China just 'months' behind U.S. AI models, Google DeepMind CEO says

Demis Hassabis, CEO of Google DeepMind, recently shared his insights on the rapid advancements of artificial intelligence in China, suggesting that the nation may be only months away from matching the capabilities of U.S. and Western AI models. Speaking on CNBC's newly launched podcast, The Tech Download, Hassabis noted that the gap between Chinese and Western AI technologies has narrowed significantly over the past couple of years. He explained that the progress made by Chinese AI labs, particularly with models like those developed by DeepSeek, has been impressive. These models have demonstrated strong performance using less advanced chips and at a lower cost compared to their American counterparts. Although the initial shock of DeepSeek's capabilities has diminished, companies like Alibaba and innovative startups such as Moonshot AI and Zhipu are also contributing to the competitive landscape. However, Hassabis cautioned that while China may be closing in, the true test lies in the ability of Chinese firms to innovate beyond existing frontiers. He emphasized, "The question is, can they innovate something new beyond the frontier?" The CEO pointed out that despite catching up, there has yet to be a significant breakthrough akin to the transformer model developed by Google researchers in 2017, which has become foundational for many large language models, including OpenAI's ChatGPT and Google's Gemini. Industry leaders, including Nvidia CEO Jensen Huang, have also acknowledged China's advancements in AI. Huang remarked last year that the U.S. is not significantly ahead in the AI race, highlighting China's strength in energy and infrastructure while noting the U.S. lead in semiconductor technology. Despite these strides, Chinese tech companies face hurdles, particularly with access to cutting-edge technology. The U.S. export ban on advanced semiconductors from Nvidia presents a significant challenge for training sophisticated AI models. Although the White House has signaled a potential approval for sales of Nvidia's H200 chip to China, it is still not at the forefront of AI technology. Analysts have expressed concerns that the lack of access to top-tier Nvidia chips could widen the gap between U.S. and Chinese AI models over time. Richard Clode, a portfolio manager at Janus Henderson, indicated that the superior U.S. AI infrastructure might lead to divergent advancements in the coming years. Furthermore, even within China, there is acknowledgment of the steep competition. Lin Junyang, technical lead of Alibaba's Qwen team, mentioned at an AI conference that there is a less than 20% chance for Chinese firms to surpass U.S. tech giants within the next three to five years, citing a significant disparity in computing infrastructure. Hassabis attributes the challenges in achieving frontier breakthroughs more to a mindset than to technological limitations. He compared DeepMind's approach to that of a modern-day Bell Labs, promoting exploratory innovation rather than merely expanding on established knowledge. He concluded by noting that while China possesses world-class engineering talent, the complexity of scientific innovation remains a formidable barrier. As a key figure in the AI arena, Hassabis's assessments underscore the dynamic and competitive nature of the global AI landscape, especially as Google continues to push forward with its AI initiatives, including the recently launched Gemini 3 model.