Kai-Fu Lee suggests Sam Altman of OpenAI is likely experiencing sleepless nights due to 01.AI’s shift to Deepseek
Kai-Fu Lee Shifts Focus of 01.AI to Deepseek’s Open-Source Models
Introduction to the Shift
In a notable change of direction, Kai-Fu Lee, the former head of Google China, has decided to pivot his AI startup 01.AI towards utilizing Deepseek’s open-source models. This strategic change, which Lee describes as crucial for the company’s future, responds to what he calls a "ChatGPT moment" in China, igniting excitement for AI technologies.
The Rise of Open-Source AI
Lee revealed in an interview with the South China Morning Post that the arrival of Deepseek’s open-source models has made a significant impact on the Chinese tech landscape. The enthusiasm generated by these models has encouraged numerous hardware and software companies in China to align their services with Deepseek’s offerings.
Following a noticeable increase in demand from Chinese CEOs for Deepseek models, Lee and his team chose to embrace this approach fully. "It became imperative for us to … embrace [Deepseek] as our primary bet," he expressed. This marked a shift from their prior focus on developing proprietary language models.
Challenge to Established Giants
Lee believes that Deepseek’s free and open-source stance poses a serious challenge to larger companies like OpenAI. He noted, "The biggest nightmare for Sam Altman is that his competitor is free." In fact, some users have already canceled their ChatGPT subscriptions in favor of Deepseek’s free offerings. This shift highlights the growing competition in the AI space, particularly between established players and innovative startups like Deepseek.
Focus on Corporate Customization
With a workforce of around 200 employees, 01.AI is now concentrating on tailoring Deepseek’s models for specific sectors, including finance, gaming, and legal industries. Lee pointed out that, unlike major corporations like Alibaba or Google that can justify large investments in pre-trained models due to their vast user bases, smaller companies struggle to do so.
The Importance of Technical Expertise
Despite changing tactics, Lee stresses that the technical skills of his team remain vital. He emphasized the complexities of training, tuning, and reinforcing learning models, which are areas where 01.AI can continue to provide value. Although the company’s revenue projection for the first quarter of 2025 is 100 million yuan (approximately $13 million), 01.AI has not yet reached profitability.
The Perception of Deepseek by Industry Giants
Recently, OpenAI and Anthropic have urged the U.S. government to impede the progress of Deepseek, claiming it operates under state control. Lee interprets this as a sign of "paranoia" among larger firms, as they see competition rising in the form of a capable and free alternative. He pointed out how companies built on what he calls a "house of cards" model are now confronting the reality of a strong competitor providing valuable offerings at no cost.
Trends in the AI Landscape
In a separate discussion with Bloomberg, Lee spoke about the ongoing consolidation in the development of AI models in both the U.S. and China. He expressed a belief that while only a few major firms would dominate the pre-training of large language models, open-source approaches like Deepseek will eventually take the lead.
Cost Structures in AI Development
A striking contrast in operating costs between companies was highlighted, with reports suggesting that OpenAI’s operational expenditures reached $7 billion in 2024, while Deepseek manages with a fraction of that budget—around two percent. This substantial difference raises questions about the sustainability of business models like OpenAI’s, especially in a landscape where cost efficiency is key.
Lee feels that Deepseek’s founder has sufficient backing to maintain operations and has managed to significantly reduce computing costs. With this kind of competitive edge, Lee suggests that Sam Altman, OpenAI’s CEO, might have considerable reasons to be concerned as the landscape continues to evolve.
In summary, Kai-Fu Lee’s pivot toward using Deepseek’s models represents a broader trend in the AI industry towards open-source solutions and emphasizes the challenges faced by traditional players. As interest in AI grows, the competition is likely to intensify, fundamentally altering the dynamics within the field.
