Kai-Fu Lee Positions 01.AI Towards Deepseek, Posing an Existential Threat to OpenAI

Former head of Google China, Kai-Fu Lee, is redirecting his AI startup 01.AI to fully embrace Deepseek’s open-source models, which he describes as an existential threat to OpenAI’s business model.

Lee’s company has shifted from its earlier strategy of developing proprietary large language models and is now exclusively relying on the open-source solutions offered by Deepseek, as he clarified in an interview with the South China Morning Post.

According to Lee, the release of Deepseek has triggered what he calls a «ChatGPT moment» in China, sparking widespread excitement around AI applications. This shift has encouraged many Chinese hardware and software providers to align their services with Deepseek’s models.

«For us, it became imperative to make Deepseek our main focus,» says Lee. This decision followed a surge in demand for Deepseek models from Chinese CEOs late in January.

Lee believes that the free, open-source approach of Deepseek fundamentally challenges OpenAI.

«The biggest nightmare for Sam Altman is that his competitor is free,» Lee asserts. «I’ve met numerous people who’ve canceled their ChatGPT subscriptions because Deepseek is available at no cost.»

Lee’s startup 01.AI, which employs 200 people, now aims to concentrate on customizing Deepseek models for clients in the finance, gaming, and legal sectors.

«A pre-trained model can only be justified with hundreds of millions of users,» Lee stated. «So Alibaba might justify it, Google can justify it, Deepseek can justify it, and ByteDance can just about justify it, but the rest of us cannot.»

Lee contends that his company’s technical expertise remains valuable despite this strategic pivot. «How you train it, how you tune it, how you conduct reinforcement learning, and how you achieve rapid inference—only companies with LLM capabilities can execute this last part,» he explains.

He anticipates a revenue of 100 million yuan ($13 million) for the first quarter of 2025, which would equal the company’s total revenue for 2024. Despite this growth, 01.AI has yet to achieve profitability.

Recently, OpenAI and Anthropic urged the U.S. government to ban Deepseek’s models, labeling the Chinese startup as «state-controlled.» Lee interprets this as a sign of paranoia. «They see their house of cards beginning to collapse because someone else has built an equally good house for free,» Lee remarks.

In a separate interview with Bloomberg, Lee emphasized the ongoing consolidation of pre-trained models in both the U.S. and China. He predicts that open-source approaches will eventually prevail, while pre-training of large language models will remain limited to a few major corporations.

Lee highlights the stark contrast in operating costs between the two companies. While OpenAI reportedly spent $7 billion on operational expenses in 2024, Lee claims that Deepseek requires only about two percent of that amount.

«Is the issue really about whose model is 1% better? I think they are all quite good. The real question is whether OpenAI’s model is sustainable at all,» Lee says.

Lee describes Deepseek as «infinitely sustainable,» as its founder has ample funding to maintain operations and has reduced computing costs by 5-10 times. «Given such a formidable competitor, I believe Sam Altman is probably not sleeping well,» Lee adds.

Source