Summary of Key Points
During the AI startup boom in 2023, top talents flocked out of large companies. By 2026, however, the trend reversed, with young AI geniuses such as Luo Fuli, Sun Tianxiang, and Yao Shunyu beginning to return to these same companies. The reason behind this shift is that the development of large models has entered a “big science” phase—requiring massive computing power, long-term funding, organizational collaboration, and practical application scenarios, which small teams find difficult to sustain. These individuals are not giving up; rather, they are pursuing challenging but meaningful projects by choosing companies that can provide them with the necessary resources and opportunities. At the same time, some others (like Zeng Guoyang) have chosen to focus on more niche areas, such as developing smaller, specialized models for specific use cases. Their common goal is to remain committed to solving problems, rather than being tied to any particular organization.
Detailed Analysis
From “Moving Out” to “Returning In”: Why Have AI Talents Changed Their Approach?
After the success of ChatGPT in 2023, AI startups became akin to a new gold rush, with many believing that small teams could develop models with just a little funding, and leaving large companies was seen as a sign of bravery. However, reality quickly showed otherwise: large models are not like mobile internet products that can be rapidly iterated on by small teams; they require thousands of GPUs, stable power supply, and continuous funding, sometimes costing millions of dollars per month (for example, MiniMax consumed $27.9 million monthly). Many startups failed to sustain this burden. Wang Huiwen’s company was acquired by Meituan, Baichuan shifted its focus to healthcare, and ZeroOneWanShi stopped pursuing massive models. Even DeepSeek, which appeared to be a startup, relied on the powerful computing resources provided by Huafang Quantization. Small teams could not afford these expenses, leading to a return of talents to large companies, which offer the necessary computational power, application scenarios, and financial support for model development and deployment.
Returning to Large Companies Is Not About “Lying Back”: They Seek Practical Resources
The return of these talents is not about high salaries or prestige; they are after the resources and authority needed to actually make things happen. For example:
- Luo Fuli (Xiaomi): Seven years ago, she was looking for a place that balanced research with practical applications—something more than just theoretical work. Xiaomi’s range of products, including smartphones, cars, and IoT devices, provided her with real-world scenarios to deploy her MiMo model (e.g., using it in car infotainment systems and home automation).
- Sun Tianxiang (Baidu): His MOSS model was one of the earliest open-source ChatGPT-like models in China, but academic laboratories could not turn it into a commercially viable product. Baidu’s strengths in search technology, cloud services, and WenXinYiYan helped transform his model from a prototype into a real product that could withstand user testing.
- Yao Shunyu (Tencent): His ReAct and Tree of Thoughts models advanced from basic chat functionality to practical applications. Tencent’s platforms, such as WeChat and gaming, enabled him to integrate these technologies into useful products (e.g., intelligent assistants within WeChat).
What they seek is real power: the ability to set research directions, allocate computing resources, select teams, and integrate models into products. If they are merely given high salaries without meaningful responsibilities, they will quickly become overwhelmed; if not, it means the companies have truly granted them that authority.
Some Choose a Different Path: Smaller Models Can Be Just as Effective
Not all talents return to large companies. Zeng Guoyang, for instance, focuses on developing smaller models with 2 billion parameters (referred to as “small cannons”). These models serve different purposes from the massive models developed by large companies. While large companies focus on building extensive infrastructures (costly but efficient), smaller models are more suitable for specific use cases on mobile devices and robots, requiring less resources and being easier to maintain. This is not a compromise; it’s about choosing a path that suits their expertise and goals.
The Core Remains the Same: They Are Committed to Solving Problems, Not Companies
Whether they join large companies or stay outside, these talents’ focus remains on solving problems. Luo Fuli’s original goal of balancing research with business hasn’t changed; she has moved from Alibaba to DeepSeek to Xiaomi, always looking for environments that meet her criteria. Sun Tianxiang has continued to work on practical applications, and Zeng Guoyang focuses on developing models for specific use cases. They are not tied to any one company; they will move if better opportunities arise. This is similar to the path of the sun across the sky each day—although its position changes, the overall pattern remains consistent. Their loyalty lies in solving problems, not in staying with a particular company.
This influx of AI talents indicates that the industry has matured. Large models require collaborative efforts akin to those in “big science” projects, and talents are choosing platforms that can support their work. Niche areas also offer opportunities; as long as they can solve real-world problems, whether within large companies or small teams, their contributions are valuable. Their stories show us that true geniuses always follow the path of solving problems, not just trends.