examines whether observed changes in Ethereum’s available block and blob capacity are drivers behind transaction throughput and fees. We recognize that realized performance also depends on network demand and transaction composition. Capacity expansions may remain underutilized. As [22] notes, protocol capacity should be interpreted as an enabling condition. Effective use depends on technological development and adoption, user demand and the organization of the entire ecosystem.

The remainder of our paper is organized as follows. Section 2 describes the Ethereum network architecture and documents the sequence of recent protocol upgrades. Section 3, Section 4 and Section 5 cover Polygon, Solana, and the L2s. Section 6 reports transaction speed and fee comparisons across blockchains. Section 7 and Section 8 estimate the effects of gas limits and blobs on speed and transaction costs. From these models, we produce forecasts for both the Mainnet and the Layer 2s for TPS and fees. Section 9 concludes.

2. Ethereum Network Improvements

The improvements covered in this section target two outcomes for the network: increasing TPS and decreasing fees. Section 2.1 covers the hard forks that introduced and expanded blobs; Section 2.2 presents blob usage data. Section 2.3 documents the increases in the gas limit. Section 2.4 reports Mainnet transaction counts and fees. Section 2.5 provides information on future upgrades.

2.1. Protocol Forks

On 13 March 2024, at 01:55:35 UTC, epoch number: 269,568, Ethereum launched Cancun-Deneb, “Dencun”, https://ethereum.org/ethereum-forks/, accessed on 1 September 2026. Cancun contains execution-layer improvements, while Deneb contains consensus-layer upgrades. For our analysis, the key change was EIP-4844 [17], which introduced blobs. Blobs are cheaper than standard transactions because the data is only stored temporarily and not processed through Ethereum’s Virtual Machine (EVM), the computation engine for the network. This provides a cheaper way for L2s to post their transaction data to the Mainnet, https://chain.link/article/eip-4844-blob-storage, accessed on 1 September 2026. The target number of blobs in Dencun was three, and the maximum number of blobs was six. These numbers have grown progressively since their introduction.

Prague-Electra,“Pectra”, was released 7 May 2025, at 10:05:11 UTC, epoch number 364,032, https://ethereum.org/ethereum-forks/, accessed on 1 September 2026. The upgrade combines execution-layer changes (Prague) and consensus-layer changes (Electra). The key updates for our analysis are EIP-7840 [23], which simplified the process for blob updates, and EIP-7549 [24], which increased the target number of blobs to six and the maximum to nine. The blob count increase under Pectra was explicitly intended to provide short-term TPS improvements [25].

Following Pectra, Fulu-Osaka, “Fusaka”, was released 3 December 2025, at 21:49:11 UTC, epoch number 411,392. The name consists of Osaka, the execution-layer upgrade, and Fulu, the consensus-layer upgrade. https://ethereum.org/roadmap/fusaka/, accessed on 1 September 2026. Under Fusaka, EIP-7935 [26] increased the gas limit from 36 million to 60 million, and EIP-7594 [27] introduced Peer Data Availability Sampling (PeerDAS). PeerDAS allows the nodes to confirm blob data is present by checking only a sample of it, rather than the full dataset. This reduces the workload on each node.

For L2 users, the intended outcome was a reduction in transaction fees and higher blob TPS, https://ethereum.org/roadmap/fusaka/peerdas/, accessed on 1 September 2026.

2.2. Blob Capacity

Fusaka also introduced a mechanism to rapidly expand blob capacity independently of other protocol changes, https://ethereum.org/roadmap/fusaka, accessed on 1 September 2026. These are called blob parameter-only (BPO) forks. These differ from traditional hard forks because BPOs adjust only a small set of blob-related parameters, such as the target and maximum numbers of blobs. Traditional hard forks require coordination and more extensive protocol changes [28].

Since Fusaka went live, BPO-1 and BPO-2 have been activated. BPO-1 activated 9 December 2025, 14:21:11 UTC, at epoch 412,672, increasing the target blob count from six to 10, and the maximum from nine to 15. BPO-2 went live on 7 January 2026, 1:01:11 UTC, epoch 419,072. It increased the target blob count to 14 and the maximum to 21 blobs, https://www.optimism.io/blog/fusaka-is-live-scaling-optimism-and-the-superchain, accessed on 1 September 2026.

We are interested in studying how the target blob count impacts both the TPS and transaction costs of the L2s. The target blob count impacts the L2 fee calculation because, when the blob usage exceeds the target number of blobs, the fee increases [17].

Figure 1 plots the daily number of target and maximum blobs per block and the average number of blobs in a block from January 2024 through March 2026.

All three series begin at zero prior to the Dencun upgrade, at which point the target number of blobs jumps to three and the maximum to six. The target and maximum remain unchanged until the Pectra upgrade in May 2025, after which the target blob rises to six and the maximum to nine. The next jump is observed at BPO-1 in December 2025, raising the target to 10 and the maximum to 14, with BPO-2 shortly following in January 2026, raising the target further to 14 and the maximum to 21.

Throughout the entire sample period, the average blob usage, shown by the blue line, remains well below both the target and maximum, generally tracking around one to two blobs per block. While the average blob usage per block shows a slight uptrend, the difference between usage and the target and maximum blob counts continues to increase.

Figure 2 complements this result as it reports the percentage of blocks per day where the blob count exceeds the target.

The proportion of blocks with the number of blobs exceeding the target declined over the sample. From mid-2025 onward, the percentage of blocks exceeding the target fell to near zero and remained there through the end of the sample period.

2.3. Gas Limit Data

We are interested in studying how the gas limit affects both transaction costs and transaction speeds on the Mainnet. When a block’s gas usage exceeds the target gas limit, defined as half of the gas limit, the following block incurs a fee increase, https://ethereum.org/developers/docs/gas/, accessed on 1 September 2026. In terms of speed, increasing the gas limit increases the transaction count, enabling each block to handle more transactions. Ethereum Foundation developer, Marius van der Wijden, has documented this relationship, https://mariusvanderwijden.github.io/blog/2024/01/11/GasLimit/, accessed on 1 September 2026. Consistent with this, Ref. [5] finds that increasing the gas limit increases TPS.

Figure 3 reports the high, low, and interquartile range (IQR) of gas used per block within each month from January 2024 through March 2026, as well as the gas limit and the target gas limit. The IQR is defined as the difference between the 75th and 25th percentiles of daily gas used per block within each month.

From January 2024 through January 2025, the gas limit remained flat at roughly 30 million, and the monthly distributions of gas used per block were narrow and stable, with the IQR averaging 6.6 million.

As the median gas limit increased to 36 million from February through June 2025, the average IQR widened to 10.9 million. The IQR remained at a similar level of 10.1 million during the 45 million gas limit period from July through October 2025. Following the increase to 60 million beginning in November 2025, the average IQR widened further to 18.3 million through March 2026.

While the medians in Figure 3 are consistently below the target, under EIP-1559, it only requires a single block exceeding the target to increase the base fee for the following block.

Figure 4 reports the daily percentage of blocks in which gas used exceeded the target from January 2025 through March 2026.

Throughout our entire sample period, 37.7% to 48.5% of blocks exceeded the gas target on any given day. The percentage was lower and less volatile in early 2024, ranging from 39.9% to 46.2% from January 2024 through June 2024. There is no clear downward trend over the sample period, indicating that blocks exceeding the gas target persisted and remained consistent despite increases in the gas limit.

2.4. Mainnet Transaction Information

With the gas increases documented, we examine the transaction activity these changes have enabled on the Mainnet in Figure 5.

Mainnet activity remained relatively flat between 32.3 and 39.6 million transactions per month from January 2024 through April 2025. Beginning in May 2025 following the Pectra upgrade, transaction counts increased notably, rising to 70.01 million by January 2026 and remaining elevated through March 2026.

Figure 6 presents Mainnet median monthly transaction fees.

Mainnet fees declined with each major gas limit increase. We will explore this relationship further in Section 8.1.

2.5. Roadmap for Future Upgrades

In February 2026, the Ethereum Foundation introduced its L1 Strawmap, a roadmap of proposed protocol upgrades for the Mainnet, https://strawmap.org, accessed on 1 September 2026. The next protocol upgrade is expected to be Glamsterdam in 2026 Q3. The dynamics of this upgrade are expected to set the stage to reach a 200 million gas limit floor once implemented, https://blog.ethereum.org/2026/05/02/soldogn-interop-recap, accessed on 1 September 2026.

EIP-7938, currently proposed for inclusion in a future hard fork, would introduce a deterministic schedule that increases the gas limit by a factor of ten every two years for four years [7]. This trajectory aligns with the Strawmap’s target of one gigagas per second.

3. Polygon

Polygon is an Ethereum Virtual Machine (EVM)-compatible side-chain that can run smart contracts and decentralized applications written for Ethereum, https://ethereum.org/developers/docs/scaling/sidechains/, accessed on 1 September 2026. Operating as a side-chain, Polygon processes transactions on its own network and will periodically post a checkpoint to the Mainnet. This differs from L2s in that Polygon does not post its transaction data to the Mainnet to be validated.

Figure 7 displays Polygon monthly transaction counts from January 2024 through March 2026.

Transaction counts showed no sustained directional trend from January 2024 through April 2025, averaging approximately 106 million transactions per month. Beginning in May 2025, transaction counts trended upward, surpassing the prior sample high of 142.7 million by November 2025 and reaching 305.6 million by March 2026.

Figure 8 reports Polygon median monthly transaction fees in U.S. dollars.

Polygon had a sustained decline in fees beginning in May 2024. Fees surged in the first quarter of 2026. In February, total daily fees on Polygon surpassed Ethereum for the first time. Industry analysis has focused on the prediction market Polymarket https://polymarket.com/, as the key driver, https://cryptorank.io/news/feed/39df8-polygon-fees-surpass-ethereum-polymarket, accessed on 1 September 2026.

4. Solana

Solana serves as the primary performance benchmark for this paper. Solana is designed to handle high transaction speed on its base layer.

Solana’s actual speed is often overstated because a substantial portion of block space is required to validate transactions in Solana’s consensus scheme. Figure 9 displays Solana’s monthly transaction counts, separated into vote and non-vote transactions, in billions from January 2024 through March 2026.

Vote transactions make up an average of 71.5% of total transactions in our sample. Solana’s roadmap plans to open more block space for non-vote transactions through their Alpenglow upgrade. This upgrade, expected in 2026 Q3, will move all vote transactions off-chain, https://solana.com/news/solana-network-upgrades, accessed on 1 September 2026. If all of the 2026Q1 vote transactions are replaced by non-vote transfers, Solana’s TPS could rise to over 4500 TPS.

We will focus our comparisons across the blockchains on the non-vote transactions which we plot in Figure 10.

Transaction counts were relatively stable throughout the sample period, ranging between 1.72 billion transactions per month in September 2024 to 3.66 billion in March 2026. A notable decline is observed beginning in August 2025, falling from approximately 2.92 billion to approximately 2.23 billion by December 2025, before recovering to approximately 3.66 billion by March 2026.

Figure 11 presents Solana median monthly transaction fees. Solana again is a benchmark with the lowest fees of any chain other than Optimism.

Fees on Solana show a sustained downtrend from March 2024, but fees fell below levels from the first two months of 2024 only in February and March of 2026.

5. L2 Networks

There are two different classes of L2 networks: optimistic rollups and Zero-knowledge (ZK) rollups. All L2s execute transactions off-chain and post the compressed transaction data to the Mainnet. The difference between the two classes lies in their verification approach.

For optimistic rollups, it is assumed that the transactions posted by the L2 are valid. There is a 7-day window during which a transaction can be challenged for review by the Mainnet. Due to the immediate validity assumption, optimistic rollups can achieve higher TPS than ZK-rollups.

Rather than assume validity upon posting data to the Mainnet, ZK-rollups provide cryptographic validity proofs for each transaction batch posted to the Mainnet, which must be verified. This method is much more computationally intensive, limiting the TPS count.

Transaction Comparison

Figure 12 plots daily TPS for optimistic and ZK-rollups from January 2024 through March 2026. Optimistic rollups grew steadily over the period, rising from 29.67 TPS in January 2024 to 195.22 TPS for March 2026. ZK-rollups followed a different trajectory, reaching 33.57 TPS in March 2024 before declining sharply to around 7.82 TPS in March 2026.

Figure 13 includes all 20 L2 networks that collect fees in ETH. We do not include L2s such as Celo, https://l2beat.com/scaling/projects/celo, accessed on 1 September 2026, that do not price fees in ETH. Of the 20 L2 networks in our sample, nine were active from 1 January 2024. Eleven entered throughout the sample period, with the last network, Ink, joining in December 2024.

The combined L2 networks have outperformed the Mainnet in transaction count through the entire period. Moreover, Arbitrum and Base individually exceeded the Mainnet’s monthly transaction count for most of the sample period, with Base surpassing the Mainnet from April 2024 onwards and Arbitrum from January 2024 onwards.

Total L2 transactions grew from 135.6 million in January 2024 to 543.8 million in March 2026. Base grew quickly to become the dominant network throughout the period and is the primary driver of overall growth. Base also leads in DEX trading, as we show in Figure 14, where Base has a larger volume than all the other L2s. Since April 2025, Arbitrum has had the second-highest transaction count among the L2s. In Q1 2026, Arbitrum, Base, and Optimism are the three networks with the highest transaction counts.

Although Arbitrum, Base, and Optimism are optimistic rollups and EVM-compatible, their schemes are different. Optimism was publicly launched on 16 December 2021, https://messari.io/project/optimism/profile, accessed on 1 September 2026. In 2024, Optimism introduced OP Stack, a shared software system that was adopted by Base, Unichain and Zora, among others. Blockchains built on OP Stack form a connected network known as the Superchain, https://eco.com/support/en/articles/11779236-what-is-the-op-stack-architecture-and-superchain-explained, accessed on 1 September 2026.

Base was developed by Coinbase and launched in August 2023. Base has access to Coinbase’s existing users, enabling a smoother transition for users moving from Coinbase to Ethereum L2 applications. Specifically, decentralized applications (dApps) built on Base have access to Coinbase’s existing products and developer tools, including wallet infrastructure. Base was the first major network built using OP Stack, but as of May 2026 it has adopted its own software stack, https://blog.base.dev/next-chapter-for-base-chain-1, accessed on 1 September 2026. In addition to this change, Base has moved away from purely optimistic proofs to using ZK and Trusted Execution Environment proofs, https://blog.base.dev/multiproofs-on-base, accessed on 1 September 2026.

Arbitrum became publicly accessible on 31 August 2021, https://messari.io/project/arbitrum/profile, accessed on 1 September 2026. It consists of two networks, Arbitrum One and Arbitrum Nova. We only include Arbitrum One here because it handles more than 95% of the volume, https://defillama.com/chain/arbitrum-nova, accessed on 1 September 2026. On 8 January 2026, Arbitrum introduced a fee-smoothing mechanism intended to reduce price volatility during periods of high demand.

6. Blockchain Comparison

We now quantify the differences between the Mainnet, its L2s, Solana, and Polygon by comparing DEX volume and TPS from January 2024 through March 2026.

6.1. Volume

To analyze trading activity, we aggregate 2026 Q1 DEX volumes across all trading pairs and convert them into USD equivalents. This measure captures the total dollar value of DEX trades executed on each blockchain. We compute this metric for the Mainnet, Base, Solana, and Polygon. We group all the other L2s, which each have less than $50 billion of DEX volume, into another group, ‘Other Layer 2’. Figure 14 therefore represents the distribution of trading volume measured in USD across DEX on each blockchain.

Solana leads with 58% of total volume, followed by Ethereum at 22%. Base has the greatest volume percentage compared to all other L2s at 12%, leaving other L2s to make up 6% of volume. Polygon has the lowest volume at 1%. Combined, Ethereum and its L2 networks account for 40% of DEX volume.

6.2. Transactions per Second

We begin by comparing 2026 Q1 TPS to the same quarter back in 2024. To study L2 speed, we aggregate the daily TPS across all L2 networks introduced in Section 5 to obtain a combined L2 TPS measure. We refer to this as L2 Total TPS.

Table 1. Network speed: average daily transactions per second.

Network	2024 Q1	2024	2025	2026 Q1
Mainnet	13.67	13.59	16.61	25.78
L2 Total	78.60	124.00	227.08	226.92
Solana	804.72	819.28	1054.77	1303.46
Polygon	45.72	41.32	44.57	91.51

Note: Annual values are computed as the average of daily TPS across each calendar year. 2024 Q1 and 2026 Q1 values cover January 1 through March 31 of their respective years.

reports the 2024 Q1, 2024, 2025, and 2026 Q1 TPS values for Ethereum, L2 Total, Polygon, and Solana.

Solana remains the fastest network throughout the sample, averaging 804.72 TPS in 2024 Q1 and rising to 1,303.46 TPS by 2026 Q1. The L2 ecosystem has closed the gap with Solana, rising from 78.60 TPS in 2024 Q1 to 226.92 TPS in 2026 Q1. Polygon doubled its TPS rising from 45.72 in 2024 Q1 to 91.51 TPS in 2026 Q1. The Mainnet grew almost as quickly from 13.67 TPS in 2024 Q1 to 25.78 TPS in 2026 Q1.

Figure 15 visualizes the TPS of all networks monthly from January 2024 through March 2026.

All three networks show uptrends in TPS. We will extrapolate time trends in Solana. For the Mainnet, we will link TPS improvements to increases in the gas limit. For the Layer 2s, we will model their trend growth using blobs.

6.3. Transaction Fees

To present a representative fee measure, we focus on the L2 Leaders, defined as the mean of daily median fees across Arbitrum, Base, and Optimism. These three networks account for 79.39% of total L2 transactions and 18% of total DEX volume in Q1 2026.

Table 2. Median transaction fees in USD.

Network	2024 Q1	2024	2025	2026 Q1	Pct. Chg.
Mainnet	$3.786300	$1.736066	$0.164527	$0.012283	−99.68%
Solana	$0.000687	$0.001288	$0.001037	$0.000496	−27.80%
Arbitrum	$0.134343	$0.006011	$0.002957	$0.002183	−98.38%
Base	$0.149274	$0.007602	$0.001658	$0.001634	−98.91%
Optimism	$0.136594	$0.007140	$0.000261	$0.000027	−99.98%
L2 Leaders	$0.180219	$0.051259	$0.002305	$0.001512	−99.16%
Polygon	$0.005742	$0.002503	$0.001070	$0.006868	19.61%

Note: Fees are the medians of daily medians across each calendar year. 2024 Q1 and 2026 Q1 values cover January 1 through March 31 of their respective years. L2 Leaders consists of the average of the daily medians of Arbitrum, Base, and Optimism. The percentage change is from 2024Q1 to 2026Q1.

reports the median of daily median transaction fees in USD for Mainnet, Solana, L2 Leaders, and Polygon across the sample period.

The L2 Leaders declined continuously across the sample, falling 99.16% from $0.180219 in 2024 Q1 to $0.001512 in 2026 Q1. Mainnet fees observed a sharp decline across the sample, falling 99.68%from an average of $3.786300 in 2024 Q1 to $0.012283 in 2026 Q1. Polygon was the only network to see an increase in fees, which occurred in 2026 Q1. Solana fees remained consistently low throughout the sample, with a median of $0.000687 in 2024 Q1 and declining further to $0.000496 in 2026 Q1.

Although the gap between the Mainnet and Solana fees has narrowed substantially, Solana continues to offer the lowest transaction costs.

We also present a cross-chain comparison of fees in native units in

Table 3. Median transaction fees in native units.

Network	Unit	2024 Q1	2024	2025	2026 Q1	Pct. Chg.
Mainnet	ETH	1.34 × 10−3	5.63 × 10−4	5.49 × 10−5	5.34 × 10−6	−99.60%
Solana	SOL	6.57 × 10−6	8.13 × 10−6	5.96 × 10−6	5.62 × 10−6	−14.41%
Arbitrum	ETH	5.40 × 10−5	1.92 × 10−6	1.00 × 10−6	1.01 × 10−6	−98.13%
Base	ETH	5.92 × 10−5	2.25 × 10−6	5.95 × 10−7	7.50 × 10−7	−98.73%
Optimism	ETH	5.73 × 10−5	2.11 × 10−6	9.62 × 10−8	1.08 × 10−8	−99.98%
L2 Leaders	ETH	6.01 × 10−5	1.69 × 10−5	7.91 × 10−7	6.37 × 10−7	−98.94%
Polygon	POL	6.46 × 10−3	4.64 × 10−3	4.66 × 10−3	6.62 × 10−2	925.10%

Note: Fees are the medians of daily medians across each calendar year. 2024 Q1 and 2026 Q1 values cover January 1 through March 31 of their respective years. L2 Leaders consists of the average of the daily medians of Arbitrum, Base, and Optimism. The percentage change is from 2024Q1 to 2026Q1.

.

The clear outlier here is Polygon, and Polygon would not be competitive if POL prices were to return to their early 2024 levels.

Figure 16 displays median transaction fees in USD for Arbitrum, Base, and Optimism from January 2024 through March 2026.

Following the Dencun upgrade in March 2024, all three networks experienced fee declines. Fees spiked again in late 2024 before falling further after the Pectra upgrade in May 2025, after which all three networks converged to their lowest levels of the sample period.

Optimism became consistently below Arbitrum in August 2024 and Base in October 2024. It remains, on average, 78.1% below Arbitrum and 75.2% below Base.

Since May 2025, Optimism’s median transaction fees have fallen below Solana’s, averaging 81.7% lower across the eleven months. All three L2s have consistently been below Mainnet fees for the entire period.

Figure 17 shows the monthly median transaction fees in USD for the Mainnet, the L2 Leaders, Solana, and Polygon from January 2024 through March 2026.

Each increase in the target blob count was associated with lower L2 transaction fees. Before the introduction of blobs through the Dencun upgrade, the L2 Leader fee was $0.2107. The introduction of three target blobs under Dencun reduced this by 95.5% to $0.0095. Subsequent increases to six blobs under Pectra and 10 blobs under BPO-1 further reduced the fees to $0.0019 and $0.0013. The most recent increase to 14 blobs under BPO-2 saw fees rise slightly to $0.0015.

As reported in Table 2, Solana provided, on average, the lowest median transaction fees across our sample period. However, there were seven months when Polygon had lower median fees than Solana, during the second and third quarters of both 2024 and 2025. During these months, Polygon was on average 10.9% below Solana.

7. Transactions per Second Forecasts

The previous sections study the historical differences in transaction speed and cost across the Mainnet and its competitors. We now examine how the gas limit affects Mainnet TPS and how the target blob count affects L2 TPS. We extend this study to include forecasts for both Mainnet and L2 TPS and compare the forecasts to our projected Solana speeds.

Our first step is to show that gas limits are causal for Mainnet TPS and blobs for Layer 2 TPS. Dickey-Fuller [29] and KPSS [30] tests confirm that the data are difference stationary. Our tests for cointegration are mixed, so we use Toda-Yamamoto [31] tests for causality. The results are shown in

Table 4. T–Y causality tests for Mainnet and Layer 2 TPS.

VAR Lags	Direction	Test	T–Y Test Significance
5	forward	gas_limit→mainnet_tps	***
5	reverse	mainnet_tps→gas_limit
7	forward	blob_target→layer2_tps	***
7	reverse	layer2_tps→blob_target

Note: VAR lag length is selected using BIC for the Mainnet and AIC for the Layer 2s. *** 99%.

:

The [31] tests do not only show causality from the protocol changes to TPS; they also show that the impact is significant for five to seven days.

To rule out spurious regression influences, we also confirm using Granger–Newbold [32] tests that the Durbin–Watson statistic exceeds the $R^2$ for the models in Table 4.

7.1. Transactions per Second on the Ethereum Mainnet

To study the impact of the gas limit on Mainnet TPS, we begin our sample in January 2025, when the gas limit began to increase toward 35 million.

Table 5. Gas limit time trend.

	Dependent Variable: Gas Limit
	Coef.	Std. Err.	$\mathit{z}$	p-Value	[0.025, 0.975]
Constant	$2.78\times {10}^7$	$6.36\times {10}^5$	43.78	0.000	[$2.66\times {10}^7$, $2.91\times {10}^7$]
Time Trend (t)	$7.44\times {10}^4$	2660.67	27.98	0.000	[$6.92\times {10}^4$, $7.97\times {10}^4$]
Observation Period	1 January 2025 through 31 March 2026
Adjusted $R^2$	0.88

Note: The gas limit is defined as the median daily gas limit per block. Standard errors are heteroskedasticity and autocorrelation, consistent (HAC) with lag length selected automatically by a Bartlett kernel.

reports the estimates from a linear time trend regression of the gas limit over the period of 1 January 2025 through 31 March 2026. The time trend coefficient of $7.44\times {10}^4$ indicates that the gas limit increases by 2.2 million units per month under the current growth trajectory.

Applying the results from Table 5, we first utilize this forecasting model for the gas limit,

$$\mathrm{Gas}{\mathrm{Limit}}_{t+\tau }=2.78\times {10}^7+7.44\times {10}^4·(t+\tau )$$

(1)

where $t=1,2,\dots ,T$ with $t=1$ corresponding to 1 January 2025 and $T=454$ corresponding to 31 March 2026, and $\tau \ge 0$ indexing days into the forecast horizon. Figure 18 presents the forecast of gas limit values until 2036.

Having established the trajectory of the gas limit under the current growth trend, we now determine the impact that the gas limit has on the Mainnet TPS.

Table 6. Effect of gas limit on Ethereum mainnet.

	Dependent Variable: Mainnet Transactions per Second
	Coef.	Std. Err.	$\mathit{z}$	p-Value	[0.025, 0.975]
Constant	2.67	1.26	2.12	0.03	[0.20, 5.15]
Gas Limit	$3.52\times {10}^{-7}$	$3.25\times {10}^{-8}$	10.83	0.000	[$2.88\times {10}^{-7}$, $4.16\times {10}^{-7}$]
Observation Period	1 January 2025 through 31 March 2026
Adjusted $R^2$	0.66

Note: The gas limit is defined as the median daily gas limit per block. Standard errors are heteroskedasticity and autocorrelation robust (HAC) with lag length selected automatically using a Bartlett kernel.

reports the regression results.

Each additional unit of gas limit capacity is associated with roughly $3.5\times {10}^{-7}$ more transactions processed per second. Therefore, a 10,000,000-unit increase in the gas limit is associated with 3.52 more transactions processed every second. Doubling the gas limit to 120 million would result in 21 additional transactions processed every second on top of the baseline. At the current gas limit of approximately 60 million, the model estimates 24 transactions per second. For March 2026 the average Mainnet TPS was 25.59.

Given the substantial gap between the Mainnet and Solana TPS, we accordingly narrow the analysis to forecast when the Mainnet will reach 100 TPS. Solving for the gas limit required to achieve 100 TPS using Table 6 estimated coefficients yields:

$$\mathrm{Gas}{\mathrm{Limit}}^{*}={\displaystyle \frac{100-2.67}{3.52\times {10}^{-7}}}\approx \mathrm{276,500,000}.$$

(2)

The model predicts that 100 TPS on the Mainnet requires a gas limit of approximately 276.5 million. At the current gas limit of approximately 60 million, this represents a 4.6-fold increase from current levels. Substituting the predicted gas limit values using Table 5 results, we define our model as follows:

$$\begin{array}{cc}\hfill {\mathrm{TPS}}_{\mathrm{Mainnet},t+\tau }& =2.67+3.52\times {10}^{-7}·\mathrm{Gas}{\mathrm{Limit}}_{t+\tau }\hfill \\ \hfill & =2.67+3.52\times {10}^{-7}·\left(2.78\times {10}^7+7.44\times {10}^4·(t+\tau )\right)\hfill \\ \hfill & \approx 12.46+0.03·(t+\tau )\hfill \end{array}$$

(3)

where $t=1,2,\dots ,T$ with $t=1$ corresponding to 1 January 2025 and $T=454$ corresponding to 31 March 2026, and $\tau \ge 0$ indexing days into the forecast horizon. We predict that Mainnet will reach 100 TPS by February 2034, as seen in Figure 19.

To provide confidence intervals for our forecast, we use the HAC standard errors for the marginal effect of the gas limit on Mainnet TPS from Table 6. The 95% confidence interval for our forecast ranges from August 2032 to a period after the end of our sample in March 2036.

We compare our Mainnet TPS forecasts to 30-day and 90-day random walks using the [33] test for forecast accuracy. The model improves upon both random walks at the 99% confidence level.

As a final robustness check, we consider the implications of a model where TPS does not scale linearly with the gas limit but rather as a square root. With this specification, the Mainnet TPS does not reach 100 TPS prior to the end of our sample in March 2036. This is consistent with our conservative forecast for the linear model.

7.2. Gigagas

Under the current growth trajectory, 100 TPS will take almost eight years to reach. However, the Ethereum Foundation has outlined a more aggressive path forward with a goal to reach one gigagas per second.

The L1 Strawmap proposes an increase in the gas limit by a factor of ten every two years, totaling a 100-fold increase over four years under EIP-7938 [7]. This upgrade is not expected to activate until 2027. Therefore, we use the time trend model in Table 5 to forecast the gas limit until 2027, at which point the EIP formula is applied.

$$\mathrm{Gas}{\mathrm{Limit}}_{t+\tau }=\left\{\begin{array}{cc}2.78\times {10}^7+7.44\times {10}^4·(t+\tau )\hfill & \tau \le 275\hfill \\ \mathrm{Gas}{\mathrm{Limit}}_{t+276}·{10}^{(\tau -275)/730}\hfill & 276\le \tau \le 1736\hfill \\ \mathrm{Gas}{\mathrm{Limit}}_{t+276}·100\hfill & \tau >1736\hfill \end{array}\right.$$

(4)

where $t=1,2,\dots ,454$ with $t=1$ corresponding to 1 January 2025 and $t=454$ corresponding to 31 March 2026, and $\tau \ge 0$ indexing days into the forecast horizon. $\tau =276$ corresponds to 1 January 2027, $\tau =1736$ corresponds to 31 December 2030. The gas limit increases by a factor of ten every two years, with the upper threshold at $100\times \mathrm{Gas}{\mathrm{Limit}}_{t+276}$ after four years. Mainnet TPS is then forecast using the first line of Equation (3)

$${\mathrm{TPS}}_{\mathrm{Mainnet},t+\tau }=2.67+3.52\times {10}^{-7}·\mathrm{Gas}{\mathrm{Limit}}_{t+\tau }$$

Figure 20 presents the forecast gas limit under the combined 1 January 2025 to 31 March 2026 trendline and EIP-7938 exponential growth schedule.

The gas limit grows gradually through 2027 following the observed 2025 trend, after which the EIP formula takes effect, producing a sharp exponential increase. Under EIP-7938, the Mainnet reaches 100 TPS by January 2028, more than six years earlier than when using current trends. The gas limit is projected to level off at 8.22 billion at the end of 2030 with the Mainnet at 2896 TPS.

7.3. Transactions per Second on the L2 Networks

We now study the impact of the target blob count on L2 total TPS. Figure 21 plots L2 total TPS alongside the target blob count from January 2024 through March 2026.

The step increases in the target blob count correspond to the Dencun upgrade in March 2024, the Pectra upgrade in May 2025, and the BPO activations in December 2025 and January 2026. From January 2024 through November 2025, TPS rose broadly in line with each increase in the target blob count. Following the second BPO activation in January 2026, TPS spiked sharply before declining through 2026 Q1, ending with TPS below pre-BPO levels.

Table 7. Effect of target blobs on L2 TPS.

	Dependent Variable: Layer 2 Total TPS
	Coef.	Std. Err.	$\mathit{z}$	p-Value	[0.025, 0.975]
Constant	128.46	8.65	14.86	0.000	[111.51, 145.41]
Target Blobs	10.75	1.52	7.09	0.000	[7.78, 13.72]
Observation Period	1 January 2024 through 31 March 2026
Adjusted $R^2$	0.37

Note: Standard errors are heteroskedasticity- and autocorrelation-robust (HAC) with lag length selected automatically by a Bartlett kernel.

quantifies the impact target blobs have on TPS.

Each additional target blob is associated with 10.75 more L2 total TPS. Increasing the blob count by four, as it was with both BPOs, results in an increase of 43 TPS. At the current target blob count of 14, the model predicts a baseline of 278.96 L2 total TPS. The average TPS since the second BPO activation was 227.77.

We now turn to our forecast of TPS and construct our projection to maintain the stepwise nature of the target blob increases. BPO-1 and BPO-2 each added four blobs to the target, 30 days apart, giving an increment rate of four blobs per 30 days defined as:

$$\mathrm{Target}{\mathrm{Blobs}}_{t+\tau }=\min \left(14+int⌊{\displaystyle \frac{\tau }{30}}⌋\times 4,300\right)$$

(5)

where t is 1 April 2026 and $\tau \ge 0$ indexes days into the forecast horizon. We apply this same increment schedule going forward until the target blob count reaches 300. The Strawmap projects one gigagas per second for the Layer 2s, so this target is very conservative. The projected target blob count is then used to forecast L2 TPS using the coefficients from Table 7:

$${\mathrm{TPS}}_{\mathrm{L}2,t+\tau }=128.46+10.75·{\mathrm{TargetBlobs}}_{t+\tau }$$

(6)

where ${\mathrm{TPS}}_{\mathrm{L}2,t+\tau }$ denotes L2 total TPS and ${\mathrm{TargetBlobs}}_{t+\tau }$ denotes the projected target blob count from Equation (5).

Figure 22 displays the forecasted L2 total TPS under our defined blob count increase structure.

We construct confidence intervals for these forecasts as we did for the Mainnet, using the HAC standard errors for the marginal effect of the blob target from Table 7.

L2 total is expected to reach 800 TPS in April 2027 at 66 blobs and 1000 TPS in August 2027 at 82 blobs. The 95% confidence interval for our forecast ranges from December 2026 to October 2027 for 800 TPS and from April 2027 to April 2028 for 1000 TPS.

We compare our Layer 2 TPS forecasts to 30-day and 90-day random walks using [33]. The model does not provide a significant improvement over the 30-day random walk, but it does improve upon the 90th day at the 95% confidence level.

Using the square root specification for the blob target, the Layer 2 total TPS reaches 800 TPS in December 2029. It does not reach 1000 TPS prior to the end of our sample in March 2036.

7.4. Solana Crossover Prediction

We now examine when L2 total will close the speed gap with Solana. For 2026 Q1, Solana had an average daily TPS of 1303.46 and the combined total for the L2s was 226.92, a difference of 1076.54 TPS.

The L2 total forecast is directly from Section 7.3. Solana TPS is extrapolated using a linear time trend fit to historical data from 1 January 2024 to 31 March 2026 shown in

Table 8. Solana TPS time trend.

	Dependent Variable: Solana TPS
	Coef.	Std. Err.	$\mathit{z}$	p-Value	[0.025, 0.975]
Constant	745.41	27.10	27.50	0.000	[692.29, 798.53]
Time Trend (t)	0.565	0.073	7.78	0.000	[0.423, 0.707]
Observation Period	1 January 2024 through 31 March 2026
Adjusted $R^2$	0.33

Note: Standard errors are heteroskedasticity- and autocorrelation-robust (HAC) with lag length selected automatically by a Bartlett kernel. Daily transaction data are obtained from our public Dune query: https://dune.com/queries/6946173anddividedby86,400forTPS, accessed on 17 July 2026.

:

$${\mathrm{TPS}}_{\mathrm{Solana},t+\tau }=745.41+0.565·(t+\tau )$$

(7)

where $t=1,2,\dots ,T$ with $t=1$ corresponding to 1 January 2024 and $T=820$ corresponding to 31 March 2026, and $\tau \ge 0$ indexes days into the forecast horizon.

Figure 23 presents historical and forecasted TPS for L2 Total and Solana until 2030.

L2 total TPS is projected to surpass Solana TPS in March 2029, with both networks at approximately 1820 TPS at the crossing point.

We construct confidence intervals for these forecasts as we did for the Mainnet, using the HAC standard errors for the marginal effect of the blob target from Table 7. The 95% confidence interval for the Solana crossover ranges from March 2028 to November 2031.

Before concluding this section, we provide some analysis of coefficient stability for both TPS forecasts.

Table 9. Stability of marginal effects on TPS.

Dep. Var.	Regressor	Range	Full	Sub-Sample	95% Interval
Mainnet TPS	Gas limit	G $>=$ 45 M	3.52 × 10−7	4.34 × 10−7	[3.23 × 10−7, 5.39 × 10−7]
Layer 2 TPS	Target Blob	January–December 2025	10.748	10.809	[6.310, 15.307]

shows that the effect of the gas limit has remained stable as the gas limits have reached 45 million. It also shows that the effect of blobs on Layer 2 TPS was stable through the end of December 2025.

Our fits for the Layer 2 TPS and blob target models are not as strong as for the gas limit on Mainnet TPS. We can also see a clear slowdown in the growth of Layer 2 TPS in Figure 21. If demand remains sluggish, the additional blob capacity we forecast may not be utilized by the Ethereum ecosystem.

8. Median Transaction Fee Forecast

The Mainnet is not expected to reach Solana speeds within the next 10 years under the current gas limit trajectory, but the outlook for transaction fees is more favorable. This section will cover predictions for Mainnet fees and compare them with Solana’s forecasted fees.

Our methodology is similar to the forecasts we generate for TPS. After finding the data to be difference stationary, we first show causality from gas limits and blobs to fees using the Toda–Yamamoto [31] tests in

Table 10. T–Y causality tests for Mainnet and Layer 2 fees.

VAR Lags	Direction	Test	T–Y Test Significance
7	forward	gas_limit→mainnet_fee	**
7	reverse	mainnet_fee→gas_limit
7	forward	blob_target→layer2_fees	*
7	reverse	layer2_fees→blob_target

Note: The VAR lag length is selected using the AIC. * 90% confidence level, ** 95%.

.

Granger–Newbold tests [32] again confirm that the Durbin–Watson test exceeds the $R^2$, ruling out a possible spurious relationship.

8.1. Ethereum Mainnet

Figure 24 plots the daily median transaction fees on the Mainnet from 1 January 2024, through 31 March 2026.

Fees have declined substantially over this period. The daily median transaction fee for 2024 Q1 was $3.786300, and the fee for 2026 Q1 was $0.012283, a 99.68% decline. Overall, as the gas limit increased, median transaction fees declined. Between 1 January 2024 and 3 February 2025, the gas limit was roughly 30 million. The median transaction fee for this period was $1.317987. Between 4 February 2025 and 20 July 2025, the gas limit was roughly 36 million. The median transaction fee for this period was $0.172602. Between 21 July 2025 and 25 November 2025, the gas limit was roughly 45 million. The median transaction fee for this period was $0.162836. Between 26 November 2025 and 31 March 2026, the gas limit was roughly 60 million. The median transaction fee for this period was $0.012753.

As of 2026 Q1, the median transaction fee is $0.012283. We are interested in forecasting when the Mainnet median fee will fall to $0.0005, which is the 2026 Q1 Solana median fee. The estimation results are in

Table 11. Determinants of Ethereum median transaction fees (USD).

	Dependent Variable: log(Median Fee USD)
	Coef.	Std. Err.	$\mathit{z}$	p-Value	[0.025, 0.975]
Constant	3.01	0.32	9.37	0.000	[2.38, 3.64]
$\mathrm{Gas}\mathrm{Limit}$	$-1.19\times {10}^{-7}$	$6.54\times {10}^{-9}$	−18.18	0.000	[$-1.32\times {10}^{-7}$, $-1.06\times {10}^{-7}$]
Observation Period	1 January 2025 through 31 March 2026
Adjusted $R^2$	0.74

Note: The gas limit is defined as the median daily gas limit per block. Standard errors are heteroskedasticity- and autocorrelation-consistent (HAC) with lag length selected automatically by a Bartlett kernel.

, which reports the impact of the gas limit on the log median transaction fee. We take the log of the median fee to ensure that the forecasted values remain positive.

A 5-million-unit increase in the gas limit is associated with fees falling by approximately 59.2%. At a gas limit of 60 million, the model predicts that the median fee will be $0.016, which is quite close to Mainnet’s average median fee of $0.013 since the gas limit reached 60 million on 26 November 2025.

To forecast the median fee, we extrapolate the gas limit from the 2025 time trend as estimated in Table 5. Therefore, recall the gas limit forecast model:

$$\mathrm{Gas}{\mathrm{Limit}}_{t+\tau }=2.78\times {10}^7+7.44\times {10}^4·(t+\tau )$$

(8)

where $t=1,2,\dots ,T$ with $t=1$ corresponding to 1 January 2025 and $T=454$ corresponding to 31 March 2026, and $\tau \ge 0$ indexes days into the forecast horizon. Our fee forecast model then becomes:

$$\begin{array}{cc}\hfill \ln \left({\mathrm{Fee}}_{t+\tau }^{\mathrm{Mainnet}}\right)& =3.01-1.19\times {10}^{-7}·{\mathrm{GasLimit}}_{t+\tau }\hfill \\ \hfill & =3.01-1.19\times {10}^{-7}·\left(2.78\times {10}^7+7.44\times {10}^4·(t+\tau )\right)\hfill \\ \hfill & \approx -0.30-0.00885·(t+\tau ).\hfill \end{array}$$

(9)

${\mathrm{GasLimit}}_{t+\tau }$ is given by Equation (8) and ${\mathrm{Fee}}_{t+\tau }^{\mathrm{Mainnet}}$ denotes the daily median transaction fee of the Mainnet.

Figure 25 displays the forecasted Mainnet median transaction fees under our model Equation (9) for the effect of gas limits on fees. We produce confidence intervals as we did in the prior section on TPS using the standard errors of our gas limit estimates.

Mainnet median fees are expected to reach $0.0005 in April 2027 at a gas limit of 89.2 million. The 95% confidence forecast interval ranges from January to September 2027.

We compare our Mainnet fee forecasts to 30-day and 90-day random walks using [33]. The gas limit model improves upon both random walks at the 99.9% confidence level.

For Mainnet fees, the square root specification predicts a crossover of the $0.0005 fees in August 2027. This is again similar to the conservative bound on the linear forecast.

8.2. Ethereum and Solana Crossover Prediction

We first estimate a time trend extrapolation for Solana in

Table 12. Solana transaction fees’ (USD) time trend.

	Dependent Variable: Log Solana Median Fee
	Coef.	Std. Err.	$\mathit{z}$	p-Value	[0.025, 0.975]
Constant	−6.4925	0.073	−88.71	0.000	[−6.636, −6.349]
Time Trend	−0.00212	0.000	−8.53	0.000	[−0.003, −0.002]
Observation Period	1 January 2025 through 31 March 2026
Adjusted $R^2$	0.51

Note: Standard errors are heteroskedasticity- and autocorrelation-robust (HAC) with the lag length selected automatically by a Bartlett kernel.

.

Having shown that Mainnet fees are projected to reach Solana’s 2025 median fee level by early 2027, we now examine when Mainnet fees can converge with Solana’s projected fee level.

Our Mainnet median fee forecast is directly from Section 8.1. We forecast Solana fees by fitting a time trend to the log of historical fees. Our model is as follows:

$$\ln \left({\mathrm{Fee}}_{t+\tau }^{\mathrm{Solana}}\right)=-6.4925-0.00212·(t+\tau )$$

(10)

where $t=1,2,\dots ,T$ with $t=1$ corresponding to 1 January 2025 and $T=454$ corresponding to 31 March 2026, and $\tau \ge 0$ indexes days into the forecast horizon. $\ln \left({\mathrm{Fee}}_{t+\tau }^{\mathrm{Solana}}\right)$ denotes the logged daily median transaction fee on Solana.

Figure 26 presents historical and forecasted daily median transaction fees for the Mainnet and Solana until 2028. Confidence intervals for the Mainnet fees are from the standard errors of the gas limit effect.

The Mainnet is projected to achieve Solana median transaction fees in August 2027, with both networks’ fees at $0.000204. The 95% confidence interval date range is between January and September 2027.

The in-sample fits are strong here, and the interval around the Solana forecasts is just slightly more than a year.

8.3. Layer 2 Leader Fee Trends and Solana Crossover

Fees on the cheapest of the largest L2 networks are already below Solana levels in the first quarter of 2026. Optimism, the third largest L2, has fees of $0.000027 for 2026 Q1, 94.6% lower than Solana’s at $0.000496.

We now consider when the larger set of L2 leaders, adding Base and Arbitrum to Optimism, will reach fee Solana fee levels. We use the same explanatory variables that we use for L2 TPS, regressing log fees on target blobs in

Table 13. Effect of target blobs on L2 fees.

	Dependent Variable: Layer 2 Leader Fees
	Coef.	Std. Err.	$\mathit{z}$	p-Value	[0.025, 0.975]
Constant	−6.0220	0.118	−51.104	0.000	[−6.253, −5.791]
Target Blobs	−0.0476	0.013	−3.803	0.000	[−0.072, −0.023]
Observation Period	1 January 2025 through 31 March 2026
Adjusted $R^2$	0.09

Note: Standard errors are heteroskedasticity- and autocorrelation-robust (HAC) with lag length selected automatically by a Bartlett kernel.

.

Our estimates in Table 13 imply that every 30 days when the blob target rises by four, L2 median fees fall by −17.34%.

We create a forecast using the coefficients from Table 13 and our blob growth rule

$$\ln \left({\mathrm{Fee}}_{t+\tau }^{\mathrm{L}2}\right)=-6.0220-0.0476·{\mathrm{TargetBlobs}}_{t+\tau }$$

(11)

where ${\mathrm{Fee}}_{t+\tau }^{\mathrm{L}2}$ denotes L2 leader fees, and ${\mathrm{TargetBlobs}}_{t+\tau }$ denotes the projected target blob count from Equation (5).

Figure 27 shows that the leading L2 networks will reach $0.0001 fees in May 2027. Our confidence bounds on the marginal effects place a 95% region between November 2026 and a date beyond our sample end of March 2036.

We compare our L2 leader fee forecasts to 30-day and 90-day random walks using [33]. The blob target model improves upon both random walks at the 99.9% confidence level.

For Layer 2 fees, the square root specification predicts a crossover of the $0.0001 fees in March 2029. The square root scaling slows down the predicted crossover by approximately 22 months.

In predicting a crossover with Solana, we use the same time trend model as Equation (10). Figure 28 displays our predicted crossover in October 2026 with a 95% confidence band between May 2026 and a date beyond our March 2036 sample end.

Layer 2 fees are falling more quickly than Solana’s. Our forecast predicts that the Layer 2 per-transaction fees of $0.000328 cross Solana’s $0.000371 at 42 blobs in October 2026.

The model fits for Layer 2 gas fees on target blobs are weaker, and this is reflected in wide confidence intervals for our forecasts.

For both crossover dates, we cannot rule out at the 95% confidence interval that they will take place outside of our terminal date of March 2036.

We close this section with a discussion of coefficient stability based on recent samples for the marginal effects in

Table 14. Stability of marginal effects on fees.

Dep. Var.	Regressor	Range	Full	Sub-Sample	95% Interval
log(Mainnet fee)	Gas limit	G $>=$ 45 M	−1.19 × 10−7	−1.27 × 10−7	[−1.46 × 10−7, −1.06 × 10−7 ]
log(Layer 2 fee)	Target Blob	January–November 2025	−4.76 × 10−2	−0.1224	[−0.244, −3.00 × 10−2 ]

. As with our Mainnet TPS forecasts, the marginal effect of the gas limit on fees has been stable since the increase to 45 million. Layer 2 fees have the weakest fit, and we find the coefficients are less stable after November 2025.

9. Conclusions

This paper documents the evolution of transaction speed and cost from January 2024 through March 2026 across the Ethereum network, its L2 ecosystem, Polygon, and Solana. This period is defined by Dencun, Fusaka, and Pectra, three major Ethereum protocol upgrades, and BPO-1 and BPO-2, two blob parameter-only upgrades.

We summarize our forecasts in

Table 15. Forecast milestones.

Metric	Milestone	Figure	Date	Value at Milestone	Adj. R2
TPS	Mainnet 100 TPS	Figure 19	2034-02	276.5 M gas limit	0.66
TPS	100 TPS Gigagas	Figure 20	2028-01	276.5 M gas limit	0.66
TPS	L2 800 TPS	Figure 22	2027-04	66 target blobs	0.37
TPS	L2 1000 TPS	Figure 22	2027-08	82 target blobs	0.37
TPS	L2 > Solana	Figure 23	2029-03	Both at 1820 TPS	L2: 0.37; Solana: 0.33
Median fee	Mainnet $0.0005	Figure 25	2027-04	89.2 M gas limit	0.74
Median fee	Mainnet < Solana	Figure 26	2027-08	Both at $0.000204	M: 0.74; S: 0.51
Median fee	L2 leaders $0.0001	Figure 27	2027-05	70 target blobs	0.09
Median fee	L2 leaders < Solana	Figure 28	2026-10	L2 $0.000328; Solana $0.000371	L2: 0.09; Solana: 0.51

.

Confidence intervals around these forecasts are larger, particularly for the Layer 2 results. The confidence intervals should not be interpreted independently. A slowdown in gas limit increases or a decline in blob utilization will move all of the estimated targets further into the future.

The ability of Ethereum to function with gas limits proposed in the Strawmap is a work in progress. Ref. [34] suggests that massive parallelism may make these goals achievable on standard hardware. Nonetheless, if transactions and fees are not able to scale linearly as we have modeled, it may substantially delay the milestones in Table 15.

In terms of speed, Solana remains the fastest network, averaging 1303 TPS in 2026 Q1. Mainnet speed has nearly doubled since 2024, reaching 25.78 TPS in 2026 Q1. We estimate that each 10-million-unit increase in the gas limit is associated with 3.52 additional TPS, meaning the Mainnet cannot reach 100 TPS under current gas limit trajectories until February 2034. The proposed Strawmaps’ EIP-7938 gas limit increase, if activated in 2027, would accelerate that timeline considerably, allowing the Mainnet to reach 100 TPS by January 2028.

We anticipate a much faster growth in TPS for the L2 networks. Extending the BPO upgrade schedule, with a 4-blob increase every 30 days, L2 TPS is projected to cross Solana’s time-trend extrapolated TPS by March 2029 at 1820 TPS.

Solana is an adapting protocol as well, and the Apenglow upgrade could provide block space for upwards of 4500 TPS. If organic transaction demand arises to fill the space on Solana, the Layer 2s will still substantially lag behind Solana in speed.

Our analysis indicates the Mainnet should reach Solana fee levels in less than two years. A 5-million-unit increase in the gas limit is associated with fees falling by approximately 59.2%. Even if EIP-7938 is not adopted, the Mainnet will reach Solana’s extrapolated time-trend fees of $0.0002 by August 2027. The L2 networks are currently competitive with Solana on fees, but we forecast declining fees across the leading networks as target blobs increase. We project that average fees across Arbitrum, Base and Optimism will fall below Solana’s by October 2026.

With projected gas limit increases, the Mainnet may soon be competitive with Solana on fees. Without a healthy L2 ecosystem, though, the Mainnet will be a sluggish alternative to faster blockchains. It will likely continue to lag Solana on speed-sensitive DEX trading.