Source: Uweb
What is the new round of Web3+AI hot spots? The following is the highlights of the interaction between Uweb co-founder Wu Yaoshan and ArkStream Capital partner Zhong Zicong. Enjoy:
1. AI investment can focus on the model and intelligent entity layer, while the computing power layer is becoming saturated
When investing in the AI field, it is necessary to build an overall framework from upstream to downstream and gradually analyze each level. The initial AI industry is not yet mature, but the current stage of AI development is obviously different. The AI track can be divided into five key levels: computing power layer, model layer, AI Agent, data set layer and tool set layer. These levels can be evaluated from the dimensions of sustainability, technological innovation, cutting-edge and user growth.
Since the beginning of this year, the attention paid to the bottom layer and model layer has been particularly prominent. The bottom layer mainly refers to the GPU computing power layer, which is regarded as the foundation of AI development, similar to the importance of oil in traditional industries. Recently, Binance and OKX have launched some new currencies, which are mainly concentrated in the computing power layer. In addition, there are many related projects in the private equity track, indicating that the attention paid to the computing power layer continues to heat up. If this bottom layer architecture is properly constructed, higher-level layers such as models, AI agents, applications, data sets, and tools can be built on it. After the computing power layer, the importance of the model layer is becoming increasingly apparent. The model layer mainly involves large language models, which lay the foundation for higher-level AI applications and agents. In the future, the focus may gradually shift to higher-level models and agents. AI Agent represents the direction of AI applications. Although it is between the model and application layers and has not yet been fully applied, its execution capabilities and potential cannot be ignored.
The model layer and the agent layer are areas that deserve special attention in the future, while the computing power layer may have reached saturation.
2. Narratives in Web 3 should focus more on the model layer, especially narratives that can combine upstream and downstream resources and bring wealth effects
The narrative of Web 3 combined with AI is currently mainly concentrated on the model layer. Especially after the heat of the computing power layer has subsided, the continuous emergence of large language models indicates that the model layer will become the next wave of hot spots. The analysis of the model layer should focus on sustainability, technological innovation, cutting-edge and future user growth.
In traditional fields, AI is already very popular, but in the field of Web 3, whether it can form a strong narrative force will determine its development potential. Successful model layer projects can often provide new solutions, integrating the computing layer and the intelligent layer to form a new ecosystem. In addition, the model layer can also promote sustainable growth by creating a wealth effect, which is also the key to some recent successful projects.
3. Web3 AI models are mostly based on Web2 open source projects. Tao promotes decentralization, but education is needed for popularization; new market narratives have not yet emerged
There are no pure large-model projects in Web3. Many of them are developed in the form of AI public chains. Most of Web3's large-model projects are actually secondary development based on Web2's traditional open source large models. The second approach is to access traditional models to provide API interfaces.
In the field of AI models, Tao is a typical AI public chain project narrative, which aims to solve the problem of over-centralization and opacity of traditional AI. By using a decentralized AI model, Tao attempts to change the limitations of traditional AI, but Tao has not yet achieved large-scale application and still needs market education.
Some smaller AI model projects exist in the private equity market, but in the public equity market, Tao is still one of the few leading projects.
Since Tao's ecosystem is already relatively mature, future room for wealth growth may be limited.
The current trend is that some projects are not completely focused on AI model development, so it depends on how they create wealth effects and improve traditional models. Tao currently stands out in promoting decentralized AI models, but there has been no particularly new narrative in the market recently.
4. AI project private placement is hot, computing power layer projects are more active in the secondary market, and model layer projects are still circulating in the primary market
The current market environment is different from the previous cycle. It was relatively easy to make money in the previous cycle, but in the current cycle, the rate of return has dropped significantly. Many capital institutions, whether in the primary market or the secondary market, have a target rate of return between 2 and 5 times, while in the previous cycle, there were relatively more opportunities to find 100-fold coins, such as the Polkadot project.
One of the new trends in the current market is that many founders who have successfully operated conventional projects have taken a fancy to the narrative in the field of AI and have launched new AI projects. Sentient is a typical example, and its current valuation in the private equity market has reached about 1 billion US dollars. Another similar project is Ritual, which was also initiated by a founder with a large capital background. New AI projects, especially those led by successful teams in the past, usually perform well in the private equity market. These teams know how to operate the market and use the wealth effect to promote project development.
At present, most AI-related projects are still in the private market stage. In the secondary market, the main circulation is related to computing power, such as IO.net. Overall, computing power layer projects are more active in the secondary market, while most projects involving AI models are still circulating in the primary market.
5.IO.net and Aethir focus on computing power services; the innovation space for computing power projects is becoming saturated, and the market may turn to a new direction in the future
Currently, two projects, IO.net and Aethir, are representative in the fields of AI and DePIN. The development directions of AI and DePIN are mainly divided into two categories:
The first category is that DePIN directly provides computing power services for training and reasoning of large language models. Such projects, such as IO.net and Aethir, have become popular at the beginning of this year. They provide high-quality computing power services by aggregating global computing power resources into a distributed network. Such projects have received widespread attention because their application scenarios are direct and easy to understand, but the challenge is how to effectively aggregate computing power and ensure its stability. IO.net and Aethir were originally competitors, but later found that cooperation was more advantageous. IO.net focuses mainly on C-end business, while Aethir focuses on B-end business, although the technical foundations of the two are similar. IO.net's computing power resources are mainly composed of RTX 4090 and some A100, and integrate the computing power resources of Filecoin and Render, but it has not yet been involved in the provision of large language models. Aethir focuses on providing high-end computing power services to enterprise-level customers.
The second type of project involves the development of AI models for wearable smart devices. This type of project builds high-value AI data training sets and feeds back to the development of large models. This approach is relatively complex and requires building a huge data network and then training large models.
Currently, the market is mainly focused on computing power projects, but the innovation space in this field may have become saturated. As the market evolves, more attention may be paid to new and more promising directions in the future.
AI Agent combines token economy and innovates tokenization; innovative gameplay focuses on mobile phone mining.
AI Agent is the third layer of AI development and is currently considered to be a direction with great potential in the future. The computing power of the first layer is already a relatively mature field, the second layer is the model, and AI Agent represents a new trend that combines an open, permissionless creator economy.
Traditional AI projects usually adopt a fee-based model, where customized agent services are developed by agent companies and charged according to the SaaS model. In the Web3 environment, the introduction of the token economy enables AI Agent to be tokenized in a decentralized manner, and users can choose and pay to use different agents according to their needs. This approach utilizes the token economy and creates a virtuous cycle, allowing projects to attract more users to participate in the early stages.
AI Agent may also involve more complex models, such as mining through mobile phones or dedicated devices, but the support of different devices may affect user participation, requiring project parties to conduct more market education and promotion in the early stages.
At present, AI Agent's innovative gameplay is mainly focused on mining through mobile phones. This method is relatively simple, but the success of the project also depends on whether it can effectively attract users, provide attractive rewards, and avoid market declines in the later stage. In the future, more AI Agent projects will enter the market, using token economy and decentralization to bring early dividends to users.
7. AI Agents are emerging in Web3, with general and dedicated device strategies running in parallel. The key to development is how to get users involved smoothly in the early stages
AI has been popular in traditional industries for a while, and in the Web3 field, it has gradually become popular since the beginning of this year. At the beginning of the year, the underlying infrastructure such as computing power was gradually improved, and then AI Agent began to emerge.
The application of general-purpose equipment in the field of AI Agent is far more extensive than that of special-purpose equipment. At present, most AI Agent projects are still in the early stages, and the use of general-purpose equipment allows more people to participate easily. Some projects with special-purpose equipment are also emerging. The challenge of such projects is how to open up sales channels, especially in markets such as Japan and South Korea that have a greater influence on KOLs. If special-purpose equipment is successfully sold, users may show higher stickiness and loyalty after the initial investment cost, making the entire project larger. In contrast, the threshold of general-purpose equipment is lower, and users do not need to invest in advance, so the project scale may be relatively small. But in the long run, special-purpose equipment may dominate larger-scale projects. Some successful projects in the past have adopted similar strategies, and now this strategy has been applied to the field of AI Agent and combined with more combined gameplay. Whether it is general-purpose equipment or special-purpose equipment, the key is how to get users involved smoothly in the early stage.
8. Web3 data processing continues the traditional logic and adds new elements such as privacy protection and decentralization
In terms of data processing, Web3 projects are basically consistent with traditional practices, especially in data cleaning and processing logic. Although Web3 introduces privacy protection, users do not care much about privacy protection, and the core cleaning and processing process is still the same as traditional. AI Agents usually have the functions of the model layer, which are themselves built based on Web2 technology and traditional software. Therefore, data processing in Web3 is largely just a microcosm of traditional data processing, but with the addition of decentralization, token economy and composability elements, but its underlying logic is still the same as traditional data processing.
9. The key feature of Web3 AI Agent is composability. The core innovation is mainly reflected in the token economy, wealth effect and the composability architecture of the ecosystem. Other aspects have not changed much compared with Web2.
Composability is a key feature of Web3 AI Agent, which is different from the traditional agent development method. Traditional agents are usually developed by R&D companies based on a user's pain point to solve a specific problem. Web3 AI Agent is defined and created by the market and the community, forming a more open and ecological system.
In the Web3 ecosystem, developers can create and combine various different Agents in an Agent Marketplace, for example, an Agent for mining or an Agent for exchanging tokens. This marketplace allows users to flexibly select and combine different Agents according to their needs, thereby achieving greater composability and customization.
Although Web3 has introduced the token economy and wealth effect, in terms of data cleaning, privacy protection and model application, Web3 still uses the technology and methods of Web2, and no significant innovation has been seen yet. The development path of AI in Web3 is similar to that of other early tracks. Development is mainly driven by the wealth effect, and there may be real technological innovation.
10. OLAS and Spectral are emerging leading AI Agent projects with market popularity and potential, and good development prospects.
Fetch AI and Ocean Protocol, both of which are old projects in the last cycle, have gradually transformed into AI Agents recently. The current market is more inclined to chase emerging narratives and projects. Among the emerging AI Agent projects, OLAS and MOL are worth mentioning. OLAS is a project deployed by Arkstream Captical in January this year. At that time, it achieved a return of nearly ten times. It has now become one of the emerging leaders. As long as they maintain their leading position, there is still broad room for future development.
Another project that Arkstream Captical has been tracking for a long time is Spectral. This project has undergone a transformation in the past three years, from its original business to the field of AI Agent. Their ability to innovate in the market is worth paying attention to because they already have a certain customer base.
When investing or paying attention to leading projects in the field of AI Agent, Arkstream Captical tends to choose emerging leaders. Although old projects may also receive market hype, emerging leaders usually have more market heat and potential.
* This article is for learning and sharing purposes only and does not constitute any investment advice.