← All posts / Industry

Thirty to One: The Lopsided Flow That Explains the US-China AI Talent Race

A Carnegie Endowment study reveals that for every 30 Chinese AI researchers in the US, only one made the reverse journey to China — Beijing's open-door visa push is landing, but almost nobody is walking through it.

Thirty to One: The Lopsided Flow That Explains the US-China AI Talent Race

For two years, China has run one of the most aggressive talent recruitment campaigns in modern science. It introduced a dedicated visa category for scientists, dangled lucrative research grants, and made attracting the world’s best minds an explicit national priority — even as the Trump administration moved in the opposite direction, making it harder for American companies and universities to hire foreign researchers.

The results, according to a major new study from the Carnegie Endowment for International Peace, are humbling for Beijing: foreigners are not coming. Worse, the number of Chinese scientists working in the United States has risen rather than declined.

The 30-to-1 ratio

The starkest number in the Carnegie data, reported by the New York Times on October 6, is the flow ratio. For every 30 Chinese AI researchers working in the United States in 2025, only one researcher in China made the opposite journey. The bilateral talent pipeline, in other words, remains overwhelmingly one-way — despite tariffs, visa friction, export controls, and two years of intensifying geopolitical hostility.

“Despite U.S.-China tensions, the number of Chinese-origin researchers working in the U.S. actually rose by four percentage points rather than declined” compared with three years earlier, said Damien Ma, the Carnegie China director who led the study.

The study’s methodology is straightforward but rigorous. Carnegie’s team analyzed authors of papers accepted at NeurIPS 2025 — the field’s premier machine learning conference, which accepted 5,823 papers from 25,677 authors this cycle — and traced where each researcher went to college, earned graduate degrees, and ultimately works. The 2025 cohort covers more than 10,000 researchers with complete career paths, roughly 40 percent of all authors. NeurIPS acceptance remains a reliable proxy for elite status; the study calls these researchers the “Navy Seals” of AI. Notably, the team used Claude Code and OpenAI Codex to automate the career-path data work, with human verification layered on top — an AI industry studying itself with its own tools.

China’s retention win, recruitment loss

The picture is not uniformly bad for China. On the retention side, the trend line has genuinely flipped: the share of Chinese-origin researchers who end up working in China rose from 57 percent in 2022 to 69 percent in 2025. China now hosts 41 percent of the world’s top AI researchers by this metric, up from 27 percent in 2022, while the US share fell from 46 percent to 34 percent. Chinese-origin talent — defined by undergraduate origin — now accounts for 57 percent of the global elite pool.

Carnegie attributes domestic retention to two forces: China’s booming AI industry makes it easier than ever for Chinese graduates to find world-class work at home, and tighter US visa policy — particularly for Chinese STEM graduate students — makes America a harder destination to reach. In other words, China is winning the talent war at home largely by default: its own best people can stay, and Washington has raised the cost of leaving.

But inbound recruitment is a different story. The visa-for-scientists program, the generous grants, the open-door rhetoric — none of it has produced a measurable counterflow of foreign researchers into Chinese labs. The one-way street runs on.

Why the door stays shut

Several structural factors work against Beijing’s pitch. Language and research culture remain real barriers for Western researchers, who can find elite positions in the US, Europe, or increasingly in other Asian hubs without uprooting into an unfamiliar system. Geopolitics compounds the problem: the same tensions that make America unwelcome to some also make China a complicated line on a CV, with potential re-entry friction into the US ecosystem afterward.

China’s own behavior sends mixed signals too. Even as it courts foreign scientists, it has been tightening exit controls on its domestic stars. Bloomberg reported in May that China restricts overseas travel for top AI professionals at firms like Alibaba and DeepSeek, and by late September those curbs had been extended to their family members. A country that fences its own talent in while inviting outsiders to jump the fence is selling a complicated value proposition.

There is also an asymmetry of need. Chinese labs are flush with homegrown PhDs and returnees; foreign hires are a nice-to-have, not a bottleneck. American labs, by contrast, remain structurally dependent on foreign-born talent — which is precisely what makes US visa policy the more consequential variable in this race.

The US paradox

The Carnegie data exposes a paradox at the heart of American AI policy. The United States is doing everything it can, policy-wise, to reduce its dependence on foreign researchers — and yet its magnetic pull keeps strengthening. The share of Chinese-undergrad researchers working in the US rose four percentage points even as Washington tightened screws on Chinese student visas.

That resilience has limits, though. South Korea and Singapore are quietly gaining: Korea now retains 77 percent of its top talent (second only to the US), and KAIST has leapt twelve places to rank as the world’s third-best producer of elite AI researchers. Europe continues to fade, and India’s share of top talent actually dropped two percentage points, with no Indian institution in the top rankings. The duopoly at the top is consolidating — but the feeder system beneath it is shifting.

What it means

The 30-to-1 ratio is the number to remember from this study. It captures, in a single fraction, the true state of the US-China AI talent race: China has solved retention and lost on recruitment; America has wounded its own pipeline and remains the destination anyway.

For Beijing, the lesson is that money and visas cannot quickly buy what decades of accumulated ecosystem — peer networks, credible institutions, career optionality — provide. For Washington, the lesson is that its lead is more fragile than it looks: the four-point rise happened despite policy, not because of it, and every visa denied is a small donation to competitor hubs in Seoul, Singapore, and yes, eventually, Shenzhen.

The talent ledger has flipped, but the traffic has not. In AI, as in physics, momentum takes a long time to reverse — and a very short time to waste.