The AI Layout Backed by Wang Huiwen Family Office: From Large Models to AI Agents
Covers multiple key links in the AI industry chain.
In recent years, the investment path of the AI industry has gradually become clear.
Take Wang Huiwen, co-founder of Meituan, as an example. In recent years, he has continued to lay out the AI industry through his family office (referred to as Wang Huiwen Family Office, with the fund name Lollapalooza Capital).
As can be seen from the official website of Lollapalooza Capital, Wang Huiwen Family Office has invested in 24 projects. From the large model company Moonshot AI, to the AI Agent startup Butterfly Effect, to AI video, AI education, AI hardware and other fields, its investment territory has covered multiple key links in the AI industry chain.
From these projects, we can see that around 2023, the core focus of the market was still the competition for large model capabilities. Who has stronger basic models, more computing power resources and larger data scale has become the focus of capital attention. However, with the gradual opening of large model capabilities, the competition for AI startups has begun to shift from "model capability" to "application implementation".
If these projects are split according to the industrial chain, several obvious directions can be seen: large models and infrastructure are the foundation, AI Agent is becoming a new application entry, content and education have become important implementation scenarios, and intelligent hardware represents the future human-computer interaction terminal.
"Lollapalooza" means an outstanding person or thing. Wang Huiwen Family Office named it this way, which reflects its persistent pursuit of top founders or excellent projects. In its investment portfolio, there are also projects that continue the heritage of "Light Year Beyond". Although Light Year Beyond did not become the Chinese version of OpenAI, the AI talents it gathered are continuing to exert influence through new startup projects.
(Screenshot from Lollapalooza Capital's official website)
01 Large models remain the technical core of AI competition
In the investment portfolio of Wang Huiwen Family Office, large model companies still occupy an important position.
Among them, Moonshot AI is one of the most concerned large model startups in China at present.
The Kimi intelligent assistant launched by Moonshot AI gained a lot of user attention in the early stage by virtue of its long text processing capability, and has also become a representative case in the competition of domestic large model products.
In July of this year, Moonshot AI released a new generation of base model Kimi K3, which has become the world's largest open source model with a total of 2.8 trillion parameters. With a number of underlying architecture innovations, it has entered the world's first echelon in scenarios such as programming, agents and long text reasoning. In a number of authoritative evaluations, Kimi K3 has surpassed many overseas closed-source leading models.
Different from traditional Internet startups, the competition of large model companies is not just product competition, but a comprehensive competition of model capability, engineering capability, data capability and commercialization capability.
For investors, investing in large model companies is to bet on the most leverage-effect underlying capability in the AI era.
In the past Internet era, search engines, social platforms and mobile operating systems defined the path for users to reach the world; while in the AI era, large models have become operating system-level platforms that drive all intelligent applications.
However, with the gradual convergence of large model capabilities, industrial competition is shifting from "whether there is a model" to "how the model is used".
This also explains why Wang Huiwen's investment territory does not stay at the model layer, but further extends to the AI infrastructure and application layers.
02 AI Infrastructure: the "water, electricity and coal" of the AI era
In addition to model companies, AI infrastructure is also an important investment direction.
SiliconFlow is an AI model service platform company, which mainly provides basic capabilities such as model invocation and deployment.
Now more and more companies want to make use of AI, and the demand for model reasoning and deployment efficiency is also rising.
Yuan Jinhui is the founder of SiliconFlow, a doctor from the Department of Computer Science of Tsinghua University. He previously worked as a principal researcher at Microsoft Research Asia, and developed LightLDA, the world's fastest topic model training algorithm at that time.
He has a lot of intersections with Wang Huiwen.
In 2023, Wang Huiwen entered the AI industry and founded Light Year Beyond, hoping to build it into the Chinese version of OpenAI. During that period, Light Year Beyond acquired 1st Technology founded by Yuan Jinhui through share swap.
A few months later, Meituan acquired the assets of Light Year Beyond. After leaving, Yuan Jinhui founded SiliconFlow at the end of 2023, focusing on AI infrastructure.
Looking back, the technical team that Wang Huiwen bet on at the beginning is now moving forward on the road of AI infrastructure.
If large models are similar to the "operating system" of the AI era, then model service platforms, computing power infrastructure and development tools are similar to the infrastructure that supports the operation of the entire ecosystem.
In addition, companies such as Boxlite also pay attention to infrastructure construction in the AI Agent era.
However, the difference is that BoxLite focuses on providing a secure, isolated and stateful operating environment for AI Agents.
As AI gradually develops from a chat tool to an agent capable of executing tasks, how to make Agents run safely, invoke tools and complete complex tasks will become a key issue in enterprise applications.
In the future AI industry, there will be both model companies and infrastructure enterprises that can support the large-scale application of AI.
03 AI Agent becomes a new hotspot
If the keyword of 2023 is "large model", then one of the keywords in the past two years is "AI Agent".
The biggest change of AI Agent is that it no longer just answers questions, but tries to complete tasks for users, such as sorting out materials, analyzing information, operating software, executing workflows, etc.
In this direction, Manus, a product under the Butterfly Effect, has become a highly concerned AI Agent product in the market.
The emergence of Manus makes the market re-discuss the possibility of general-purpose agents.
In addition to Manus, Wang Huiwen's investment portfolio also includes many startups that start businesses around Agents.
For example, Pine AI focuses on phone agents, helping users handle bills, customer service communication and other affairs through AI; Trooly explores user research services based on Voice Agent; Efflora focuses on agents that can perform software engineering tasks.
These projects cover multiple scenarios such as personal assistants, enterprise services, and software development.
From the perspective of industrial trends, AI Agent is becoming an important bridge connecting model capabilities and commercial applications.
In the future, the way enterprises use AI may no longer be simply invoking a chatbot, but deploying multiple digital employees who can complete specific work to collaborate with humans.
04 AI content entrepreneurship: generative AI enters the production process
In addition to basic models and Agents, AI content generation is also an important direction in Wang Huiwen's investment portfolio.
In recent years, AI-generated images, videos and texts have developed rapidly, and the content industry has become one of the earliest industries affected by AI.
Among them, Pixverse is a representative project in the field of AI video generation.
As video becomes one of the core forms of Internet content consumption, AI video generation is considered to have great commercial potential.
Advertising marketing, short videos, e-commerce content, film and television production, education and other fields may all become important application scenarios for AI videos.
Compared with traditional content production methods, AI lowers the threshold of creation, enabling individual creators and small teams to complete content production that previously required professional teams.
In addition, Sand AI is also an AI company focusing on video generation models and products, while CreativeFitting has launched AI short drama related product Reel.AI.
Cao Yue, founder of Sand AI, once served as co-founder of algorithms at Light Year Beyond and former head of the vision model center of a top AI research institution, focusing on basic vision models and multimodal large model research. After the end of the Light Year Beyond project, Cao Yue founded Sand.ai in early 2024, focusing on the autoregressive direction to explore video generation.
These projects reflect a trend:
AI is gradually entering the content production process from an auxiliary creation tool.
(Summary of Wang Huiwen Family Office's investment projects, produced by Duozhi)
05 AI Education: AI Tutor is changing the way of learning
Educational technology is also a direction in Wang Huiwen's investment territory.
In the past ten years, the core logic of online education is to realize the scale of high-quality content through the Internet.
The emergence of AI has enabled educational technology to evolve towards more personalized learning services, bringing more possibilities.
Yuanfudao is one of the important enterprises in China's online education field, and it is also exploring intelligent hardware products and AI education directions.
At the same time, Cosmo represents a type of overseas AI education startup, whose direction is to use AI to provide personalized tutoring in English and mathematics.
Cosmo (formerly known as ClassUp) was founded in Silicon Valley, USA in 2023 by two Chinese alumni of Princeton University, Annie Chen and Jessica Zou. Shuai Ke, CEO of the company, was previously co-founder of Yuanfudao.
Duozhi learned that in July of this year, Cosmo has achieved first-order profitability for several consecutive months, and its cash flow turned positive in the first month. In addition, Gakku, an AI English product developed by the company, has been officially promoted.
Compared with traditional online education, the core advantage of AI Tutor is that it can adjust teaching methods in real time according to students' conditions.
In the past, one teacher faced dozens of students; in the future, one AI Tutor can face one student and continuously adjust teaching strategies according to learning data.
Of course, AI education is still in the exploratory stage, and it is worth looking forward to whether it can tap greater potential in the future.
06 AI Hardware: define the next generation of interaction methods
In addition to software and applications, AI hardware is also an important direction.
In Wang Huiwen's investment portfolio, Viture is an XR glasses company, while Li Auto represents the direction of intelligent vehicles.
Intelligent vehicles have become an important application scenario for AI technology.
From intelligent driving, to AI cockpit, to in-vehicle intelligent assistants, automobiles are becoming new computing platforms.
At the same time, XR devices are also considered to be one of the potential future AI interaction entrances.
If AI assistants need to accompany users for a long time in the future, in addition to mobile phones, terminals such as glasses and automobiles may all become new carriers.
(Sorting out of Wang Huiwen Family Office's investment projects, produced by Duozhi)
07 An AI industry chain investment map is taking shape
On the whole, Wang Huiwen Family Office's AI investment presents an obvious industrial chain layout:
Upstream, there are large model companies such as Moonshot AI, and infrastructure companies such as SiliconFlow;
In the midstream, there are AI Agent companies such as Manus and Pine AI;
Downstream, there are application companies for specific scenarios such as Pixverse, CreativeFitting and Cosmo;
At the same time, hardware directions such as intelligent vehicles and XR devices are also becoming an important part of the AI ecosystem.
In addition to the invested projects, "robotics and embodied intelligence" is also mentioned on the official website. Although this field has not yet appeared in the investment portfolio, it may also be a track that Wang Huiwen Family Office focuses on.
Due to Wang Huiwen's deep participation in AI startups, Lollapalooza Capital's investment is often interpreted by the outside world as his judgment on the future industrial direction. However, in the fast-changing AI field, he cannot invest in all projects correctly. He admits that there is no consensus in the AI industry yet, and there are also projects he "can't understand", such as Lovart, Genspark and Emochi, which correspond to different directions of AI design tools, general AI agents and AI interactive entertainment respectively. They represent the rapidly emerging new product forms in the AI era. Wang Huiwen said: "(They are all) growing very fast, with both positive and negative voices very distinct."
From the perspective of Wang Huiwen Family Office's investment portfolio, it has not bet on a certain type of direction. This also reflects the real state of current AI startups: the industry is still in the early exploratory stage, and no path has been verified as the final answer.
From large models to Agents, and then to various AI-native applications, different directions are evolving simultaneously. The future winners will be those companies that can continuously convert AI capabilities into user value.
This article is from the WeChat official account "Duozhi Wang" (ID: duozhiwang), written by Wang Shang, authorized for release by 36Kr.