Are Rollups Overvalued or Undervalued? An Analysis of Rollup’s Revenue and Cost Structure

Published in

IOSG Ventures

14 min readJul 11, 2024

Background

The Ethereum Rollup L2 ecosystem is thriving, with a total daily TVL surpassing $37 billion. However, the short-term price performance of Rollups has not met expectations. In terms of FDV, mainstream Rollups such as Arbitrum reached a FDV of $8 billion, Optimism $7.4 billion, Starknet $7.1 billion, and zkSync FDV at $3.7 billion, while Solana’s FDV is at $77 billion.

From a revenue perspective, Ethereum’s revenue reached $2 billion in 2023, while Arbitrum and Op Mainnet generated annual revenues of $63 million and $37 million respectively. Newcomers like Base and zkSync, which entered the market with strong performances this year, earned $50 million and $23 million in the first half of 2024, while Ethereum generated $1.39 billion in the same period, showing that the gap has not narrowed. Rollups have yet to achieve a revenue scale comparable to Ethereum’s.

One contributing factor is that the apps on Rollups have not been sufficiently attractive to users, an issue common across most chains. Our question is: how effectively are Rollups fulfilling their role as infrastructure for mass adoption, and is their value being underestimated due to currently low levels of activity?

Everything still goes back to the original proposition: the emergence of Rollups was driven by Ethereum’s increasing congestion and costs reaching levels unacceptable to users. Rollups were inherently designed to reduce costs. Beyond the security purpose, Rollups also boast a disruptive cost structure that becomes more economical as transaction volume increases.If this principle can be effectively realized, Rollups may hold irreplaceable value.

This article briefly analyzes the current economic structure of Rollups and looks forward to future possibilities.

Rollup Business Model

Overview

Rollups use the Sequencer as the gateway for cash flow, charging users fees on Rollups transactions to cover the costs incurred on both L1 and L2, and to generate additional profits.

On the revenue side, the fees include:

Base fees (including congestion fees)
Priority fees
Fees to cover L1 costs

Additionally, the protocol can capture potential revenues through strategies that include:

MEV fees

On the cost side, the expenses include the relatively minor L2 costs and the more substantial L1 costs, such as:

Data Availability (DA) costs
Verification costs
Excution costs

What sets Rollups apart from other L2 solutions is their cost structure. The largest proportion, the DA costs, are seen as variable costs that fluctuate with the amount of data submitted to L1, while verification and execution costs are more often considered fixed costs essential for maintaining Rollups operations.

We aim to clarify the marginal costs of Rollups, that is, to what extent the additional cost of an extra transaction is less than the average cost per transaction. This analysis is crucial to validate the phrase “the more users, the cheaper the Rollup becomes.”

The reason behind this is that Rollups handle data in batches, compress data, and aggregate verifications, which theoretically leads to lower marginal costs compared to other L1s. The fixed costs of Rollups should be well amortized over each transaction, making them negligible when transaction volumes are high, but this also requires our validation.

Rollups Revenue

Transaction Fee Income

The primary revenue for Rollups comes from L2 transaction fees. These fees are intended to cover the operating costs of Rollups and generate a portion of profits to hedge against the long-term fluctuations in L1 gas costs. Some Rollups also charge transaction priority fees, allowing users to expedite urgent transactions.

Arbitrum and zkSync use a First-Come, First-Served mechanism, where transactions are processed in the order they are received. OP stack has adopted a more flexible approach to this issue, allowing transactions to “jump the queue” by paying a priority fee.

For users, L2 base fees are determined by minimum fee during periods of low activity. During busy times, congestion fees are charged based on each Rollup’s assessment of the congestion level, which often increases exponentially.

Since the costs of Rollups L2 are extremely low (consisting only of off-chain engineering and operational costs) and the charges are quite flexible, nearly all the income used to pay L2 fees becomes profit for the protocol. Due to the centralization of the current sequencer, the governance organization can relatively freely decide on fee parameters to meet their short-term needs.

MEV Revenue

MEV transactions are divided into malicious and non-malicious. Malicious MEV includes front-running transactions like sandwich attacks. Non-malicious MEV involves back-running transactions such as arbitrage and liquidations.

Unlike L1, Rollups do not offer a public mempool; only the sequencer can see transactions before they are finalized, so only the sequencer has the capability to initiate MEV on L2. Since most L2s currently run their own centralized sequencers, the occurrence of malicious MEV is unlikely for the time being.

According to research by Christof Ferreira Torres and others, which involved replaying transactions on Rollups, it was concluded that Arbitrum, Optimism, and Zksync do engage in on-chain non-malicious MEV activities. These three chains have collectively generated $22 million in MEV value, making it a significant source of revenue worth noting.

Source: Rolling in the Shadows: Analyzing the Extraction of MEV Across Layer-2 Rollup

Fees to cover L1 costs

This portion of fees is charged by Rollups to cover L1 costs. Besides predicting L1 gas to cover the costs of L1 data, Rollups also incur additional fees as a reserve to hedge against future gas price fluctuations, which fundamentally represents an revenue for Rollups. For example, Arbitrum adds a “Dynamic” fee, while OP stack multiplies the fee by a “Dynamic Overhead” coefficient. Before the EIP4844 upgrade, these fees were estimated to be about one-tenth of the DA costs.

Revenue Sharing

Base, due to its use of the OP stack, has a special revenue-sharing model with OP Superchain. Base commits to giving the greater of either 2.5% of its total income or 15% of the profits (after deducting the costs associated with submitting data to L1) from L2 transactions to the OP stack. In return, Base will participate in the on-chain governance of both the OP Stack and Superchain, and will receive up to 2.75% of the OP token supply. Recent data indicates that Base contributes about 5 ETH per day to Superchain’s revenue.

It is apparent that Base provides a significant proportion of revenue to Optimism. Beyond mere cash flow, a healthy network effect also makes the OP Stack ecosystem more attractive to users and the market. Although some of Arbitrum’s metrics such as TVL or stablecoin market cap may exceed those of Base + Optimism, it can no longer surpass the latter in terms of transaction volume and revenue. This is evident from their P/S ratios — considering Base’s revenue, the P/S ratio of $OP is 16% higher than that of $ARB, reflecting the additional value the ecosystem adds to $OP.

Rollups Costs

Ethereum L1 Data Costs

Each chain has a specific cost structure, but they can generally be divided into execution costs, Data Availability costs, and verification costs (ZK Rollups).

Execution costs

These mainly include state updates between L1 and L2, and cross-chain interactions.

DA Costs

This involves posting compressed transaction data, state roots, and ZK proofs to the DA layer. Before the EIP4844 upgrade, the primary cost for L1, especially for protocols like Arbitrum and Base (over 95%), and for Zksync (over 75%), and Starknet (over 80%) came from DA costs. After EIP4844, DA costs significantly decreased, with reductions varying by Rollups mechanisms, ranging from 50% to 99%.

Verification Costs

Primarily relevant for ZK Rollups, these costs are for verifying the reliability of Rollups transactions using ZK approach.

Other Costs

These mainly include off-chain engineering and operational costs. Given the current operation of Rollups, the cost of node operation is close to that of cloud server costs, which are relatively low (comparable to corporate AWS server costs).

Comparison of L2 Profits with Other L1 Data

By now, we have a general understanding of the overall revenue and expenditure structure of Rollups. We can compare this with Alt L1s. We have selected the average weekly data from Rollups including Arbitrum, Base, zkSync, and Starknet as Rollups Average Performance.

The overall profit margins of Rollups are similar to those of Solana and show a clear advantage over BSC, reflecting the excellent performance of Rollups’ business model in terms of profitability and cost management.

Comparison among Rollups

Overview

Rollups show significant differences in fundamental performance in different stages. For instance, when there is an expectation of airdrop, Rollups experience a substantial increase in transaction volume. This increase is accompanied by a significant rise in both revenue and cost.

Most Rollups are still in their early stages, where short term profitability isn’t as crucial as ensuring financial sustainability and supporting their long-term competition. This aligns with Starknet’s current attitude of not charging users additional fees for profit.

However, since mid-March, 2024, Starknet has been operating at a continuous loss. What are the underlying reasons for these losses, and will this continue over the long term?

Let’s delve deeper into this question. The marginal cost structures of Rollups vary due to the specific Rollup mechanisms each chain uses. Differences in data compression techniques and other computational mechanisms also contribute to cost disparities.

We aim to compare costs within Rollups to help us horizontally assess what characteristics different Rollups possess.

Cost Structures of Different Rollup Types

ZK Rollups

ZK Rollups primarily differ in their verification costs, which often can be considered as their fixed costs. These costs are difficult to offset through fee allocation and are a root cause of Rollups running into financial deficits.

Source: David Barreto @Starknet, Quarkslab, Eli Barabieri, IOSG

We will use Starknet and zkSync as examples.

Starknet

Starknet uses its proprietary verification service, SHARP, to handle transaction ordering, confirmation, and block production. After these steps, transactions are batched and processed through SHARP to construct transaction proofs, which are then sent to an L1 contract for verification. Once approved, the proofs are forwarded to the Core contract. In Starknet, the fixed costs for verification and DA are derived from the block and batch processes, respectively.

Source: Starknet community — Starknet Costs and Fees

In Starknet, variable costs increase with the number of transactions, primarily due to DA costs, which in theory should not incur additional expenditures from users. However, in fact, Starknet charges transaction fees per write operation, but its DA costs are determined only by the number of memory units updated, not by the frequency of updates to each unit. Thus, Starknet has previously overcharged for DA costs.

The collection of transaction fees and the payment of operational costs occur at different times, which can lead to potential losses or profits.

Therefore, as long as transactions continue to occur, Starknet needs to consistently produce blocks and pay the fixed costs associated with blocks and batches. Additionally, the more transactions there are, the higher the variable costs that need to be paid. Fixed costs do not significantly increase marginal costs.

Source: Eli Barabieri — Starknet User Operation Compression

Due to the computational resource limits per block (Cairo Steps), Starknet’s gas fee calculation method is based on the resources used and the data volume, covering both fixed and variable costs. However, the cost per block or batch is difficult to allocate to each transaction, but since a block gets finalized once a certain level of computational resources is reached (triggering fixed costs), a portion of the fixed costs can be calculated and charged based on the amount of computational resources used.

However, due to the limitations on block time, if the transaction volume is insufficient (the computational load in a single block is low), computational resources do not effectively reflect the costs that need to be distributed, so fixed costs cannot be fully covered. Furthermore, “computational resource limits” are subject to changes with Starknet network parameter upgrades. The significant short-term losses after EIP4844 exemplify this, with losses easing only after the computational resource parameters included in the charging fees were adjusted.

Starknet’s charging model is not effective in covering fixed costs with each transaction. Therefore, when Starknet’s mainnet updates and transaction volumes are extremely low, it experiences losses.

zkSync (zkSync Era)

After the Boojum upgrade, zkSync Era shifted from block verification to batch verification and only storing state differences, effectively reducing verification and DA costs. The process is basically similar to Starknet, where the Sequencer submits batches to the Executor contract (state differences and DA commitments), and the Prover nodes submit verifications (ZK proofs and DA commitments). The batches are executed after verification is passed (every 45 batches); the difference is that Starknet incurs verification costs for both blocks and batches, whereas zkSync incurs verification costs only for batches.

Cost Comparison between zkSync and Starknet

The batch size in Starknet is much larger compared to zkSync Era, with zkSync Era limiting each batch to 750 or 1,000 transactions, while Starknet has no transaction limits per batch.

From the table, it is clear that Starknet boasts a more robust scaling capability. The computational resource limits per block allow it to process more transactions and batches, enhancing its performance in high-frequency trading and scenarios that involve large volumes of simple operations. However, Starknet encounters high fixed costs during periods of low transaction volumes. Conversely, zkSync benefits from its high compression efficiency and flexible block resources, which provide advantages in adapting to fluctuations in L1 gas prices and during periods of low activity. However, zkSync faces limitations in the speed of block production.

For users, Starknet’s charging model tends to be more friendly, with less correlation to L1 and stronger economies of scale. zkSync’s fees are more cost-effective but are subject to greater fluctuations with L1. For Rollups, during phases of low activity, Starknet’s high fixed costs could lead to losses, whereas zkSync is better suited to such scenarios. Starknet is more suited for handling large volumes of high-frequency transactions and controlling costs at the same time, whereas zkSync’s current mechanism may slightly lag in scenarios with high volumes.

Optimistic Rollup

The cost structure of Optimistic Rollup is relatively straightforward. Without the cost of verification, users only need to pay for the computational costs on L2 and the DA costs for publishing data to L1. Each block, or several blocks, periodically upload a state root to L1, which tends to be a fixed cost. While the upload of compressed transactions represents a variable cost that is predictable and distributed evenly across each transaction.

Compared to Zk Rollup, it has lower fixed costs, making it more suitable for scenarios with medium transaction volumes. However, since each transaction requires a signature, this results in higher DA or variable costs. In the phase of high activity, the optimistic Rollup’s advantage on marginal costs becomes smaller.

Based on the current adoption scale, the fixed costs of ZK Rollups may lead to a higher cost compared to Optimistic Rollups, thus increasing the costs for users. However, the scalability advantages of ZK Rollups are significant: as transaction volumes increase, verification costs gradually diminish, and the marginal costs saved will eventually exceed those of Optimistic Rollups. Besides, running Validiums/Volitions and requiring only state differences for DA, along with faster withdrawal speeds, are better for scalability and the RaaS ecosystem.

Data Comparison

Revenue

From the table we can find that per transaction, Base has higher revenues, Starknet has lower revenues. It is noted that before the EIP4844 upgrade, Arbitrum had higher revenue per transaction, while after the upgrade, Base’s revenue per transaction increased.

Costs

Looking at the cost per transaction, before EIP4844, Base had excessively high transaction costs due to elevated DA costs, effectively resulting in higher marginal costs. The cost advantages expected from economies of scale were not evident. After the EIP4844 upgrade, with a significant reduction in DA costs, the cost per transaction for Base decreased significantly, making it the lowest among all Rollups. Comparing OP and ZK, it is evident that OP Rollups benefited more from the upgrade.

StarkNet’s DA costs are reduced by about 4 to 10 times, slightly less than OP Rollups. This is consistent with our theory: ZK Rollups did not benefit as much as OP Rollups in the EIP4844 upgrade. The performance of ZK Rollup costs after the EIP4844 also reflects the impact of fixed costs.

Profit

According to the data, Base has the highest profit margin due to its economy of scale, significantly surpassing Arbitrum. Among ZK Rollups, Starknet, due to its low transaction volume, cannot cover its fixed costs currently, resulting in negative transaction profits, while zkSync, although profitable, is also limited by fixed costs and has lower profits than OP Rollups. The EIP4844 upgrade did not directly enhance profit margins — the main beneficiaries will be the users, who will see a substantial reduction in their cost expenses.

Future

Cost Side

Currently, most Rollups are still in the early part of their margin curve, where marginal costs and the average fixed costs decrease with increasing transaction volumes. However, in the future, as transaction volumes in L2 ecosystems rise, the increase in average transaction costs due to network capacity will lead to a gradual uptrend in marginal costs (as evident from Base’s performance from March to May). This is a crucial issue that cannot be looked at for the long-term development of Rollups.

In the short term, for Rollups, reducing marginal costs more effectively is the best way to win in the competition. Among the strategies, adjusting revenue and cost models according to market conditions is a good solution.

Revenue Side

To sustain long-term competitiveness, protocols should ideally avoid charging users extra, potentially even subsidizing users to make users’ expenses both low and stable. Although priority fees could increase revenue, sufficient chain activity is required.

Post-EIP4844, some Rollups, like Arbitrum, saw significant revenue declines due to nearly eliminated hidden profits from DA data fees. The revenue model for Rollups becomes simpler, focusing mainly on extracting income from L2 fees.

Overall, the business model of Rollups indeed will benefit from economies of scale, especially for ZK Rollups. The current market conditions are not conducive for Rollups to leverage their advantages.