Not a race to the bottom

Introduction

I have been thinking of the economic impact of EIP 4844 for some times. The highly anticipated upgrade is expected to reduce the callData cost of posting L2 transactions to L1 by at least an order of magnitude through introducing another data format for layer 2 to store information of transactions.

While I agree to the fact that the layer 2 businesses will become incredibly lucrative after the reduction of the L1 cost; what does not change is that the fact that top line fees will be reduced substantially; and the only way to layer 2s to hold valuation or even to become more valuable is to depend on an expansion of trading multiples; in the anticipation of stronger future growth and profitability.

It makes me question why the market is unanimously excited by the upgrade and even force fitting a narrative as we approach the implementation of it. I then become more interested to understand the unit economics of rollups and would love to dedicate this piece to analyse the above questions.

Status Quo of Layer 2 Economics

Top line fees for layer 2 are broadly defined as total gas fee paid by users and gas fee consists of two components; the L2 execution fee and the L1 security fees which refers to the cost of publishing transactions to the layer 1.

Currently the average transaction fee per user for optimistic rollups is at 0.20 USD with an average operating margin of close to 50%; and the numbers for the newly minted Base chain is slightly higher than existing leaders Optimism and Arbitrum. (Thanks Artemis.xyz for the data!)

It is also important to note that operating margin of optimistic rollups increases as transaction count goes up since L2 execution gas becomes the bottleneck; and hence the margin does not linearly scale with the number of on chain activities but exponentially. As friend.tech has been gaining massive traction over the past few months; operating margin of Base is therefore at a premium to Optimism and Arbitrum.

Zero knowledge rollups would also incur an additional cost to verify proofs; and from Dune we could derive the average cost of verifying proofs for zkSync which is at roughly 0.06 USD per transaction; and assuming it has the same L1 data costs as the optimistic rollups we could also derive its execution cost.

For zero knowledge the average proving costs distributed to each transaction in the table is just a rough estimation from the data in the previous two months. The relationship of proving costs decreases linearly with the number of transaction since more transaction could share the proving cost of the same batch.

If we consider the effect of the two above factor, the transaction costs relationship for zero knowledge rollup would actually look like a U curve; when there is too little transaction the fixed proving cost would increase the per transaction cost; and when there is a network congestion L2 execution cost goes up.

That being said, we would only be using the average numbers over the previous two month for simplicity and illustration of the thought experiment.

How Fees Look Like After The Upgrade

Let’s not get into the technical nitty-gritty and say the EIP 4844 upgrade would reduce data cost by an order of magnitude which is 10x or even more as estimated; that would bring the L1 data cost from an average of 0.1 USD per transaction to 0.01 USD.

Assuming the current execution cost / operating margin (in real dollar term) charged by the rollups remains the same and that rollups are passing 100% of the cost savings to users; this would bring the total gas fee per transaction down to 0.10 USD on average and an operating margin of 80% to 85%.

The picture looks a bit different for zero knowledge rollups as the operating margin for zero knowledge rollups are natively lower than that of optimistic rollups even prior to the EIP 4844 upgrade at the 20% level. Assuming similar condition as aforementioned after the upgrade, the operating margin would only slightly improve to close to 40%; and fee per transaction would be at 0.12 USD

Initial Hypothesis on Rollup Economics Post EIP 4844

Now we have two ends of the pendulum; on one end is that rollups decided not to pass users any of the savings and the transaction fees per transaction would stay at the same level; and on another end is that rollups do choose to pass 100% of the savings transaction fees would come down by more than 50%.

Neither do I think it is the right time for rollups to milk fees out of users by passing 0% of the savings; nor do I think they would pass 100% of savings simply because rollups are still projects run by profit seeking organisations; I thought fees would eventually settle somewhere in the middle of the two ends, but where exactly?

Also I thought the elasticity of users switching platforms due to higher gas fees would be would be plateauing for obvious reasons; so I would argue rollups would reduce fee per transactions until volume increase driven by bots activities or native user increase could at least compensate the loss to at least make up for the fee reduction.

How Would Optimistic Rollup Respond?

Optimism and Arbitrum are currently the two leaders in the layer 2 verticals and let’s say they have the absolute pricing power of transaction fees; so how they plan to price transaction fees after the EIP 4844 upgrade would be crucial to the entire market.

Let’s put the volume surge after cost reduction aside; since it could be very tricky to actually quantify the impact even looking at historical upgrade such as Bedrock Upgrade in June 2023. The increase of on chain transaction could also be driven by some native ecosystem development so it would be tough to strip away the effect of organic growth and size the impact separately.

I thought another way we could think about the impact is to size the volume increase percentage required to offset the top line reduction assuming optimistic rollups would really pass a meaningful portion of the cost savings to users.

As aforementioned zero knowledge rollups generally charge a premium to optimistic rollups since users also have to pay for the verification of proofs; and the cost would still exist after the said upgrade. As a result fee per transaction on zero knowledge would still be more expensive than that on optimistic rollups.

Therefore it would be economically logical to assume that optimistic rollups would price transaction fees until a level that they are still competitive enough to be equivalent or even underpricing their zero knowledge competitors; primarily to make up for the reduction of top line revenue.

From the table above we could derive that on top of the reduced data cost and existing operating profit; optimistic rollups would enjoy the privilege of expanding the operating margin at most 0.04 USD and still could be cheaper than or equal to zero knowledge rollups. Under this scenario optimistic rollups would be operating at least 50% operating margin and would only be passing less than 50% of the savings to users.

How Would Zero Knowledge Rollups Respond?

Theoretically zero knowledge rollups should also adjust operating margin downwards to maintain competitiveness against optimistic rollups even with an additional fixed cost in the unit economics stack to try underpricing optimistic rollups if we are in a perfectly efficient economy. However I like to think otherwise.

Prior to EIP 4844 zero knowledge rollups is already operating with a compressed operating margin compared to optimistic rollups; and after the reduction of cost post upgrade the operating margin could slightly improve from 20% to 40%; and hence theoretically speaking they have the privilege to reduce operating margin to at least be as competitive as optimistic rollups if they would like to.

However, I thought it is not to their best interest to try underpricing optimistic rollups because they do not have the cost structure benefit to do that and they could easily operate at a loss if they decided to do this since optimistic rollups could easily defend; and from there on we could see a race to the bottom and optimistic rollups could easily reign.

Also, trying to be competitive would also further eat not only into their top line revenue but also they would need even more volume to compensate the fee reduction. As of now user activities are still mainly on optimistic rollups like Arbitrum and Base not to mention the rollup games have been growing with more newer and more well funded chains launching in the next 12 months.

As a result I thought transaction fee per users would ultimately settle at the aforementioned range; which is close to 0.12 USD; with optimistic rollups operating at an 50% margin; and zero knowledge rollup operating at 40% margin.

In term of volume increase in percentage terms to compensate the top line fee reduction; the numbers do not look outrageously crazy; for Optimism and Arbitrum it only requires an additional 45% on chain activities to make up for it; and of course a part of it could comes from more bots activities since the cost for capturing arbitrage reduces; and also part of the valuation gap could come from the market pricing rollups at a slightly higher multiples.

Concluding Thoughts

In a perfectly efficient economy optimistic rollups are expected to stay competitive by continuously underpricing other optimistic rollups. Therefore fees per transaction could continue to trend downwards. However, like I said above as users’ behaviours are expected to become less sensitive to the decline in transaction fees; the pricing war between rollups would cease as marginal increase of volume plateaus.

If the above logic makes sense optimistic rollups are not going to pass the entirety of the cost savings to users after the EIP 4844 upgrade; but would rather pass roughly 46 of the savings to users. All it takes is an additional 45% transactions to make up for it.

Of course the numbers above are quite static since it’s a function of what has happened over the past two months; we have experienced the social summer on Base driven by friend.tech and Worldcoin’s launch on Optimism which leads to spikes of transaction fees. Also the verification cost for zero knowledge proofs could be quite tricky to model given its relationship with transaction count. But I hope this framework or thought experiment could be an interesting angle to dissect the entire economic impact of the EIP 4844 upgrade.