Leading AI researchers and executives in China are presenting conflicting views on the country's ability to catch up with the United States in artificial intelligence, with the central bottleneck being access to advanced computer chips. Demis Hassabis, CEO of Google's DeepMind, stated that China's AI capabilities may lag only a few months behind the U.S., citing rapid progress from tech giants like Alibaba and startups such as Moonshot AI, Zhipu, and DeepSeek.
However, this optimistic assessment is countered by prominent Chinese AI leaders. Tang Jie, founder of Zhipu, told a Beijing conference that "the truth may be that the gap is actually widening." Justin Lin, who leads development of Alibaba's Qwen AI model, estimated the odds of any Chinese company surpassing American leaders like OpenAI and Anthropic in the next three to five years at 20% or less.
The core challenge stems from U.S. export controls that prevent Nvidia from selling its most advanced chips directly to China. When Nvidia launched its new Rubin hardware in January, it named several American firms as buyers but excluded all Chinese AI developers. Chinese companies are exploring workarounds, including renting computing power from data centers in Southeast Asia and the Middle East to access these chips.
According to UBS analysts, China's biggest internet companies spent approximately $57 billion on capital projects last year, with much directed toward AI. This represents roughly one-tenth of what American companies invested. Justin Lin highlighted the resource disparity, noting that "a massive amount of compute at OpenAI and other American companies is dedicated to next-generation research, whereas we are stretched thin."
Despite these challenges, Chinese firms have demonstrated remarkable efficiency. DeepSeek has published research on methods to build larger AI models with fewer chips, and according to Epoch AI, models from DeepSeek and Alibaba have closed the performance gap with top American models to just four months, down from seven months on average in recent years.
The recent U.S. approval for Nvidia to sell its H200 chip to China is unlikely to significantly alter the landscape, as industry insiders note the H200 is two generations behind the Rubin line and too weak for training top-tier AI models. Nvidia's China revenue dropped 45% year-over-year to about $3 billion in the most recent quarter, though the company achieved overall revenue of $57 billion.
Looking forward, a potential long-term threat to Nvidia's dominance could emerge if Chinese developers, forced to use domestic chips, create open-source software tools that gain global adoption, potentially undermining Nvidia's CUDA software platform that locks developers into its ecosystem.
