Talent: The Ultimate Variable in the US-China AI Competition

In the narrative of US-China AI competition, computing power blockades and energy bottlenecks have long dominated the headlines. Yet mounting evidence suggests that what truly determines the outcome of this technological contest is neither the number of GPUs nor the power supply to data centers — it is talent. More critically, the trend is reversing.

The Overlooked Ultimate Variable

Over the past two years, US chip export controls on China have tightened layer by layer, from the H100 to the H20, with the blockade line continuously moving downward. Meanwhile, the global battle over power supply for AI data centers has intensified. These two issues have captured the vast majority of attention from policymakers and media alike.

But a 2025 report from the McKinsey Global Institute points out that the total number of top-tier AI researchers worldwide does not exceed 50,000, with core engineers possessing large model pre-training experience likely numbering fewer than 5,000. This figure reveals a harsh reality: Chips can be manufactured, power plants can be built, but cultivating top AI talent takes 10 to 15 years and cannot be rapidly replicated through capital.

Data from Stanford University's 2025 AI Index Report is even more telling: among the world's most influential AI papers, the disconnect between researchers' nationalities and their work locations is undergoing profound change. This means whoever can attract and retain this "scarce resource" holds the lifeline of AI competition.

The Reversal Has Begun

For a long time, the flow of China's top AI talent followed a single pattern: domestically educated, advanced studies in the US, employed in America. According to earlier statistics from the MacroPolo think tank, approximately 29% of the world's top AI researchers received their undergraduate education in China, but more than half of them ultimately chose to stay and work in the United States. This "talent deficit" was once regarded as the greatest hidden concern for China's AI development.

However, multiple recent data points indicate that the trend is reversing.

First, the return flow is accelerating. Between 2023 and 2025, the number of high-caliber AI professionals returning to China has grown significantly. Recruitment agency data shows that the proportion of Chinese researchers from Silicon Valley's top AI labs returning home has nearly tripled compared to 2020. Among the core teams of prominent Chinese companies such as DeepSeek, Moonshot AI, and Zhipu AI, the share of returnee technical leaders continues to climb.

Second, the American "push factor" is intensifying. Tighter visa policies, rising geopolitical sensitivity, and scrutiny cases targeting Chinese-American scientists are eroding America's appeal to Chinese AI talent. Several Chinese AI researchers, who spoke on condition of anonymity, stated that "a sense of belonging and security" has become a significant factor in their consideration of returning to China.

Third, China's "pull factor" is strengthening. The explosive growth of China's domestic AI industry provides an unprecedented stage. From compensation packages to research freedom, from startup ecosystems to policy support, China is building a more competitive talent attraction framework. Cities including Beijing, Shanghai, and Shenzhen have successively launched dedicated AI talent programs, and the comprehensive packages offered to some top researchers now match or even exceed those in Silicon Valley.

The Deeper Logic of the Talent Race

Why is talent a more fundamental variable than computing power? The answer lies on three levels.

From a technical perspective, breakthroughs in AI are still essentially "human breakthroughs." Whether it is the creation of the Transformer architecture or DeepSeek's innovations in reinforcement learning, these advances are driven by the insights of a very small number of brilliant researchers. Computing power is an amplifier, but without original ideas, even the most GPUs are nothing but silent silicon.

From an industrial perspective, deploying AI applications requires a large number of interdisciplinary professionals who understand both technology and real-world scenarios. China boasts the world's richest landscape of application scenarios but has long faced a structural contradiction of "having scenarios but lacking talent." The return of talent is bridging this gap, dramatically improving the efficiency of connecting technology with markets.

From a strategic perspective, the effectiveness of computing power blockades is being eroded by technological substitution and architectural innovation. DeepSeek's case of achieving high-performance models at relatively low computing costs has already proven that top talent can find "asymmetric pathways" under resource-constrained conditions. In other words, the higher the talent density, the lower the marginal effectiveness of computing power blockades.

Concerns and Opportunities for Both Sides

For the United States, the risk of talent drain should not be underestimated. America's AI leadership has been built largely on its global talent siphoning effect. If this effect weakens, its innovation engine will face a "fuel shortage" dilemma. Recently, multiple US lawmakers and technology leaders have called for immigration policy reform to provide faster green card pathways for AI talent.

For China, while a window of opportunity has opened, the challenges are equally daunting. How to build a world-class basic research environment, how to foster an academic culture that encourages originality rather than imitation, and how to cultivate homegrown top talent while attracting returnees — the answers to these questions will determine whether the talent return can be converted into a lasting competitive advantage.

Looking Ahead

Over the next five years, the focal point of US-China AI competition will gradually shift from "hardware blockades" to a comprehensive contest over "talent ecosystems." Chips face the ceiling of Moore's Law, power is constrained by physics and the environment, but the creativity of talent has no upper limit.

The trend is reversing, but the magnitude and speed of that reversal still depend on the policy wisdom and institutional flexibility of both sides. What can be foreseen is that the nation capable of providing the best soil for the world's brightest AI minds will ultimately secure an unshakable position in this contest of the century.

In this race with no finish line, the scarcest resource has never been silicon — it is ideas.