At the 2026 World Economic Forum in Davos, Demis Hassabis, CEO of DeepMind, provided a new assessment of the global AI landscape: **The overall technical strength of Chinese AI companies is now very close to the West, with the gap narrowing to about six months**—this judgment is more precise than previous claims of being "months behind," and reflects the rapid catching up of China's large models.
Hassabis specifically praised the **DeepSeek R1 version**, calling its performance "impressive," which even caused a stir in Silicon Valley, leading to fluctuations in the stock prices of chip giants like NVIDIA. However, he also pointed out that although Chinese companies have performed well in engineering implementation, training scale, and product application, **they have yet to demonstrate truly breakthrough original technological capabilities**. "They can efficiently reproduce and optimize existing architectures, but they have not yet proven the ability to define the next generation of paradigms," he emphasized. He highlighted that scientific disruptive innovations—such as new foundational architectures after Transformer or the underlying theories of embodied intelligence—are still dominated by the West.
Notably, Hassabis expressed a positive attitude towards recent U.S. policy adjustments. He pointed out that **the Trump administration has eased restrictions on AI chip exports to China**, which not only helps ease global supply chain tensions but also opens up a larger market for U.S. tech companies, including Google and NVIDIA, forming a "technology export—data feedback—model iteration" virtuous cycle.
As a core participant in Google's Gemini AI assistant project, Hassabis revealed that DeepMind is fully committed to driving AI's transition from "digital intelligence" to "physical intelligence." His team continues to invest heavily in robotics, aiming to enable AI systems to perceive, reason, and act in the real world, with the goal of achieving key breakthroughs in embodied intelligence.
This speech is an objective acknowledgment of China's AI progress, as well as a clear reminder: In today's era where computing power, data, and engineering capabilities are becoming increasingly similar, **the real moat is not who runs faster, but who sees further**. As the global AI competition enters deeper waters, the ability to raise new questions and explore new paths will become the key to determining the landscape of the next decade.
