Written by: Roy, Lyv, Aki Network Research
Just a month or two ago, several major updates of zkEVM came out one after another, and more and more discussions began to wonder: Does zkEVM already represent the future direction of Ethereum, or is it already the limit state of Ethereum?
However, the development of Ethereum's technical route always implies interesting business logic, and is often accompanied by compromises and regrets. After Aki Network Research conducted an in-depth study of the driving force behind Ethereum's current L1-L2-L3 structure, we found that we may need to re-examine the previous questions. In fact, the problem we face may be more extensive: Does the Ethereum universe represent the final form of blockchain?
1. Why Layer 2？
Ten years after the launch of the Bitcoin network, our blockchain world still has not solved the capacity problem.
The TPS of Ethereum mainnet is only in the low double digits now, which is not only several orders of magnitude lower than the target TPS of 100,000 of ETH upgrade route, but also far behind the tens of thousands of TPS of mainstream Visa/MasterCard payment networks in traditional finance. Under such a processing speed, Web3 represented by blockchain is still moving forward under heavy burden.
Obviously, the first problem is that as more and more people use the Ethereum network, the network will become more and more congested with limited TPS, and the average confirmation time for each person's transaction will be longer. The second problem is that since Ethereum's underlying transaction confirmation is a bidding mechanism, there are only two situations for gas fees: getting more and more expensive, and suddenly becoming very expensive.
A network must at least be usable, that is, it must be able to quickly and cheaply complete basic information exchange functions, so that the significance of the network's existence and the discussion of its future development will be meaningful. In order to solve the blockchain capacity problem, the industry now has two solutions: one is the various Layer 2 related solutions we will talk about today, and the other is Danksharding sharding, which we will talk about later.
After the expansion, Ethereum gas fees will be cheaper, and more importantly, after the gas fees are cheaper, many things that were previously unfeasible will become feasible, and this is what we are more interested in.
2. What should Layer 2 look like?
This problem is not only a technical problem, but also a political and economic problem, so let us define it clearly here:
It is at least a blockchain. This means it can issue native tokens and have a complete community, which can retain users who are loyal to this L2 and create a foundation for its own Layer 3.
It inherits the security of the Ethereum network from L1.
Its purpose is to expand the Ethereum network.
Putting points B and C together, we can see that the reason why Layer 2 is safe and reliable in theory is that it inherits the security of Ethereum itself. Layer 2 publishes data to the Ethereum mainnet to achieve the final Settlement. This is also the most important point that distinguishes it from various previous expansion solutions represented by Plasma, that is, Data Availability can be achieved on the mainnet.
From an economic perspective, Ethereum "divides" its economic value, and one of the anchors of Layer 2 tokens is that they nominally work for Ethereum. Layer 2 itself is not a unique ecological form of Ethereum. For example, Sei Network, as the L1 of the Cosmos ecosystem, is currently working hard to develop SVM's Layer 2 C. For other L1 blockchains, this idea and demand are not obvious for the time being.
3. What does Layer 2 on Ethereum look like now?
3.1 The most popular L2 expansion solution: rollups
Many people have used Rollups, and its technical routes are divided into Optimistic Rollup and ZK Rollup. In the former route, there are many well-known chains such as Arbitrum/Optimism, etc., and everyone should have interacted with various famous projects under them.
Although OP Rollups has reduced the cost of each transaction to 5-10% of the mainnet through the execution layer off-chain, the poor user experience during the Arbitrum airdrop shows that we still have a long way to go. After all, hundreds of thousands of people interacting with L2 in a few hours can increase the gas fee of each L2 transaction by 5-10 times, which almost completely offsets the potential of EIP-4844 to save 10 times the transaction fee for L2. And a smooth user experience under such intense network pressure is the most basic requirement for Web3 to welcome the next billion users.
ZK Rollup has been very famous in the past two years, but relatively few things have been implemented. This is due to the compatibility problem of EVM. For details, please refer to our previous article "Zero-knowledge Proof and zkEVM: Where did it come from and where is it going?". However, the good news is that this problem has been basically solved now. ZK Rollup and the former are at least on the same starting line in terms of development difficulty. We can foresee that many developers are willing to deploy their own dapps on ZK L2.
On the user side, ZK Rollup is the L2 of the Ethereum mainnet, and its shorter withdrawal waiting time and more mathematically secure background are very attractive. However, from the commercial perspective of ordinary users, the attractiveness of withdrawal time and security needs to be considered. As the L2-native Web3 ecosystem continues to prosper, more and more new users may become "L2-native users", and may not even need to have any knowledge of L1 transactions and ecosystems (imagine a new user of StepN based on L2). The demand for L2 to cross-chain back to L1 may become increasingly marginalized and low-frequency in the future as the L2 ecosystem becomes more and more abundant.
3.2 Having discussed the theoretical aspects, where are we at in terms of L2 in practice?
OP Stack: In order to compete with ZK Rollups armed with zkEVM, Optimism, a member of the OP Rollups technology route, launched OP Stack. We hope to make the pie bigger. We believe that the idea of OP Stack can be compared to the multi-chain shared security and decentralized ecosystem that the Cosmos ecosystem wants to achieve.
It is worth mentioning that OP Stack is a highly open technology stack, but its economic convergence is very low. It is even possible that a certain L2 developed based on it has only a very weak connection with the Ethereum ecosystem, and at the same time has no direct economic relationship with Optimism and is not connected to the OP shared sorter - of course, it is possible that the OP shared sorter can have a huge economic and politically correct persuasiveness after decentralization.
For details about various types of zkEVM, see: "Zero-knowledge proof and zkEVM: Where do we come from and where do we go?"
Starknet launched a Type 4 expansion solution, which is different from all the solutions in B in that they are not EVM compatible. The advantage is that in theory they are indeed faster and cheaper than Type 2 solutions. For developer friendliness, the Starknet team made the following improvements:
Cairo 1.0: This is the most significant upgrade of Cairo since the release of Cairo in 2020 by the Starknet team. Testnet is now online. Cairo 1.0 is a high-level language that looks a lot like Rust, which means that Cairo 1.0 has gotten rid of the shackles of the original Cairo and has greatly enhanced developer friendliness. But this also means that many developers have lost their place to learn Cairo after more than two years. On the other hand, the new Cairo 1.0 is not yet complete, and it seems that it will take another 3-6 months to be more complete.
KAKAROT: This is an EVM solution based on Cairo 1.0 with a combat power of up to 9000 (according to the developer), but it is just getting started.
4. Now that we have talked about Layer2, what is L3?
4.1 First of all, we need to answer, what problem is the L2 to L3 structure trying to solve?
Let's first look at the economic principle of Rollup: Rollup needs to pay a very high fixed cost to submit a batch to the chain. If the overall traffic of L2 is not large enough, this Rollup will either choose to wait longer to integrate a batch of high enough value, or increase the transaction cost of each participant in this batch to cover the cost of packaging this batch. This situation is like carpooling, either waiting for everyone to arrive, or paying more money to the driver to leave early. Then the ideal state is to have enough people to carpool, so that as a customer, the waiting time is short, the average cost is low, and the driver can also make more money in the same amount of time.
In such a scenario, L2 actually solves this problem: there are many ideas that have insufficient traffic in the initial stage or low value of a single transaction. If they are directly deployed on the main network or build an L2, it is not economically feasible; Choosing to become an L2 L3 can significantly reduce your own costs. This itself is an economic consideration. The underlying logic is very similar to a dApp's choice of whether to deploy it directly on Optimism or use OP Stack to create a dedicated L2.
Only when the infrastructure is cheap enough can the ecological environment flourish. On the other hand, this question itself also answers the current situation of the blockchain industry from one perspective, that is, the application ecology dominated by exchanges and DeFi Lego, because only projects that directly involve money can rationalize the operating costs of the blockchain network.
4.2 The real logic of each company’s L2 and L3 grand plan
Let's go a little deeper. This is not just a simple carpooling-departure model. Because of the magic of tokenomics!
Let's take a simple example. L2 centralized sequencers are doing a good job now, so why must each household's roadmap be decentralized? Of course, security factors are a natural political correctness, but what most people avoid talking about is that decentralization is the only way to make a profit.
Only after the decentralized structure is established, can there be a token staking mechanism for validators and sequencers, and verification fees for provers and verifiers, and can the native tokens of L2 be circulated through the L3 ecological development. Whether it is Op Stack or Arb Orbit, and the later L3 concept of the zk series, etc., they all follow the relatively traditional business model of the blockchain industry and are not innovative concepts.
Traffic in Web3 is money, which is actually even simpler and more direct than Web2. The original purpose of L2 was to expand the capacity of L1, so they naturally inherited the traffic of L1. Then the question is, since it is the traffic provided by L1, most of it will eventually return to L1. Because traffic is money, what is given to you is the native assets of L1, so it will always return to L1 in the end. In the blockchain industry, this phenomenon is explained as "inheriting the orthodoxy of Ethereum." That is, only assets that can be confirmed by L1 in the end are real "money."
The word "legitimacy" sounds very feudal, and the hierarchical relationship of L1-L2-L3 actually has subtle similarities with the feudal system. Liquidity goes from L1 to L2, so it is difficult for L2 to truly exist independently of L1, and the interaction between L1-L2 is not simply reflected in the deposit/withdrawal from L1. The truth is that every transaction of L2 strengthens the credit and value of the L1 native token, which can be called the seigniorage of L1. This also explains why all Ethereum L2s are now using ETH as the gas fee payment currency.
However, the logic of L2-L3 is a little different from that of L1-L2. The birth of L1-L2 is to expand the capacity of L1, while L2-L3 is to divert traffic to L2. This is a very realistic problem, because the current traffic of L2 is not large enough. Going back to the logic of carpooling, drivers cannot send cars efficiently. And facing an already large ecosystem, such as Coinbase's Base, L2 such as Optimism finds it difficult to incorporate it into its own L3 development ecosystem. It can only try to capture the huge value of the former through the cross-chain convenience brought by Op Stack (potential user diversion) and the ecological development fund contribution promised by Base.
4.3 Rollup-over-rollup or Validium is expected to dominate the L3 expansion solution
Before we start talking about this topic, we need to explain this question first: If ZK Rollup is so excellent, is it possible to develop a multi-layer ZK-Rollup structure to above L3, or even achieve unlimited expansion?
Let me first state the conclusion: it cannot, because even if the complexity of ZKP computational proof can be continuously reduced through ZKP nesting, the degree of compressibility of the data itself is limited.
We look at this from two perspectives: one is the process of computational proof, and the other is data (availability).
In our current Rollup solution, no matter which route is used, the process of computing proof itself can be rolled up. In the ZKP scenario, the Prover computes proofs for each block, and the Verifier only needs to perform a little calculation to verify the correctness of the Proof. In this scenario, it is theoretically feasible to make a ZKP that proves the "correctness of the ZKP".
But the situation is quite different when it comes to data availability (DA).
Because the data must be uploaded to the main network so that it can be verified and checked by users. Under this premise, Rollup nesting is meaningless. The basic principle of Rollup is to compress the transaction data that needs to be stored on the chain. The advantage of this is that the overall data becomes smaller, so the cost is reduced. However, such compression has a limit: the "bottom line" of compression is that data availability (Data Availablility) holds, that is, any user can theoretically use these compressed data on the chain to independently reproduce the changes made by this Rollup to the main network state to verify its correctness.
If we make a ZKP to prove the correctness of ZKP, then from the perspective of DA, we need to upload the compressed content of each ZKP to the chain. In this way, the workload is not reduced, so it is better to just make a whole ZKP.
But this does not mean that it is technically impossible to do a "Rollup over Rollup" structure. In summary, in the same layer or in the same dApp, the simple stacking of Rollups is meaningless.
In the L2-L3 scenario, L3 uploads the transaction batch to L2, and L2 packages several different L3 batches and L2 native transactions into a new batch and uploads it to L1: In this way, L2 performs ZKP of "several ZKPs from different entities". In this scenario, although the compression of the data itself has reached the limit allowed by DA, from an economic perspective, merging the economic value of several batches to exceed the lower limit of the gas fee required for packaging and uploading to L1 (we mentioned this departure principle in Section 4.1) is a commercially reasonable behavior, which can greatly reduce the transaction costs of each participant.
Because we think the possible structure of Layer 3 is Rollup-over-rollup or validium:
A. Top-down structure: Vitalik’s idea is that L2 is a widely applicable General Purpose Rollup, while L3 needs to provide more customized services: either for special purposes, such as privacy computing; or a special rollup built for the special data structure of some dApps; or validium, which is cheaper but requires weak trust assumptions.
B. Bottom-up structure: If we consider a more App-oriented narrative, assuming that a dApp as Ethereum L3 wants to gain stronger autonomy after becoming popular, it can have many options:
Cosmos app chain - becomes L1 directly, almost completely independent;
Op Superchain / zkSync Hyperchain - Become a parallel L2, with a high probability of participating in the shared sorter and maintaining economic system connections;
Maintain the status of L3, but issue coins and follow the L2 decentralization roadmap again;
The essence of route B is that after dApp generates huge traffic, it needs to consider its own development interests more than the ecological value brought by the platform, especially if the dApp itself is the leader in the platform. Here we can optimistically assume that as the infrastructure is further improved, non-financial dApps and dApps that create more real scenarios that utilize Web3 and have real use value can completely take this path.
4.4 What is the significance of L1 in the L2-L3 structure?
Speaking of this, if L3 realizes large-scale Web3 migration in the form of App Chain as envisioned, and truly generates traffic that is economically meaningful to L2, then this traffic is the real native asset for L2, that is, the resources that are truly born and raised here, and its assets are provided with the so-called "legitimacy" by L2.
Here we have to ask a philosophical question: if there are enough assets that will never return to L1, then why do we need L1?
The first aspect is that L1 provides valuable traffic at the moment, which has been mentioned a lot in the previous article. The second aspect is that L1 provides credit endorsement, and L2 places the final settlement on L1, so users can tolerate L2's high degree of centralization for performance.
If a certain L2 is declared to run independently from L1 one day, it is not unimaginable as modular blockchains are becoming more and more popular. Then, this L2 will have to become similar to the current Ethereum network, bearing the tests, challenges and costs of security, decentralization and other dimensions.
Therefore, from a technical and economic perspective, the "enclave system" of L1-L2-L3 is seamless and interconnected.
4.5 Now that we have discussed the structure, let’s talk about the current application situation.
When a new public chain starts operating, the fastest-running ones will inevitably be scam projects that are just reskins of scams, because they do not have any actual workload.
Most public chain project owners have no way to deal with this: after all, to open a snack street, stall owners need to be stationed. If someone is running fast and acting like a real person, you should not greet him with a smile, and it would be inappropriate not to support him. As a result, he ran away after you supported him, and now the users are blaming you. This can only be described as the pain of development. For public chain projects like Starknet that are used to running slowly, such pain may have to wait until the Cairo 1.0 document is written clearly before they can experience it.
Let’s talk about the good direction. What new applications can we expect in the L2-L3 architecture?
Smart wallet: As the technical debt of the underlying protocol before L1, it is mainly because it is expensive and troublesome to deploy and operate a smart contract on L1 as a wallet for individual users. After improvements made through ERC-4337, it is believed that AA at the L2 level will be implemented more smoothly because the operating cost of L2 is greatly reduced.
Chain games: With the support of Validium, support for sub-cent scenarios (gas fees for a single transaction are less than one cent) makes it possible to have truly fun normal games on Web3.
Validium is a weak-trust expansion solution, in which "weak trust" is reflected in the fact that its prover node theoretically selectively provides data for user verification. In the context of blockchain games, this is a point where concessions can be made for efficiency and performance, because not every scene in the game has great economic value, such as the player's non-property operations (such as walking, talking, and fighting).
Such blockchain games can be widely used in various types of L2 or L3, and the deployment form is mainly considered from the perspective of the game itself and the data structure, similar to the economic considerations we mentioned earlier.
Finally, let's talk briefly about the recently popular Rollup-as-a-Service (RaaS): Web3 services themselves can capture a lot of value, which is completely different from the scenario of web2. Imagine that you rent a server on AWS and open a portal website. You only need to pay AWS rent every month.
However, the world of Web3 is different. Not only do you have to pay rent to AWS, but you also need to pay protection fees represented by MEV for daily operation and every time you update and upgrade the website! Instead of feeding this value to the RaaS provider, it seems to be a more economically logical choice for dApp to deploy an application chain by itself, becoming a parallel L2, or an L3 hanging on L2.
5. L2-L3 is so good, what is the cost?
5.1 Centralization
A. Screening and KYC
The current status of various Rollups is that their sequencers are very centralized.
The underlying blockchain uses private key signatures to send information, which ensures that the sequencer cannot forge a transaction. However, the centralized sequencer still has enormous power. For example, it can choose to refuse to package transactions of a certain type of user, sort all transactions according to its own interests, and decide whether to accept external assets brought by cross-chain protocols to participate in transactions.
For example, in Base, L2, in addition to the sandwich attack common in centralized sequencers, Coinbase may require that only users who have passed Coinbase KYC can transfer funds to BASE. Base can also use whitelists to refuse transactions that package funds of non-KYC users to enter L1, thereby blocking the possibility of non-KYC users withdrawing funds from BASE.
B. MEV
This is a problem that all blockchains need to solve. Because L2 is not used by enough people, many mitigation or democratization MEV solutions that are feasible in L1 are not feasible here, while many solutions to grab MEV that are no longer feasible in L1 are still feasible in the current L2. A centralized sequencer system will inevitably lead to such a result, and the decentralization of the sequencer itself requires highly qualified project parties to take the lead in leading the industry direction.
5.2 L2 is still not cheap enough
Even if we realize all the visions of L3 expansion solutions represented by Rollup-over-rollup or validium, we still cannot solve all problems. The cost of a single interaction still cannot meet the sub-cent scenario; after the demand increases, the cost will still rise.
5.3 Client Diversity
Ethereum does not specify what the block generation and verification clients should be like. The advantage of this is that if the clients are diverse enough, then if one client has a bug, it will not affect the overall security. However, various L2s are still in the early stages of client diversification, after all, the decentralized roadmap of prover/sequencer is still in the conceptual stage.
5.4 Upgrade and Update
If the mainnet needs to be upgraded and updated, L2 must also be upgraded. This will bring great challenges to the governance and trustless goals.
6. End Game and End “Game”
At this point in this article, the exploration and analysis of the L1-L2-L3 structure is basically complete. As mentioned at the beginning, the question we are really trying to answer is: Is the Ethereum universe the end of blockchain?
6.1 The Ethereum universe is most likely not the end of the blockchain industry
The simple answer is no, we believe Ethereum Universe is only a medium-term solution for Web3’s financial needs.
First of all, we believe that the current L1-L2-L3 structure has a low mid-term performance ceiling under the constraints of the blockchain impossible triangle, and it is difficult to support the high-frequency, low-economic value interaction needs of large-scale users, such as the scale of social networking and games in Web2. In essence, human behavior itself also follows a strong head effect, that is, most of the economic value is concentrated in a very small number of actions.
Blockchain cannot be just about finance. Web3 itself is a concept created for blockchain to truly go beyond its original scope. Therefore, the industry cannot just play a zero-sum game in a small circle. At present, games are one of the few (if not the only) tracks that can go beyond their original scope, because they can create demand out of thin air, specify rules, and create user stickiness without the support of strong real-world scenarios.
In simple terms, Web3 games are currently divided into several categories:
Serious traditional games, stripping the economic system and putting it on the chain, this range is very wide, from casual chess and cards to turn-based battle games, to higher-dimensional SLG, RTG, and even open world
Make the game lighter, focusing on X to Earn
After the game is developed, NFT will be issued first to see if IP can be created first to provide funds for subsequent development.
Full-chain games, actual delivery such as low-end Minecraft or Battle Royale on the chain
Making the Web3 characteristic gambling model into a gamified form
How to convince traditional game manufacturers to take various economic and legal risks and costs, abandon existing distribution channels and profit methods, and operate a brand new system is a major bottleneck in the development of Web3 games. Web3 games developed by small teams are faced with extremely strong homogeneous competition and the problem of Web3 user growth.
The other four solutions seem to require strong economic drivers to maintain healthy operation. As mentioned earlier, most of the traffic (or liquidity) borrowed from L1 must be repaid, and the pure Ponzi model will not retain real users. If everyone comes for money, there is no need to talk about ideals, let alone making the game more fun.
So, taking a step back, if we don’t aim to make money directly, can we just focus on making games? We think this approach may be one of the correct answers.
However, as mentioned earlier, the current L1-L2-L3, the hierarchical structure of Ethereum, brings economically driven traffic, which is essentially to reap the highest APY/ROI in the system over a period of time. This is a different manifestation of the same economic driving force as the sector rotation, NFTFi, Chinese concepts, meme craze, etc. that appeared in this bull market.
At the same time, this structure cannot technically support Dapps with high frequency and low value interactions such as games. In other words, launching games on the popular L2s on the market may not be the best solution. This is the biggest paradox that this idea needs to overcome - places with too much liquidity are a double-edged sword for the durability of games.
If we have done a satisfactory job in creating the demand for web3 finance and solving the problems of web3 finance, then we still have a long way to go in terms of games. It’s not that there are no good games, but there is no suitable soil for these games to grow.
6.2 If the answer to finance is Ethereum L123, what is the answer to games?
The Gamefi track is in a period of confusion, and the lessons learned from the past are all wrong answers. However, from the perspective of elimination, Ethereum's economic segregation system is not suitable for the birth of truly playable games. We can place our hopes on a structure that is more economically loose, has lower requirements for the availability of some data, and is more tolerant of centralization. For example, Cosmos, Binance Greenfield, and Starknet's zkVM.
Some people think that due to the progress and innovation of L2/zkEVM/L3, the public chains of non-Ethereum system have lost their value. We think it is just the opposite. Their greatest value is that they are not Ethereum. The characteristic of the blockchain world is the tendency of self-disintegration, because capital flows extremely freely, technology is open and barriers are limited.
Ethereum maintains the centripetal force of the system with its economic enclave system, but this expansion is not unlimited. On the other hand, the Ethereum system may not like low-value use cases, and intentionally or unintentionally keeps the threshold at a level where finance can survive and other industries struggle to move forward, so the price of its native currency cannot be low.
In short, the Ethereum universe may be able to capture most of the behaviors with extremely high economic value in Web3, but it may be overkill for random behaviors with low economic value. Some people may think: "Not all behaviors need to be on the chain, the cost of chaining is too high." But in fact, this sentence should be expressed as: "Not all behaviors need to be on the Ethereum ecosystem, Ethereum has its own tasks and missions." This means that the ecosystem outside of Ethereum, or the remote areas of the Ethereum universe, may also grow unexpected stamens.
