LONDON (Realist English). China’s artificial intelligence models may now lag behind US and Western competitors by only “a matter of months”, according to Demis Hassabis, the chief executive of Google DeepMind, challenging the long-held view that China remains far behind in the AI race.
Speaking on CNBC’s newly launched podcast The Tech Download, Hassabis said Chinese AI capabilities are much closer to the global frontier than many assumed just a year or two ago. “Maybe they’re only a matter of months behind at this point,” he said, citing rapid advances by both Chinese tech giants and startups.
China’s progress came into sharp focus last year when DeepSeek unveiled a high-performing model built on less advanced chips and at significantly lower cost than US counterparts, briefly unsettling global markets. Since then, firms such as Alibaba, along with startups including Moonshot AI and Zhipu, have released increasingly competitive models.
Despite that progress, Hassabis argued that Chinese firms have yet to demonstrate genuine frontier innovation. “They’ve shown they can catch up and be very close to the frontier,” he said. “But can they actually innovate something new … that gets beyond the frontier? I don’t think that’s been shown yet.”
He pointed to the transformer architecture — developed by Google researchers in 2017 and now underpinning models such as OpenAI’s ChatGPT and Google’s Gemini — as an example of the kind of foundational breakthrough China has not yet replicated.
Other technology leaders have also acknowledged China’s advances. Jensen Huang, chief executive of Nvidia, said last year that the US was “not far ahead”, noting China’s strength in infrastructure and AI models even as Washington retains an edge in advanced chips.
Access to hardware remains a major constraint for Chinese firms. US export controls restrict sales of Nvidia’s most advanced semiconductors, which are crucial for training cutting-edge AI systems. While Washington has signalled it may allow exports of Nvidia’s H200 chip to China, it still falls short of the company’s top-tier products. Domestic alternatives from Huawei continue to lag in performance.
Some analysts believe these constraints will widen the gap over time. Richard Clode, a portfolio manager at Janus Henderson, told CNBC that US AI infrastructure is likely to pull ahead as it continues to scale, predicting that current conditions may represent “peak relative Chinese AI capability versus the US”.
Even Chinese executives acknowledge the challenge. Lin Junyang, a technical lead on Alibaba’s Qwen model, said recently that there was less than a 20% chance a Chinese firm would surpass US tech leaders in AI within three to five years, citing US computing infrastructure that is “one to two orders of magnitude larger”.
Hassabis, however, suggested the gap in breakthroughs reflects mindset rather than technology alone. He compared DeepMind to a “modern-day Bell Labs”, focused on exploratory scientific innovation rather than simply scaling existing approaches. “China definitely has world-class engineering,” he said, adding that true invention is “about 100 times harder than copying”.
DeepMind, founded by Hassabis more than a decade ago and acquired by Google in 2014, has been central to Alphabet’s recent AI push. In November, Google unveiled Gemini 3, its latest model, as it sought to reassure investors that it remains competitive with rivals such as OpenAI.
