More than a year after the release of ChatGPT, discussions about AI+Crypto have become heated again recently. AI is considered one of the most important tracks for the bull market of 2024-2025, with even Vitalik Buterin himself publishing an article “The promise and challenges of crypto + AI applications” to discuss potential exploration directions for AI+Cryo in the future. This article will not make too many subjective predictions but will simply sort out the AI and Crypto combination entrepreneurial projects observed over the past year from a primary market perspective, to see from which angles entrepreneurs have entered the market, what achievements have been made, and which areas are still being explored.

I. The Cycle of AI+Crypto

Throughout 2023, we discussed nearly dozens of AI+Crypto projects, where a clear cycle could be observed. Before the release of ChatGPT at the end of 2022, there were few blockchain projects related to AI in the secondary market, with the main ones being older projects like FET and AGIX. Similarly, the primary market saw few AI-related projects.

From January to May 2023, there was the first concentrated explosion period of AI projects, mainly because ChatGPT had a huge impact. Many old projects in the secondary market pivoted to the AI track, and nearly every week, there were discussions on AI+Crypto projects in the primary market. However, this period’s AI projects felt relatively simple, many of which were just “reskinned” and “blockchain-converted” projects based on ChatGPT, lacking any core technological barriers. Our in-house development team could often replicate a project’s basic framework in just a day or two. This led us to discuss many AI projects during this time, but ultimately, we did not make any moves.

From May to October, the secondary market started to bear, and interestingly, the number of AI projects in the primary market also significantly decreased until the last one or two months when the number began to rise again, with discussions, articles, and more about AI+Crypto becoming richer. We re-entered a “boom” where we could encounter AI projects every week. After six months, we noticeably felt that the new batch of AI projects had a significantly improved understanding of the AI track, commercial scenario implementation, and integration of AI+Crypto compared to the first AI Hype period. Although the technological barriers are still not strong, the overall maturity has reached a new level. It wasn’t until 2024 that we finally made our first bet on the AI+Crpyto track.

II. In the AI+Crypto Track

Vitalik, in his article on “Forecasts and Challenges,” predicts from several relatively abstract dimensions and perspectives:

• AI as a player in the game
• AI as a gaming interface
• AI as the rules of the game
• AI as a gaming target

We, however, summarize the AI projects currently seen in the primary market from a more concrete and direct angle. Projects in AI+Crypto mostly revolve around the core of Crypto, which is “technical (or political) decentralization + commercial assetization.”

Decentralization needs no introduction; it’s all about Web3. Based on the type of assetization, it can be broadly divided into three main tracks:

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

Assetization of Computing Power

This is a relatively dense track, encompassing various new projects as well as pivots from older projects, such as Akash from Cosmos and Nosana from Solana. The token prices surged post-pivot, reflecting the market’s optimism towards the AI track. RNDR, although primarily focused on decentralized rendering, can also serve AI purposes, so many classify RNDR and similar computing power-related projects under the AI track.

The assetization of computing power can be further subdivided into two directions:

Decentralized computing for AI training, represented by Gensyn.

Decentralized computing for AI inference, represented by most pivots and new projects.

An interesting phenomenon observed in this track, or rather a skepticism chain, goes as follows:

Traditional AI → Decentralized Inference → Decentralized Training

Those with a traditional AI background are skeptical about decentralized AI training or inference. And within the decentralized space, those focusing on inference doubt the feasibility of decentralized training. The main reason lies in the technical challenges, as AI training (especially for large models) requires massive data and, more critically, high bandwidth for data communication. Presently, training large Transformer models necessitates a matrix of high-end GPUs (like the 4090 or H100 for AI) plus NVLink and professional fiber optic switches for 100G-level communication channels, casting doubt on the feasibility of decentralization for such tasks.

• Those who came from traditional AI majors are not optimistic about decentralized AI training or reasoning.
• Those who use decentralized reasoning are not optimistic about decentralized training.

The reason is mainly technical, because AI training (especially large model AI) involves massive amounts of data, and what is more exaggerated than the data requirements is the bandwidth requirements caused by high-speed communication of these data. In the current Transformer large model environment, training these large models requires a large number of 4090-level high-end graphics cards/H100 professional AI graphics cards purchased computing power matrix + 100G-level communication channels composed of NVLink and professional optical fiber switches. You Saying that this thing can be implemented in a decentralized manner, hmm…

AI reasoning requires far less computing power and communication bandwidth than AI training. The possibility of decentralization is naturally much greater than that of training. This is why most computing power-related projects are engaged in reasoning, and training is basically only Gensyn. , a big player like Together that has raised over 100 million yuan. But equally, from the perspective of cost performance and reliability, at least at this stage, centralized computing power is still far better than decentralized reasoning.

It is not difficult to explain why, when looking at decentralized reasoning and decentralized training, they think “you can’t do it at all”, while traditional AI looks at decentralized training and reasoning and thinks “training is technically unrealistic” and “reasoning is commercially unreliable”. Spectrum”.

Some people say that when BTC/ETH first came out, everyone also said that this model of all distributed nodes being counted is not reliable compared to cloud computing, but didn’t it work out in the end? Then it depends on the future needs of AI training and AI reasoning for the dimensions of correctness, non-tamperability, and redundancy. Simply focusing on performance, reliability, and price cannot be better than centralization for the time being.

Assetization of Models

This track is crowded and relatively easier to understand compared to computing power assetization. The popularity of ChatGPT and applications like Character.AI have demonstrated the potential of large language models. Users can seek knowledge from historical figures like Socrates or Confucius, chat with celebrities like Elon Musk or Sam Altman, or even engage in romantic conversations with virtual idols like Hatsune Miku or Raiden Shogun. This magic is all thanks to large language models, with the concept of AI Agents becoming deeply ingrained through Character.AI.

What if these agents, like Confucius, Musk, or Raiden Shogun, were NFTs?

Isn’t this AI X Crypto?!

This embodies the AI X Crypto concept. It’s more about the assetization of agents created from large models rather than the models themselves, as large models cannot be directly placed on the blockchain. The focus is on mapping agents onto NFTs to create a sense of “model assetization” in the AI X Crypto space.

Currently, there are agents for learning English, dating, and more, along with derivative projects like agent search and marketplaces. A common issue in this track is the lack of technical barriers, as many projects simply NFT-ize the Character.AI concept. Integration with blockchain is often minimal, similar to how GameFi NFTs on Ethereum might only store a URL or hash in their metadata, with models/agents hosted on cloud servers. Trading on the blockchain essentially involves ownership rights.

Despite these challenges, model/agent assetization will likely remain a major track in AI x Crypto, with hopes for more technically robust and blockchain-integrated projects in the future.

Assetization of Data

Data assetization is logically most suited for AI+Crypto, as traditional AI training primarily utilizes visible data available on the internet, or more precisely—data from the public domain traffic, which might only account for 10-20% of the total. A significant portion of data actually resides within private domain traffic (including personal data). If this traffic data could be utilized for training or fine-tuning large models, we could undoubtedly have more specialized agents/bots in various vertical domains.

The most familiar slogan of Web3 is “Read, Write, Own!”

Therefore, under the guidance of decentralized incentives through AI+Crypto, releasing individual and private desire traffic data for assetization to provide better and richer “feed” for large models sounds like a highly logical approach. Indeed, several teams are deeply engaged in this field.

However, the biggest challenge in this track is that data is not as easily standardized as computing power. For decentralized computing power, the model of your graphics card directly translates into a certain amount of computing power, whereas the quantity, quality, and purpose of private data are difficult to measure across various dimensions. If decentralized computing power is like ERC20, then the assetization of AI training data for decentralized AI feels more like ERC721, mixed with many projects and traits like PunkAzuki, making liquidity and market development significantly more challenging than with ERC20. Thus, projects working on AI data assetization are facing considerable difficulties.

Another noteworthy aspect of the data track is decentralized labeling. Data assetization applies to the “data collection” step, and the collected data needs to be processed before feeding it to AI, which is where data labeling comes in. This step is currently a centralized, labor-intensive task. By decentralizing this process through token rewards, turning this labor work into decentralized labeling to earn, or dispersing the work in a manner similar to crowdsourcing platforms, is a concept being explored. A few teams are currently cultivating this field.

III. The Missing Puzzle Pieces in AI + Crypto

Let’s briefly discuss, from our perspective, the missing pieces in the AI + Crypto sector.

1. Technological Barriers: As previously mentioned, the majority of AI + Crypto projects have almost no barriers compared to traditional AI projects in the Web2 space. They rely more on economic models and token incentives, focusing their efforts on front-end experience, market, and operations. While there is nothing wrong with this—decentralization and value distribution are indeed strengths of Web3—the lack of core barriers often gives off an “X to Earn” vibe. We still look forward to more teams with core technologies, like RNDR’s parent company OTOY, making significant strides in the Crypto space.

2. Current State of Practitioners: Based on our observations, some teams in the AI x Crypto sector are very knowledgeable in AI but lack a deep understanding of Web3. Conversely, some teams are very Crypto Native but have a shallow understanding of AI. This is very similar to the early Gamefi sector, where teams either had a deep understanding of gaming and sought to adapt Web2 gaming to blockchain or were well-versed in Web3, focusing on innovating and optimizing various earning models. Matr1x was the first team we encountered in the Gamefi sector with a double-A understanding of both gaming and Crypto, which is why I previously mentioned that Matr1x was one of the three projects in 2023 that I decided on “right after the discussion.” We look forward to seeing teams with a double-A understanding in both AI and Crypto in 2024.

3. Commercial Scenarios: AI X Crypto is at a very early stage of exploration. The various assetizations mentioned are just a few broad directions, each with potential sub-sectors that can be meticulously explored and segmented. The combination of AI and Crypto in current projects often feels “stiff” or “rough,” failing to leverage the best competitive advantages or combinability of either AI or Crypto. This is closely related to the second point mentioned above. For example, our in-house development team conceived and designed a more optimal combination method. Unfortunately, despite reviewing many projects in the AI sector, we have yet to find a team that has entered this niche. Therefore, we can only continue to wait.

Why could our VC think of certain scenarios before the market’s entrepreneurs? Because our in-house AI team includes seven masters, five of whom hold Ph.D.s in AI. As for the ABCDE team’s understanding of Crypto, well, you know…

In conclusion, although from a primary market perspective, AI x Crypto is still very early and immature, this does not prevent us from being optimistic about its prospects in 2024–2025. AI x Crypto could become one of the main sectors in the next bull market. After all, if AI liberates productive forces and blockchain liberates production relations, what better combination could there be? :)

Disclaimer：

1. This article is reprinted from [ABCDE], All copyrights belong to the original author [ABCDE]. If there are objections to this reprint, please contact the Gate Learn team, and they will handle it promptly.
2. Liability Disclaimer: The views and opinions expressed in this article are solely those of the author and do not constitute any investment advice.
3. Translations of the article into other languages are done by the Gate Learn team. Unless mentioned, copying, distributing, or plagiarizing the translated articles is prohibited.