Amid the AI craze, what changes have taken place in the public chain infrastructure track?

Recently in the primary market, the hottest track is undoubtedly AI, followed by BTC. 80% of the projects discussed every day are concentrated in these two tracks. Personally, I can talk about 5 or 6 AI projects a day at most.

It is foreseeable that the AI bubble will reach its peak in the next two years. With the launch of hundreds of new AI projects, the market value of the AI track will reach its peak. When the bubble finally bursts and everything is in chaos, unicorns that truly find the fit between AI and Crypto will be born, pushing this track and the entire industry forward.

Therefore, in the current overheated AI environment, let’s calm down and take a look at what changes have taken place in the Infra level in recent months, especially in the public chain Infra track. Some of the new things are worth talking about.

1. ETH, or further deconstruction of the monolithic chain

When Celestia first proposed the concept of modularity and the DA layer, the market actually took a lot of time to digest and understand it. Now this concept has long been deeply rooted in people's minds, and various RaaS infrastructures have spread to an exaggerated stage where the number of infrastructures > the number of applications > the number of users.

The execution layer, DA layer, and settlement layer have undergone some different technical progress in the past few months. Each layer has derived new technical solutions, and even the concept of the settlement layer is no longer exclusive to ETH. We will briefly talk about the representative technologies of each layer.

2. Execution Layer

The hottest concept in the execution layer is undoubtedly Parallel EVM — represented by Monad, Sei, MegaETH, and some existing projects such as FTM and Canto are also planning to upgrade in this direction. However, just as not all ZK projects will protect privacy, the projects labeled Parallel EVM actually have different technical routes and ultimate goals.

Take a picture of Sei as an intuitive demonstration. It is obvious that under optimistic circumstances, the performance improvement is still very obvious when changing from the existing sequential processing to parallel processing.

Parallel EVM can actually be divided into several different technical routes.

From the perspective of how transactions are parallel — There is nothing new under the sun, except the difference between a priori and a posteriori

Solana and Sui are representatives of a priori, requiring transactions to explicitly declare which parts of the chain state they have modified, so that before packaging the block, it can detect whether there is a state conflict (such as access to the same AMM pool), and if so, discard the conflicting transactions.

Post-hoc is also called optimistic parallelism, represented by Aptos BlockSTM, which means that transactions are accepted first assuming that there is no conflict, and then tested after execution. If conflicting transactions are found, the transaction is declared invalid, the result is refreshed, and re-executed, and this step is repeated until all transactions in the block are executed. Sei, Monad, MegaETH, and Canto use similar solutions.

We have also seen parallel solutions in the primary market for state conflict scenarios (such as access to the same AMM pool mentioned above), but the project seems relatively complicated and we are not sure whether it is commercially feasible. It is still under evaluation.

**Based on the degree of attention paid to Parrallel EVM - **it can also be divided into two schools

• One is Monad, represented by Sei, which takes how to trade in parallel as the main expansion idea, that is, Parallelization is the main narrative. For example, in addition to optimistic parallel processing, Monad also has a specially developed MonadDB, and asynchronous I/O is specifically used for parallel processing.
• Another is the idea of Fantom, Solana, and MegaETH. Parallelization is one of the expansion solutions, but it is only one of them. Parallelization is an auxiliary narrative, and performance improvement depends more on other technical solutions.

For example, Fantom's Sonic upgrade focuses on the FVM virtual machine + and the optimized Lachesis consensus mechanism. Solana's next phase will focus on the modular architecture of the new Firedancer client, optimized network communication mechanisms and signature verification, etc.

The goal of MegaETH is to realize Realtime-blockchain. First, based on the newly developed Reth high-performance client of Paradigm, further optimization and improvement have been made in the state synchronization mechanism of the full node (only synchronizing the state difference instead of all data), the hardware design of the Sequencer (a large amount of high-performance RAM with storage function for state access to avoid slow disk I/O), the data structure improvement of Merkle Trie, etc., which is equivalent to a comprehensive and all-round improvement in software, hardware, data structure, disk IO, network communication, transaction sorting and parallel processing, pushing the performance ceiling of EVM to the limit and approaching "Realtime Blockchain".

Three. DA layer

There is no particularly large technological iteration at the DA layer, so the competition in this track is far less intense than that at the execution layer. After all, there are only a few main players.

ETH's CallData has been upgraded to Blob, and the fees of each L2 have dropped significantly. Now ETH has become a "not expensive" DA.

Celestia’s greater role is that after its launch, as the first project to propose the concept of the DA layer, it raised the DA track from the ceiling of 2 billion FDV to 20 billion, and from then on, the pattern and imagination were opened up. Many new Layer2 Appchains naturally chose Celestia as their first choice of DA.

Avail has become independent from Polygon. Technically speaking, it is more like an "enhanced version of Celestia". For example, it uses Polkadot's Grandpa+BABE consensus mechanism. Compared with Celestia's Tendermint, it can theoretically support the decentralization of more nodes. It also supports Validity Proof, which Celestia does not support, etc. Of course, the technical differences are far less important than the ecology. Avail still needs to catch up in terms of ecology.

EigenDA was also launched together with the EigenLayer mainnet two days ago. As one of the strongest narratives and the most commercially cooperative projects in this round, I personally feel that the adoption rate of EigenDA will not be low. Theoretically, as long as it "feels safe and cheap", not many projects really care whether you use Validity Proof or Fraud Proof, whether DAS supports it, etc.

What is more worth mentioning is the following three DAs

• Near DA — Near is a magical public chain. It was originally designed for sharding and is still doing it now, but while doing sharding, it also developed DA — it is cheaper than Celestia and supports fast settlement of L2; chain abstraction — Near recently launched chain signatures, allowing users to request signatures for transactions on any chain through a single NEAR account; AI — Founder illia is one of the eight Transformer members, the one who was patted on the shoulder by Mr. Huang at the NVIDIA conference. He is now planning to hire AI engineers and will release a near.ai related announcement next month... Hexagonal warrior, I'm also thrown into the DA track.
• BTC&CKB — Because the first layer of BTC does not support smart contracts and cannot be directly settled, dozens of BTC EVM Layer2s are now basically using BTC as DA. The only difference is whether to throw ZK Proof directly onto BTC or throw the Hash of ZK Proof onto it, as if they cannot call themselves "BTC Layer2" if they don't do this. Recently, I actually encountered a new project saying "I don't pretend anymore, I am ETH L2, DA settlement is all on ETH, but I serve the BTC ecosystem!", which is quite funny... The only alternative expansion plan is RGB++, which was withdrawn by CKB. In this framework, CKB has become a DA-like existence, and BTC has become the settlement layer of RGB++ because of the black technology of UTXO isomorphic binding.
• New DA — Let me talk about two new ideas for DA that I have seen, but I won’t mention the names of the projects. One is to combine DA with AI. In addition to being a high-performance DA, it can also serve as a storage layer for large AI models, training data, and training trajectories. The other is to improve the error correction code mechanism of the underlying DA such as Celestia, which can provide a more robust network status in an unstable state such as a dynamic network (several nodes are randomly disconnected in each round).

Four. Settlement Layer

Originally, this layer was almost exclusively owned by ETH, while DA had competition from Celestia, which had its own L2 implementation. However, other chains like Solana and Aptos do not have L2 for settlement, and BTC’s L2 cannot be used and BTC cannot be used for settlement. Currently, ETH is basically the only settlement layer you can think of.

However, this situation will soon change. We have seen several new projects moving in the direction mentioned at the beginning of the article, and some old projects have also begun to transform in this direction, namely - ZK verification/settlement layer - to further deconstruct ETH (steal ETH's business).

Why did such a concept come about?

The reason is that running a contract on ETH L1 to verify ZK Proof is not the best choice in theory.

Technically, in order to verify the correctness of ZK Proof, developers need to write verification contracts based on Solidity according to the ZK project and the ZK Proof they choose. This requires relying on a lot of cryptographic algorithms, such as the need to support different elliptic curves. These cryptographic algorithms are usually relatively complex, and the EVM-Solidity architecture is not the optimal platform for implementing these complex cryptographic algorithms. For some ZK projects, the cost of writing and verifying these verification contracts is also very high.

This has hindered some ZK ecosystems from natively joining the EVM ecosystem to some extent, so ZK-friendly languages such as Cario, Noir, Leo, and Lurk can only be verified on their own Layer 1. At the same time, it is always difficult to turn around when updating or upgrading these things on ETH.

In terms of cost, although the “protection fee” DA paid by L2 accounts for the majority, ZK contract verification also requires Gas fees, and verification on Ethereum is definitely not a cheap option. In addition, ETH Gas fees soar from time to time and become a “noble chain”, which will also greatly affect the verification cost.

As a result, new ZK verification/settlement layer concept projects have emerged. These new projects are still relatively early, with Nebra being the representative. Old projects have also pivoted in this direction, such as Mina, and Zen, which just passed the new proposal.

The overall idea of most projects in this track is basically:

• Support multiple ZK languages;
• Support ZK aggregation proof, which is more efficient and cheaper;
• Faster Finality time;

At present, it seems very likely that the ZK settlement layer and the decentralized Proof Market will be tied together. After all, you need computing power in addition to technology. We may see some settlement layer projects cooperating with Proof Market projects, or the settlement layer with computing power may directly build a Proof Market, or the Proof Market with technology may build a settlement layer. The market will have the final say on how to proceed.

There should be many articles online about other areas of Infra, such as OEV in the Oracle and MEV fields, and ZK light client in the interoperability field, so I will not go into details here.