^{Getty Images}

China’s artificial intelligence capabilities are far closer to those of the United States and other Western nations than many observers previously believed, according to Demis Hassabis, CEO of Google DeepMind. Speaking to CNBC, Hassabis said that leading Chinese AI models may now trail their U.S. counterparts by only “a matter of months,” a significant shift from perceptions just a few years ago.

As the head of one of the world’s most influential AI research labs and a key figure behind Google’s Gemini models, Hassabis’ assessment carries substantial weight. It challenges the long-held narrative that China remains structurally far behind in advanced AI development.

A Narrowing Performance Gap

Hassabis explained that Chinese models have progressed rapidly, particularly over the past two years. He noted that the gap appears much smaller today than it did one or two years ago, driven by aggressive investment, strong engineering talent, and a growing ecosystem of AI startups and large technology firms.

This progress became especially visible last year when Chinese AI lab DeepSeek released a high-performing model that surprised markets. The model demonstrated competitive reasoning and language capabilities despite being trained on less advanced chips and at significantly lower cost than many U.S. alternatives. While the initial shock has faded, the underlying message remains: China can build efficient and capable models even under constraints.

Since then, Chinese technology giants such as Alibaba, along with startups including Moonshot AI and Zhipu, have launched increasingly sophisticated large language models that perform well across benchmarks and real-world applications.

Catching Up Versus Leading the Frontier

Despite acknowledging China’s rapid catch-up, Hassabis drew a clear distinction between narrowing the gap and redefining the frontier of AI research. He argued that Chinese firms have not yet demonstrated the ability to produce fundamental breakthroughs that push the technology beyond existing paradigms.

According to Hassabis, the critical question is whether Chinese labs can originate transformative ideas, rather than refining or scaling known approaches. He pointed to the transformer architecture, introduced by Google researchers in 2017, as an example of the kind of breakthrough that reshaped the entire AI landscape and enabled modern large language models such as ChatGPT and Gemini.

In his view, China has proven it can get very close to the frontier, but it has not yet shown evidence of inventing the next foundational leap that would move the frontier itself.

Industry Leaders Acknowledge China’s Progress

Hassabis is not alone in recognizing China’s advances. Nvidia CEO Jensen Huang has previously said the U.S. is “not far ahead” in the AI race. Huang highlighted that while the U.S. maintains a strong lead in advanced chips, China is competitive in infrastructure, energy, and AI model development.

These comments reflect a growing consensus among industry leaders that China is no longer an AI laggard, even if the U.S. retains advantages in key areas.

Chip Restrictions and Structural Challenges

One of the most significant obstacles facing Chinese AI companies remains access to advanced semiconductors. U.S. export controls restrict the sale of Nvidia’s most powerful chips, which are critical for training large-scale frontier models.

Although U.S. authorities have signaled openness to approving sales of Nvidia’s H200 chip to China, this processor still falls short of Nvidia’s most advanced offerings. As a result, Chinese firms continue to rely on a mix of older hardware, efficiency optimizations, and domestically produced chips.

Companies such as Huawei have made progress in developing homegrown alternatives, but performance gaps with Nvidia’s top-tier chips persist. Some analysts believe that over time, this hardware disadvantage could cause the AI capability gap between the U.S. and China to widen rather than shrink.

Richard Clode, a portfolio manager at Janus Henderson, has argued that the U.S. may already be at peak relative vulnerability, with its superior AI infrastructure likely to compound advantages over the coming years as models are iterated and scaled.

Even Chinese executives have acknowledged these challenges. Alibaba’s Qwen technical lead Lin Junyang recently suggested there is less than a 20 percent chance that a Chinese firm will surpass U.S. tech giants in AI within the next three to five years, citing a U.S. computing infrastructure that is one to two orders of magnitude larger.

Innovation Culture Versus Technical Limits

Hassabis, however, attributes the lack of frontier-defining breakthroughs more to innovation culture than to hardware constraints. He compared DeepMind to a “modern-day Bell Labs,” emphasizing an environment designed for exploratory, high-risk scientific research rather than incremental scaling.

He acknowledged that China possesses world-class engineering talent capable of executing at scale, but argued that original scientific invention is far harder than replication or optimization. Creating something genuinely new, he said, is exponentially more difficult than improving what already exists.

DeepMind and the Global AI Race

Hassabis remains one of the most influential figures in global AI. DeepMind, which he founded over a decade ago and which Google acquired in 2014, has been central to Alphabet’s AI strategy. The lab’s work underpins Google’s Gemini models, including the recently launched Gemini 3, which has been positively received as Google seeks to demonstrate competitiveness with rivals such as OpenAI.

As the AI race intensifies, Hassabis’ remarks underscore a nuanced reality: China is much closer to the U.S. than many assume, but leadership in AI will ultimately depend on who defines the next frontier, not just who reaches the current one fastest.

‍