Amid the intensifying U.S.-China technological competition, DeepMind CEO Demis Hassabis recently gave an interview to CNBC's "Tech Briefing," offering an unexpected assessment of China's AI development: the gap between China's large models and those in the U.S. has narrowed to just "a few months," far from the so-called "generational lag" that some have claimed. This judgment directly challenges the underestimation of China's AI capabilities by certain Western media.

Hassabis specifically praised Chinese companies such as DeepSeek, Alibaba, and Moonshot, saying their model performance was "impressive," with training scale and reasoning capabilities approaching the global forefront. He acknowledged that China has even gained local advantages in AI infrastructure investment, engineering implementation, and the richness of application scenarios, demonstrating a strong catching-up speed.

However, he also pointed out a key dividing line: although China's AI excels in technical implementation and product iteration, it has yet to produce a truly "disruptive" original paradigm—scientific breakthroughs from 0 to 1, rather than optimization and evolution from 1 to N. In his view, scientific innovation is far more challenging than technical imitation, and currently, the Chinese AI ecosystem focuses more on efficient replication and rapid application, rather than exploring new architectures or fundamental theories.

Notably, Hassabis attributed this gap to "thinking patterns" rather than purely technical restrictions. Although he acknowledged that U.S. export controls on high-end AI chips have indeed limited China's ability to train ultra-large models, potentially widening the gap in the future, he emphasized that the real bottleneck lies in whether to encourage high-risk, long-term basic exploration. "Innovation requires a culture that tolerates failure and free interdisciplinary thinking. This is harder to replicate than computing power," he said.

This viewpoint has prompted deep reflection in the industry. On one hand, China's AI leads globally in the efficiency of implementation in vertical scenarios such as e-commerce, finance, and government affairs. On the other hand, in next-generation architectures after Transformer, the fundamental logic of embodied intelligence, and AI for Science, the agenda is still set by the U.S. Hassabis' comments are both an acknowledgment of China's engineering capabilities and a warning: to move from "parallel running" to "leading," it must shift from "doing quickly" to "thinking deeply."

Today, as the global AI competition enters deeper waters, computing power may determine short-term speed, but intellectual depth determines long-term height.