Deconstructing Rollup Economics: A First Principles Analysis

9 minute read

This essay aims to delve into the economics of four prominent Layer 2 solutions—Arbitrum, Base, StarkNet, and zkSync Era—examining both their revenue and cost sides. Specifically, it analyzes data publishing costs, encompassing the impact of DA solutions and the ramifications of EIP-4844. Additionally, the essay considers revenue generated by L2 native token, an aspect often overlooked in similar discussions.

Table of Contents

Background & Status Quo

Among the aforementioned solutions, Arbitrum and Base employ Optimistic Rollup (ORUs), whereas StarkNet and zkSync Era utilize zero-knowledge proofs (ZK), and they are commonly referred to as ZKRs. This delineation is pertinent to the ensuing discussion on gas and security costs. My previous ZK Rollup market analysis is available here.

Which rollup has made the most money in recent months?

In terms of revenue, Token Terminal¹ reports that Arbitrum held the top position in revenue generation over the past six months, until February 2024, when it was surpassed by Base.

Source: Token Terminal

In terms of on-chain profit, a similar trend is observed, with Base gaining its dominance since March 2024. This essay will delve into the reasons behind this shift.

Source: Token Terminal

In analyzing revenue and cost, this essay adopts the framework proposed by Barnabé Monnot², as depicted in the following chart. Additionally, it incorporates recent developments, such as shared/decentralized sequencers, which were not initially part of the framework.

Source: Barnabé Monnot & Delphi Digital³

Revenue

Revenue of L2 solutions comprises transaction fees, MEV, native token-related revenue streams, among others.

Transaction Fees

Transaction fees are levied on users for processing transactions on the L2 network. Some L2s offer gas-free transactions to incentivize usage, such as Immutable X, although this does not apply to the four L2s discussed here. Transaction fees typically encompass three components:

L1 data publication fee: Direct payment to L1.
L2 operator fee: Actual cost incurred by the rollup in executing transactions.
L2 congestion fee: Additional fees users incur during periods of rollup congestion.

These latter two components are commonly referred to as gas fees for L2. Before EIP-4844, transaction fees can be largely calculated as:

(Post EIP-4844, rollups are anticipated to transition to data blobs instead of calldata.)

It’s worth noting that most L2s impose a gas price floor to prevent network spam. For instance, Arbitrum’s floor ranges from 0.01 to 0.1 gwei⁴, while zkSync Era’s is 0.3 gwei⁵.

Due to differences in technical design, operational strategy, and congestion levels, the four L2 solutions exhibit varying gas fee structures. For instance, in the trailing 90 days, Arbitrum One, Base, and zkSync Era recorded average gas fees per second of $2.40, $4.50, and $1.75⁶, respectively, compared to Ethereum Mainnet’s $130.46. This disparity, coupled with transaction volumes, yields differing revenue figures from transaction fees: $2.15m, $12.92m, and $1.06m for Arbitrum, Base, and zkSync Era, respectively, in April 2024⁷.

MEV

Maximal extractable value (MEV) refers to the maximum value that can be extracted from block production, minus the standard block reward and gas fees. First, let’s consider a basic scenario—a single rollup without any shared services (such as DA or sequencers). In this context, the majority of MEV on this rollup (referred to as “Intra-domain MEV” by Electric Capital⁸) accrues to the rollup itself, as searchers predominantly allocate their revenue to the rollup sequencer via MEV auction⁹. This trend is notably pronounced when the rollup operates a centralized sequencer, where most MEV accrues. Presently, all of Arbitrum, Base, StarkNet, and zkSync Era manage their sequencers internally or through affiliate entities¹⁰ ¹¹ ¹² ¹³, thereby capturing the majority of intra-domain MEV.

The presence or absence of native tokens within these L2 solutions also influences the percentage of MEV accrues to them. Among these four solutions, Base and zkSync Era lack a native token, resulting in a portion of their MEV flowing to the L1 token (ETH). Further discussion on revenue associated with tokens will be provided in the subsequent section.

However, recent developments introduce complexity. For instance, Arbitrum, StarkNet, and zkSync Era plan to adopt decentralized sequencers (DS) or shared sequencers (SS) in the future.

How would SS change the value accrual mechanism?

Before addressing this query, it’s imperative to highlight another category of MEV—cross-domain MEV, which stems from transactions among different rollups. SS facilitates cross-rollup interoperability, and cross-domain MEV serves as an incentive for rollups to adopt SS, as SS can rebate a portion of it back to rollups. While concerns may arise regarding the potential impact of SS on a rollup’s ability to extract intra-domain MEV, Sreeram Kannan previously posited that SS would rebate all intra-domain MEV back to rollups¹⁴. Therefore, in conclusion, as shared services mature, MEV revenue is anticipated to increase, rather than diminish.

Token-Related Revenue

Native tokens can bring revenue to L2¹⁵. Primarily, by issuing newly-minted tokens, provided they hold value, an L2 network can generate additional revenue. It’s important to note that the value of a network or protocol doesn’t necessarily translate into the value of its token (as discussed in the author’s prior article), so the question arises:

How can an L2 token accrue value?

The first avenue is through the payment of transaction fees (gas), akin to how ETH operates for Ethereum L1. This not only furnishes native tokens with an evident utility but also introduces a burning mechanism that reinforces the value of native tokens. However, among the four L2 solutions discussed, only Starknet adopts this approach¹⁶, as the other three exclusively accept ETH for gas, including Base and zkSync, which lack native tokens.

The second method involves endowing governance rights, a feature currently embraced by Starknet and Arbitrum. The value stems from token holders’ capacity to participate in key decision-making processes in the future.

The third avenue entails leveraging L2 native tokens to establish a decentralized sequencer (DS) network. This facet is often undervalued, as none of the four solutions discussed presently employ a decentralized sequencer. The sequencer can be selected through a PoS methodology, as is the case in Ethereum L1. This staking mechanism is a typical way for cryptos to allocate resources and coordinate efforts, essential for constructing DS. With a decentralized network, a distributed group of sequencers and token holders who are staking would earn new block rewards and MEV, thereby creating a revenue stream that enhances the value of native tokens.

Other Revenue Sources

There are additional revenue streams from off-chain sources. Primarily, before an L2 achieves operational revenue balance, external funding serves as the primary source. According to CryptoRank¹⁷, Arbitrum, StarkNet, and zkSync Era have collectively raised $123.70M, $282.50M, and $458.00M, respectively, from private investors. Base, on the other hand, is backed by its parent company, Coinbase. Furthermore, following the token launch, these projects can obtain additional funding by selling tokens from their reserves. For instance, StarkNet has allocated approximately 13%, 18%, and 18% of STRK tokens for research and development, community initiatives, and strategic endeavors, respectively¹⁸. These reserves can be utilized to cover future costs in these areas.

In the long run, L2s may generate revenue by offering developer tools and services for projects developed on top of them. Although currently challenging due to the highly competitive market landscape, this revenue stream may become feasible as the market consolidates and top-tier L2s acquire greater bargaining power.

Cost

On the cost side of the economics, various components are considered below. It’s important to note that certain terms may lack precise definitions or have different interpretations in colloquial usage, which will be clarified further below.

Rollup Operating Cost

Rollup operating costs encompass the expenditures associated with managing the infrastructure and computational tasks essential for the operation of the rollup protocol. This includes the tangible costs of running nodes, such as hardware and maintenance.

From another perspective, the cost incurred by a rollup node is akin to that of an L1 (Ethereum) node. Executing tasks within the rollup also entails expenses measured in gas, which users pay as transaction fees, as discussed earlier in the revenue section. While the cost profile may vary among the four L2 projects discussed, they generally adhere to Ethereum’s gas model, with nuances based on design and usage of EVM.

It’s important to note that this doesn’t encompass the costs L2s incur for utilizing L1, which will be addressed below.

Data Availability & Data Publishing Cost

The major fees that L2 rollups need to pay L1 for is the data availability (DA) & data publishing cost. DA ensures rollup nodes have access to network data necessary for updating the state root on L1. Unlike node operating costs, which resemble those of L1 nodes, data publication expenses are a new element in the rollup economy. After compiling a new batch of transactions, the L2 operator uploads a condensed summary of the set to the base layer. The cost of publishing data is charged by the base layer. Pre-EIP-4844, this summary was posted as “calldata”; post-EIP-4844, a new blob mechanism was introduced to save block space.

Source: GenesiSee¹⁹

It’s notable that while DA cost is sometimes considered a security cost for L2, it serves a broader function by ensuring that transactions (or states) are available to all relevant parties, extending beyond security.

Data publishing costs differ between ORUs and ZKRs. ORUs (Arbitrum & Base) transmit compressed full L2 transaction data (with signatures), L2 state roots, and fraud proofs (only in disputes) to L1. ZKRs (StarkNet and zkSync Era) only need to post state differences, plus a validity proof with every batch, demonstrating the associated state root’s validity.

To delve into the economics of DA & publishing costs, several questions arise:

Which rollup offers better data compression and gas-saving abilities to reduce DA costs?

Both ORUs and ZKRs mitigate DA costs, albeit differently. Generally, ZKRs tend to offer more significant gas savings compared to ORUs, especially with a high transaction volume. While ORUs are cheaper per batch (40,000 gas vs 500,000 gas per batch), ZKRs can compile more transactions into a single proof.

For an apples-to-apples comparison, in the ORUs category, Base employs the zlib algorithm, while Arbitrum uses Brotli²⁰. According to OP Labs, both achieve similar performance, boasting approximately a 39% compression rate and a 43% fee savings²¹.

Source: OP Labs

In the ZKRs category, StarkNet doesn’t employ additional compression to state diffs, but zkSync Era does, theoretically giving zkSync Era an edge in gas savings.

In summary, the gas-saving ability ranking is:

zkSync Era > StarkNet > Base ≈ Arbitrum

How does EIP-4844 affect the DA & publishing cost?

Historically, DA costs related to calldata constituted the largest portion of L2 costs, sometimes exceeding 90%. Proto-Danksharding, also known as EIP-4844 implemented in March 2024, introduced new features to significantly reduce rollup fees. This upgrade provides dedicated storage for rollups to publish transaction data in ‘blob’ format. Unlike calldata, which is permanently stored, blobs are removed after 18 days. EIP-4844 also introduced a new “data gas fee market” to determine the price of blob transactions²².

Blobs now offer a cheaper way for Layer 2 Rollups to post transaction data to Ethereum, reducing transaction costs by more than 90% for some rollups. To illustrate the cost savings, below are charts showing the composition of on-chain costs for Arbitrum and zkSync Era (denominated in ETH), where the blue area represents the cost of calldata, which becomes almost invisible after EIP-4844:

On-chain costs composition for Arbitrum. Source: L2BEAT²³

On-chain costs composition for zkSync Era. Source: L2BEAT²⁴

What are the alternative solutions for DA? And what are the tradeoffs for those solutions?

Currently, Ethereum rollups often rely on Ethereum base layer for data availability due to its high level of security and decentralization. However, due to the high cost mentioned above, several alternative solutions are under development. Some notable projects include:

EigenDA: Based on re-staking principles, EigenDA draws upon core concepts and libraries such as the Ethereum Danksharding codebase. It is often regarded as the next most secure option to Ethereum.
Celestia: Constructed by utilizing the Cosmos SDK.
AvailDA (previously known as the Avail Project): A low-cost, expandable blobspace DA solution.

Additionally, Arbitrum, StarkNet, and zkSync Era have introduced their own DA solutions: zkPorter, Arbitrum’s Anytrust, and Starkex DAC. Most of these solutions are off-chain DA, except for EigenDA. There is a security-cost tradeoff in the spectrum below:

General-purpose rollups and app-specific rollups may choose different solutions based on their preferences:

To sum up the answers to the few questions above, although DA & publishing costs traditionally make up the majority portion of L2 rollup costs, EIP-4844 and emerging DA solutions will significantly reduce these costs.

Proving Cost

There is a proving mechanism, sometimes referred to as a challenging system, for rollup transactions and L2 state roots. Using cryptographic proofs, the correctness of rollup transactions and state roots can be verified by L1.

There are two types of proving mechanisms, each with different costs:

Validity proofs, used by ZKRs, are employed to immediately verify the validity of transactions. It is a fixed cost regardless of transaction activity since it is submitted with every batch to Ethereum L1.
Fraud proofs, also known as fault proofs, are utilized by ORUs. ORUs assume all transactions are valid, and fraud proofs present evidence that a state transition was incorrect. This is a variable cost, occurring only when there are challengers. On average, this cost is negligible for ORUs using fault proofs.

Source: BNBChain.org²⁵

This is also why ORUs, in general, have a lower fixed cost than ZKRs, although the tradeoff is the dispute time delay. Aligned Layer is a project working to bring down the proving cost for ZKRs. Aligned Layer receives proofs from different validity proof systems, verifies them, sends the final result to Ethereum, and posts the data into a DA layer²⁶ ²⁷. It provides a decentralized network of verifiers, offering fast and affordable verification of proofs.

Off-Chain Costs

In addition to on-chain costs discussed above, there are a few off-chain costs. Off-chain here refers to costs incurred by the foundation or business entity behind that L2. These include:

marketing and promotion
legal and compliance
research and development
and other overhead

Concluding Thoughts

To summarize, Arbitrum, Base, StarkNet, and zkSync Era derive revenue from transaction fees, MEV, token issuance, and off-chain funding, while incurring costs related to rollup operation, L1 publishing, proving mechanisms, and off-chain business operations, as depicted in the chart below:

Ideally, an L2 rollup should not only achieve an overall surplus (profit) but also aim for a surplus from transactions, as indicated by the red triangle above. This approach can support token price stability and reduce volatility.

In particular, this essay concludes that DA and publishing costs have decreased due to both EIP-4844 and alternative DA solutions, which are more cost-effective. On the revenue side, transaction fees are expected to decrease as DA costs decrease, emphasizing the growing importance of MEV in the future revenue of rollups. The emergence of DS and SS introduces some complexity. DS may reduce MEV on L2 rollups but can also enhance the value of L2 native tokens. SS is likely to facilitate cross-domain MEV and offer rebates to L2 rollups.

Returning to the initial question posed at the beginning of this essay—

why did Base outperform other L2s in terms of revenue and profit starting from March?

The answer also lies in EIP-4844. Louthing pointed out that the OP Stack would prevail in the EIP-4844 fee war ²⁸, resulting in the most significant reduction in DA costs. Base, being one of the most mature L2s built on the OP stack, was well-positioned to onboard new users, leading to a surge in activity following the EIP-4844 update.