Musings on the future of rollups

When talking about rollups and layer ones the classical analogy from biology about parents and offspring is not out of place. The parent(L1) secures the child(rollup) until it is big enough to secure itself. Child also resembles it's parent in it's traits, since it is safe to assume many early users who shape its culture are also users of underlying L1, at least in the early days.

Nomenclature For easier understanding and better imagery, I will use the terms "layer one / L1" and "rollup / L2" interchangeably with "parent" and "child" respectively.

Current state of rollups

Currently rollups are either app specific (Dydx, Deversify) or in it’s infancy(Arbitrum, Optimism). The sequencer (block producer) is currently centralized. In generalized rollups there is no token yet and development teams are in control of the protocol updates. Said simply, currently L2 seem like a quite different animal to L1s. However, the roadmaps of all generalized rollups include decentralizing the sequencer (rollup term for block producer) and introducing a token. All major rollup teams are also dedicated to decentralize rollup governance.

Differences & Similarities
Put simply, a fully developed generalized rollup is a separate chain that outsources its security to the parent L1.

Why rollups will become L1s

Inherent tension between children and parents

Classical analogy from biology about parents and offspring is not out of place here. The parent(L1) secures the child(L2) until it is big enough to secure itself. Child also resembles it's parent in it's traits, since it is safe to assume many early users who shape its culture are also users of underlying L1, at least in the early days. In biological systems *children are dependent on their parents, but only up to a point after which they become fully self sufficient. Will it be any different with rollups?

What makes a mature rollup different from L1 is that it outsources its security to its parent chain. It of course has to pay for this in terms of tx costs on the main chain. It would seem like a symbiotic relationship, but therein lies inherent tension. It is often said that every rollup makes the parent block space more valuable, which should be bullish for the whole ecosystem. But the darker side of this dynamics is that the child chain is leaking value to its parent. It's reasonable to assume that a market cap of a successful rollup would increase as its usage increases. However, if rollup token holders are rational and want to increase its utility hence its value, it would make sense for them to leave the parent L1 entirely. Why? They are basically leaking value out to the parent chain when paying for the security. This value could be kept within the child chain by simply keeping its state internally and implementing EIP1559-like fee burn mechanism. But wait! doesn't it compromise the security of a rollup?

The child chain at its mature rollup stage already has its own execution and consensus layer. So if it where to leave its "parent" the only extra work for the validators would be to keep the state of rollups on their nodes instead of posting it to L1. They would loose the security guarantees of L1, but as the market cap of the rollup's token continues to increase, this is less and less of an issue. The question rollup token holders will start to ask themselves is,
How much security is "enough" for our rollup? Do we at this stage really need to rely on the security of the L1 chain?

Depending on the use case, the answer would seem to turn to NO at some point in its life.

Children growing up, Rollup -> L1

Due to the inherent tension described above, L1s could serve only as a springboard for L2s to bootstrap security and leverage network effects of the underlying L1 and then grow up into separate L1s as the chain matures.

There are other forces at play, so what may keep the rollup centric future alive?

Once a child always a child, Rollup centric future

• Cross-shard interoperability If I understand Danksharding correctly, this construction will make possible a transaction executed on one rollup to be conditional on the state of another rollup, put simply, bring back the synchronous transaction across rollups and protocol interoperability, making it feel as if all rollups are just a part of one big chain!!! This is currently lost both on rollups and in L1 to L1 communication. If we really can bring this back to rollups, I would see this as the strongest argument of rollup centric future!!!
• Less friction (not really)
  One really annoying thing with all L1 is that you need to get their token to transact on the network. In rollup you can pay for tx costs in the token of the parent chain. However, there is no fundamental reason why would this status quo not change. Zksync is already making it possible to pay for tx in multiple tokens and there is no reason why "rollup turned L1" would disallow paying for tx in parent's token.
• Political alignment between rollups and L1s If there is a big overlap in the userbase of rollup and its parent, meaning many people use both it may simply be a political decision to stay as a rollup to be more closely aligned. Analogous of what we see nowadays play out, majority of miners are not in fact rational but "honest", meaning they behave by the "norms" of their chain and follow the protocol rules as intended and are not in fact profit maximizing.

Better aligned rollup constructions

If currently publicized rollup construction becomes predatory, it does not mean that rollups are fundamentally broken. These constructions do not lead to rollups becoming L1s, at least not from the reason described above.

• Staying centralized - is it a problem??? There is no fundamental reason why rollups need to be decentralized. The whole point of a rollup is the ability to force exit to parent chain. There are degrees of centralization with even the most centralized one party rollup being quite usable for most use cases. Let's take one party system like DyDx rollup. One party orders transactions, holds state, and progresses the chain. WHEN in the future there are no backdoors to upgrade rollup code, the most DyDx (or any other ZKrollup) can do, is censor transactions or halt the chain. Both of which are not ideal, but with the ability to exit to parent chain, not catastrophic either. There are also different constructions ensuring any / all these functions to be more robust. Data availability could be solved just by dumping the transactions to Celestia or another data availability layer and there could be other incentives in place for sequencers / provers to step in if the main one fails. It can also be a federated system of few sequencers / provers that ensures some redundancy in the system. Or even some fail safe mechanism, that kicks in, if some cog in this machine fails. Thus if a rollup stays fully / partially centralized, it's main security guarantee is the ability to exit to parent chain. This makes it dependent on the main chain, no matter if it has a token or not.
• No token - What is the need of a token in rollup? Not having a token seems to be currently off the table, since everyone loves a token ;) and all VCs & investors want to cash out at some point in a big way. There is however, no inherent need for a rollup to have it's own token. Arguments why tokens are needed usually are along the lines "We need it to decentralize Governance / Sequencer / (choose whatever else)". But obviously for anything a token is needed in a rollup, the token of the parent chain can be used just as easily as the token of the child chain. Put simply, there is no fundamental need for a rollup to have token even if we want to decentralize everything. So if it plays out that rollup introducing their tokens is parasitic to the main chain, new rollups will spring up that will use the codebase of existing rollups and just replace their token with parent token. Then the holders of the parent token would have large incentive to move to this new rollup since it grow their bags and not the bags of VCs of the original rollup.

So how will it all play out?

At this stage, there is little doubt that every rollup will have its own token. Right now, all rollups seems to be very aligned with Ethereum, but there is no token yet. Rollups are far from being mature and will continue to be dependent on the L1 for a long time. The narrative of a rollup centric future is also going strong and will not change over night even when incentives will.

I see two forces one pulling the rollups away from L1s and the other joining them together. The force pulling rollups apart from L1 is high fees for calldata. This could be to a large extent neutralized by drastically lowering them both with EIP4488 and sharding. The other force pushing them together is implementation of cross rollup interoperability (Danksharding), making rollups behave like a part of a super chain rather than a separate chain, thus strictly better to small L1s from interoperability perspective.

In other words, if Ethereum development were to stop today and rollups were to continue developing this would become an issue. However, Ethereum development seems to be picking up steam and going strong.

So lo and behold to the rollup centric future!!!