ABCDE: Viewing AI+Crypto from the Perspective of the Primary Market

Author: Laobai, ABCDE

More than a year after the release of ChatGPT, discussions about AI + Crypto have become lively again in the market. AI is seen as one of the most important tracks for the bull market in 2024-2025. Even Vitalik Buterin himself published an article titled "The promise and challenges of crypto + AI applications" to explore possible directions for AI + Crypto in the future.

This article will not make too many subjective predictions but will simply provide an overview of the entrepreneurial projects that combine AI and Crypto observed over the past year from the perspective of the primary market. We will look at the specific angles entrepreneurs are taking to enter the market, what achievements they have made, and what areas are still being explored.

1. The Cycle of AI + Crypto

Throughout 2023, we discussed nearly dozens of AI + Crypto projects, among which a clear cycle can be observed.

Before the release of ChatGPT at the end of 2022, there were very few blockchain projects related to AI in the secondary market. The main ones that came to mind were a few established projects like FET and AGIX, and there were similarly few AI-related projects in the primary market.

From January to May 2023, it can be said that this was the first concentrated outbreak period for AI projects. After all, the impact of ChatGPT was immense, leading many old projects in the secondary market to pivot towards the AI track, and in the primary market, we could almost discuss AI + Crypto projects every week. Similarly, the AI projects during this period felt relatively simple; many were based on "skinning" ChatGPT and "chain reform" projects, with almost no core technical barriers. Our in-house development team could often replicate a project's basic framework in just a day or two. This led to many discussions about AI projects during this time, but ultimately no investments were made.

From May to October, the secondary market began to turn bearish. Interestingly, the number of AI projects in the primary market also sharply decreased during this period, only becoming active again in the last month or two, with discussions and articles about AI + Crypto becoming rich once more. We have re-entered a "prosperous scene" where we can encounter AI projects weekly. After six months, it is evident that the new batch of AI projects has a better understanding of the AI track, the landing of business scenarios, and the combination of AI + Crypto has significantly improved compared to the first wave of AI hype. Although the technical barriers are still not strong, the overall maturity has stepped up a level. It was only in 2024 that we finally made our first bet in the AI + Crypto track.

2. The Track of AI + Crypto

In his article on prospects and challenges, Vitalik Buterin provided predictions from several relatively abstract dimensions and perspectives:

• AI as a participant in the game
• AI as the game interface
• AI as the game rules
• AI as the game objective

We summarize the AI projects currently seen in the primary market from a more concrete and direct perspective. Most AI + Crypto projects revolve around the core of Crypto, namely "decentralization (or political decentralization) + assetization in business."

There’s not much to say about decentralization; it's Web3… Based on the categories of assetization, it can be roughly divided into three main tracks:

• Assetization of computing power
• Assetization of models
• Assetization of data

Assetization of Computing Power

This is a relatively dense track, as there are not only various new projects but also many old projects pivoting, such as Akash from Cosmos and Nosana from Solana. After pivoting, their tokens have surged, reflecting the market's optimism about the AI track. Although RNDR primarily focuses on decentralized rendering, it can also serve AI, so many categorize RNDR and similar computing power-related projects into the AI track.

The assetization of computing power can be further subdivided into two directions based on the use of computing power:

• One is represented by Gensyn, focusing on "decentralized computing power for AI training."
• The other is represented by most pivoted and new projects, focusing on "decentralized computing power for AI inference."

In this track, an interesting phenomenon can be observed, or rather, a disdain chain:

• Traditional AI → Decentralized inference → Decentralized training

Traditional AI professionals are skeptical about decentralized AI training or inference.

Decentralized inference professionals are skeptical about decentralized training.

The main reason lies in the technology, as AI training (specifically for large models) involves massive amounts of data, and even more exaggerated is the bandwidth demand created by the high-speed communication of this data. In the current environment of Transformer large models, training these large models requires a computing power matrix equipped with numerous high-end GPUs like the 4090 or H100 professional AI GPUs, along with NVLink and specialized optical fiber communication channels forming a hundred Gbps communication pathway. You might wonder if this can be decentralized, hmm…

The demand for computing power and communication bandwidth for AI inference is far less than that for AI training, making the possibility of decentralized implementation much greater. This is why most computing power-related projects focus on inference, with training being primarily handled by major players like Gensyn and Together, which have raised over a hundred million. However, from the perspectives of cost-effectiveness and reliability, at least at this stage, centralized computing power for inference is still far superior to decentralized solutions.

This explains why decentralized inference professionals think decentralized training is unfeasible, while traditional AI experts view decentralized training as "technically unrealistic" and "commercially unreliable."

Some might say that when BTC/ETH first emerged, everyone also claimed that the model of distributed nodes calculating everything was relatively unreliable compared to cloud computing, yet it succeeded in the end. It will depend on the future demands for correctness, immutability, and redundancy in AI training and inference. Simply competing on performance, reliability, and price, it is currently indeed impossible to surpass centralized solutions.

Assetization of Models

This is also a crowded track for projects and is more easily understood compared to the assetization of computing power. After the rise of ChatGPT, one of the most well-known applications is Character.AI. You can discuss knowledge with historical figures like Socrates and Confucius, chat casually with celebrities like Elon Musk and Sam Altman, or even have romantic conversations with virtual idols like Hatsune Miku and Raiden Shogun. All of this showcases the charm of large language models. The concept of AI Agents has been deeply ingrained in people's minds through Character.AI.

What if these Agents, like Confucius, Musk, and Raiden Shogun, were NFTs?

Isn't this AI X Crypto?!

Thus, rather than saying it is the assetization of models, it is more about the assetization of Agents built on large models. After all, large models themselves cannot be put on-chain; it is more about mapping Agents based on the models into NFTs to create a sense of "model assetization" in AI X Crypto.

Currently, there are Agents in the community that can teach you English, as well as Agents that can engage in romantic relationships with you. Various types of Agents, including search and marketplace projects, can also be seen.

The common issues in this track are, first, the lack of technical barriers. It is essentially just the NFTization of Character.AI. Our in-house tech experts can create an Agent that talks and sounds like BMAN in just one night using existing open-source tools and frameworks. Second, the integration with blockchain is very light, somewhat akin to Gamefi NFTs on ETH, where the metadata may essentially just store a URL or hash, while the model/Agent resides on cloud servers, and the on-chain transaction is merely about ownership.

The assetization of models/Agents will still be one of the main tracks of AI x Crypto in the foreseeable future. We hope to see projects with relatively strong technical barriers and a closer, more native integration with blockchain emerge in the future.

Assetization of Data

Logically speaking, data assetization is the most suitable for AI + Crypto because traditional AI training mostly relies on visible data available on the internet, or more precisely, data from public domains. This data may account for only 10-20% or less, while more data actually resides in private domains (including personal data). If this traffic data can be used to train or fine-tune large models, we can certainly have more specialized Agents/Bots in various vertical fields.

What is the slogan that Web3 excels at? Read, Write, Own!

Therefore, through AI + Crypto, under the guidance of decentralized incentives, releasing personal and private traffic data and assetizing it to provide better and richer "food" for large models sounds like a very logical approach, and indeed, there are a few teams deeply engaged in this field.

However, the biggest challenge in this track is that data is much harder to standardize than computing power. For decentralized computing power, the model of your GPU can directly translate into how much computing power it provides, while the quantity, quality, and usage of private data are difficult to measure across various dimensions. If decentralized computing power is like ERC20, then the assetization of decentralized AI training data is somewhat like ERC721, and it resembles a mix of many projects like Monkey PunkAzuki with various traits, making liquidity and market creation much more challenging than ERC20. Thus, projects focused on AI data assetization are currently struggling.

Another noteworthy aspect of the data track is decentralized labeling. Data assetization acts on the "data collection" step, and the collected data needs to be processed before being fed to AI, which is the data labeling step. This step is currently mostly a centralized, labor-intensive process. By using decentralized token rewards to transform this labor work into a decentralized model, such as labeling to earn or similar to a crowdsourcing platform, is also a potential approach. We have seen a small number of teams currently working in this area.

3. Missing Pieces in AI + Crypto

To briefly discuss from our perspective, the missing pieces in this track are as follows.

First is the technical barrier. As mentioned earlier, the vast majority of AI + Crypto projects have almost no barriers compared to traditional AI projects in Web2. They rely more on economic models and token incentives for front-end experiences, spending effort on market and operations. This is certainly understandable, as decentralization and value distribution are strengths of Web3, but the lack of core barriers inevitably gives a sense of X to Earn. We still look forward to seeing more teams like RNDR, whose parent company OTOY has core technology, make strides in Crypto.

Second is the current state of practitioners. Based on current observations, some teams of entrepreneurs in the AI X Crypto track understand AI well but have a shallow understanding of Web3. Conversely, some teams are very Crypto Native but lack depth in the AI field. This is very similar to the early Gamefi track, where some understood games and thought about Web2 game chain reform, while others understood Web3 and focused on various gold mining model innovations and optimizations. Matr1x was the first team we encountered in the Gamefi track that understood both games and Crypto well, which is why I previously mentioned that Matr1x was one of the three projects I decided on after just discussing in 2023. We hope to see teams that understand both AI and Crypto in 2024.

Third is the commercial scenario. AI X Crypto is in an extremely early exploratory stage, and the various types of assetization mentioned above are just a few major directions, each of which has tracks that can be explored and subdivided in detail. The various projects currently seen in the market combining AI and Crypto feel somewhat "stiff" or "rough," failing to leverage the optimal competitiveness or combinability of AI or Crypto, which is closely related to the second point mentioned above. For instance, our in-house R&D team has thought of and designed a better combination method, but unfortunately, after reviewing so many AI track projects, we still haven't seen any team enter this niche area, so we can only continue to wait.

What? You ask why we, as a VC, can think of certain scenarios before the entrepreneurs in the market? Because our in-house AI team has seven experts, five of whom hold PhDs in AI. As for the ABCDE team's understanding of Crypto, you know…

Finally, I want to say that although from the perspective of the primary market, AI x Crypto still seems very early and immature, this does not prevent us from being optimistic that in 2024-2025, AI X Crypto will become one of the main tracks of this bull market. After all, AI liberates productivity, and blockchain liberates production relationships. Is there a better way to combine these two? :)