DeepMind says Chinese AI firms trail West by six months

DeepMind says China still behind

At the World Economic Forum in Davos on January 20, 2026, Demis Hassabis said Chinese AI companies were roughly six months behind the leading Western frontier, according to his public interview with Bloomberg.

That may not sound like much, but in fast-moving AI research, half a year can mean several major model releases. Hassabis argued that while Chinese firms are catching up quickly, they have not yet moved ahead of the cutting edge.

Why six months matters in AI

In artificial intelligence, progress often comes in rapid jumps. A few months can bring major gains in reasoning ability, efficiency, or multimodal features that let systems handle text, images, and more at once.

Hassabis’ timeline suggests Chinese firms are close but not leading. Being just behind means they can study what frontier labs release, but still have to prove they can set the pace rather than follow it.

DeepSeek’s splash drew huge reactions

Hassabis referenced the reaction to DeepSeek’s R1 model, which launched in early 2025 and drew global attention for its low reported training cost relative to many U.S. models according to industry reporting and technical papers.

Even so, he described the response as a massive overreaction. In his view, the model showed strong engineering and fast progress, but did not clearly push beyond the most advanced work already happening in the West.

Impressive work but not frontier leading

Hassabis acknowledged that DeepSeek’s model was impressive. He said Chinese labs are getting better at reaching the same level as frontier systems, showing real technical skill under challenging conditions.

However, he added that they have not yet shown they can innovate beyond that frontier. In other words, they are closing the gap, but still have to demonstrate breakthroughs that set new global standards.

Chip limits have shaped China’s path

Chinese AI firms have been operating under strict limits on access to the most advanced US chips. Those semiconductors are key to training and running the largest, most powerful AI models used by leading labs.

As a result, researchers in China have been pushed to explore unconventional methods and alternative system designs. These constraints have influenced how their models are built and optimized.

US policy shift could change things

In December 2025 the U.S. announced it would permit limited exports of certain Nvidia H200 class processors to China under licensing conditions, a move codified by the Department of Commerce in January 2026, while tighter controls remain on the most advanced chip families.

The change could give Chinese companies more room to use high-end processors, though the most advanced AI chips will still be blocked on national security grounds, keeping some limits in place.

Nvidia sits at the center

NVIDIA is a key player in this debate. The company has argued that if its top chips remain banned, China will speed up work on domestic alternatives instead of relying on US suppliers.

The H200 received conditional export approval from the United States but is subject to licensing rules and national security safeguards, and shipments have faced additional delays and conditional approvals in China, underscoring how hardware access remains politically fraught.

Not everyone supports looser exports

At Davos Anthropic CEO Dario Amodei warned that exporting advanced AI chips to China would be reckless and used a stark comparison reported in interviews, saying it would be like selling nuclear weapons to North Korea.

Those comments highlight how divided leaders are over how open the AI hardware trade should be. The debate blends technology, military concerns, and economic competition in one high-stakes conversation.

Chinese startups are gaining momentum

Despite limits, Chinese AI startups are building speed. Chinese AI firms, including MiniMax, have completed Hong Kong listings, and other firms, such as Zhipu, have filed or pursued listings in recent months, reflecting strong investor interest in the sector, though exact listing status varies by company.

These moves suggest that even if they trail the frontier, Chinese firms are still scaling up quickly. Public listings can provide more capital to fuel research, products, and global expansion.

Google fights back in the AI race

Google has been working to close its own gaps in AI. After missing some early excitement, the company released a new model that drew strong reviews and even prompted OpenAI CEO Sam Altman to declare a code red.

DeepMind plays a central role in these efforts, helping power Google’s Gemini assistant. The system is designed to connect with data from products like Gmail, Search, YouTube, and Photos for more personalized help.

Davos stage set the tone

Hassabis shared his views during a public interview at the World Economic Forum in Davos. Speaking with Bloomberg’s Emily Chang, he framed the AI race as highly competitive but still led by a small group of Western frontier labs.

The setting mattered. Davos is where tech leaders, policymakers, and investors compare notes on global power shifts. His comments were not just technical analysis but part of a broader conversation about who is shaping the future of AI.

Low cost models shook assumptions

Part of the buzz around China’s DeepSeek came from reports that its R1 model was built at a fraction of the cost of comparable systems in Silicon Valley. That raised questions about whether top-tier AI really needs sky-high budgets.

Hassabis still sees Western labs as ahead, but cheaper development paths could change how the race unfolds. If strong models can be trained with fewer resources, more players could compete at higher levels than before.

Which AI moves defined our last year? Explore the strategic AI decisions that shaped the year 2025.

Robotics could be the next leap

DeepMind is also looking beyond chatbots and assistants. Hassabis said robotics and physical intelligence are a major focus, as AI systems begin interacting more directly with the real world instead of just digital text.

He predicted a breakthrough moment could come soon, though he noted how hard it is to match the reliability and dexterity of the human hand.

It’s worth reading how a viral deepfake of Sam Altman became one of the most-viewed clips on OpenAI’s new Sora video app.

What do you think about the global AI race right now? Share your thoughts.

This slideshow was made with AI assistance and human editing.

Don’t forget to follow us for more exclusive content on MSN.

Read More From This Brand:

• Doctors warn AI use may be linked to psychosis
• Lenovo AI glasses could change presentations with teleprompter support
• Meet the two fastest-growing AI chatbots now, not ChatGPT