Why Top AI Researchers Are Fleeing America for China

The numbers that should alarm Washington

China now employs more top-tier AI researchers than the United States. It is the first time this has happened, and the data behind it is difficult to dispute.

A report from Carnegie China, released on September 23rd, analyzed publication records from NeurIPS 2025 — the most prestigious conference in artificial intelligence. Of the 25,677 authors who published there, 1,280 had complete education and work histories. Of those, 4,174 were working in China and 3,514 in the US. The ratio in 2022 had been 46 to 27 in America’s favor. Three years later, it flipped to 34 to 41.

This is not a marginal shift. It is a structural one.

How the flip happened

Two forces drove the reversal.

First, China’s AI industry grew fast enough to absorb its own graduates. In 2022, 56.6% of Chinese AI researchers stayed in China after graduation. By 2025, that number rose to 68.7%. The sample of Chinese graduates nearly quadrupled — from 1,460 to 5,903 — reflecting the explosive expansion of programs at Tsinghua, Peking University, and other institutions. These graduates found jobs at Baidu, Alibaba, Tencent, and dozens of startups that no longer need to poach from abroad.

Second, US visa policy became a headwind. The report notes that STEM graduate students from China face increasing difficulty obtaining US visas. Nonimmigrant visa rejection rates for Chinese applicants in technical fields have climbed steadily since 2018, and the screening process has grown more opaque. A researcher who might have come to MIT in 2019 now thinks twice — or goes to Shanghai instead.

The result: China went from exporting talent to hoarding it, while the US went from magnet to less attractive option.

The institution rankings tell an even sharper story

In 2022, Google was the number one employer of top AI researchers. MIT, Stanford, and Carnegie Mellon filled the rest of the top ten alongside a handful of Chinese universities. American institutions dominated.

In 2025, the top two spots belong to Peking University and Tsinghua University. KAIST of South Korea surged to third — a twelve-place jump from 2022. Shanghai Jiao Tong took fourth. MIT dropped to fifth. Stanford fell to sixth. Google slipped to seventh.

Five of the top ten institutions are now Chinese universities. The US has just four when you combine universities and companies. This is not a temporary fluctuation. Chinese universities have been systematically investing in AI labs, recruiting海外returnees, and tying research output to industrial partnerships for over a decade. The payoff is visible now.

The irony of net flows

Here is what makes the picture more nuanced than a simple “China wins, US loses” narrative.

The US remains the world’s largest net importer of AI talent. In 2025, it absorbed 2,300 incoming researchers while losing only 155. China, by contrast, lost 1,847 researchers abroad while gaining just 118. It is the world’s largest net exporter.

So China has more top AI researchers working within its borders, but it is still losing its best and brightest to the US and elsewhere. The difference is that fewer are leaving now than before — and those who do leave are increasingly choosing to return. Meanwhile, the US continues to pull talent from around the world, but the pool of Chinese talent arriving has shrunk.

Retention rates illustrate this. American-trained researchers stay in America at a rate of 88.7%. Korean graduates stay in Korea at 77.2%. Chinese graduates stay in China at 68.7%. China is retaining more of its own, but it still loses a significant share.

Where Korea fits — and why it matters

Korea deserves more credit than the data suggests. It is the only country outside China and the US where AI researcher representation increased between 2022 and 2025. Korean-born researchers now make up 5.7% of the global top-tier AI workforce, up from 4.8%.

KAIST’s rise from 15th to 3rd in institutional rankings is extraordinary. It signals that Korea has found a niche: producing high-quality AI researchers at a scale and speed that few countries can match. The Korean government’s sustained investment in STEM education and research infrastructure has paid off.

But there is a catch. Korea’s AI talent retention rate of 77.2% is strong regionally, but still below the US and Europe. Korean researchers who do go abroad often do not return. And while KAIST produces talent, the domestic industry — dominated by Samsung and Hyundai — has struggled to create the same kind of open, fast-moving AI ecosystem that attracts researchers in Silicon Valley or Beijing.

What happens next

Three scenarios are plausible.

In the optimistic scenario for the US, visa restrictions ease and the flow of Chinese talent resumes. America reclaims its position as the undisputed gravity center for AI research. This is unlikely in the current political climate, where technological competition with China remains a bipartisan priority.

In the base case, China continues to retain more of its graduates while the US gradually loses its edge in attracting them. The gap in domestic researcher numbers widens. American companies compensate by recruiting more heavily from India, Europe, and other regions — but those pools are smaller and more contested.

In the worst case for the US, Chinese AI companies begin matching or exceeding American compensation and research freedom, pulling even researchers who already moved to the US back toward China. This is not speculative. Chinese AI labs already offer competitive packages and the intellectual environment is less constrained by export controls and classification rules.

For Korea, the trajectory is ambiguous. KAIST’s rise proves the country can produce world-class AI talent. But producing talent and deploying it productively are different things. If Korea’s tech ecosystem cannot offer researchers the same kind of impact and compensation as China or the US, the gains from KAIST’s ranking improvement will remain largely symbolic.

The talent war is no longer about who attracts the most immigrants. It is about who keeps the researchers they produce. China is winning that battle. The US still wins the immigration battle. Korea is quietly becoming relevant in both — but relevance is not the same as leadership.