(Bloomberg) — Chinese entrepreneur Kai-Fu Lee expects just three AI models to survive a domestic shakeup triggered by the advent of DeepSeek, which up-ended the existing template for artificial intelligence development.
Most Read from Bloomberg
The venture investor and founder of 01.AI is betting on DeepSeek, Alibaba Group Holding Ltd. and ByteDance Ltd. dominating the Chinese market ultimately. And the US might have four leaders including xAI, OpenAI, Google and Anthropic, the prominent computer scientist told Bloomberg Television.
DeepSeek — a capable model supposedly trained at a fraction of the cost of OpenAI’s — has accelerated the pace of development globally since its January emergence. The Chinese startup demonstrated the effectiveness of the open-source framework and encouraged more developers to adopt that approach, Lee said.
That’s in part because of US sanctions intended to keep the most capable Nvidia Corp. chips out of China — forcing local model developers from DeepSeek to Moonshot and Zhipu to Minimax and Baidu Inc. to innovate.
“I would argue that the sanctions so far hasn’t worked,” said Lee, who formerly headed Google’s China business. “I would say DeepSeek actually benefited from having less resources and able to make training and inference at five to 10 times lower the expense incurred by OpenAI and others.”
Lee’s own company began by building large AI platforms but recently changed direction, saying the pursuit of massive, trillion-parameter models was too capital intensive. Instead, he’s steered it toward partnerships, including with Alibaba and DeepSeek, to build industry-specific applications. It wants to build what Lee called the operating layer atop DeepSeek’s underlying technology.
“OpenAI’s operating costs of $7 billion in 2024, DeepSeek’s probably operated with 2% the operating expense,” said Lee, a Carnegie Mellon-trained computer scientist who’s been a steadfast AI advocate. “Is OpenAI’s model even sustainable?”
Lee compared that with DeepSeek, which is offered to developers for free or at a fraction of the cost. “So with that kind of formidable competitor, I think Sam Altman is probably not sleeping well,” he said.
VCs in both the US and China are moving away from funding pioneer models, he added. Instead, they’re more enthusiastic about funding AI infrastructure, application companies and enterprise and consumer apps. In China, local governments are stepping in to boost AI development after Beijing designated the sector a priority.
