Wu Yonghui spent 17 years at Google. He rose to vice president of research at DeepMind, earned the rare title of Google Fellow, and then, in February 2025, walked out to run foundational AI research at ByteDance. A few months later, Yao Shunyu, a 27-year-old who helped build OpenAI's first AI agent products, took a reported 100-million-yuan package to become Tencent's chief AI scientist. He reports directly to Tencent president Martin Lau.
These are the names people noticed. The number that matters is the one behind them: three AI headhunters told the Financial Times they relocated more than 30 US-based researchers to China over the past twelve months, up from single digits a year earlier. Separately, at least 85 established scientists joined Chinese research institutions from the US during 2025.
What's pulling them east
The pitch from Chinese labs is not about salary alone, though the money is real. Yao's deal, if the Bloomberg reporting holds, puts his total compensation around $14 million. Wu Yonghui, according to South China Morning Post, was drawn by the promise of "new, exploration-oriented" work, which is what senior researchers always say. More telling is where he landed: ByteDance's Seed division, a team that barely existed before ChatGPT launched and is now structured like an independent research lab reporting to the CEO.
But the deeper pull is deployment. China is not debating whether to put AI into factories and taxi fleets and financial systems. It's already doing it. Shenzhen, in particular, has become something approaching a robotics city-state. The Shenzhen government counts over 57,000 robotics-related enterprises in the city. Its 2025-2026 action plan targets 3,000-plus AI companies and more than 10 unicorns by the end of this year. A pilot production line for humanoid robots just launched in the Longhua district, capable of assembling a unit in two hours.
For an engineer who spent years training models on benchmarks, the chance to watch a robot you helped design actually pick up boxes in a Zeekr electric vehicle factory is a different kind of career satisfaction. Steve Hsu, a computational mathematics professor at Michigan State University, has pointed out that Shenzhen alone hosts at least 100 humanoid robotics companies, making physical proximity almost a job requirement for anyone in the field.
The push from the other side
There is a less flattering version of this story, and it involves American immigration policy. In May 2025, the US announced it would actively revoke visas for Chinese students in fields including AI and semiconductors. The Rest of World interviews with returnees paint a consistent picture: H-1B anxiety, political tension, and a general sense that the welcome mat has been pulled back. "The political tensions have made the working and living environment, particularly those in North America and some in Europe, not as comfortable, not as safe and ideal as it used to be," said Lili Yang, an assistant professor at the University of Hong Kong who studies reverse migration of Chinese researchers.
Lizzi Lee of the Asia Society Policy Institute calls the trend a "slow acceleration." That framing may be too generous. When Jensen Huang told a private dinner of Taiwanese tech executives that China has "one million people working on this 24/7" compared to "maybe 20,000 in Silicon Valley," the leaked remarks, reported by entropytown, triggered days of argument. His numbers are squishy (the definition of "working on AI" does a lot of heavy lifting), but the direction is not in dispute.
The pipeline problem
The individual departures grab headlines. The scarier data point is upstream. About 20 percent of Tsinghua University's engineering graduates now apply for PhD programs in the US, down from roughly 50 percent before Covid. That number, cited in Financial Times coverage of the broader trend, represents a structural shift in where the next generation of Chinese AI talent plans to build its career. They're not leaving and coming back. They're just not leaving.
Carnegie Endowment research tracking 100 Chinese-origin AI researchers who were US-based in 2019 found that many have stayed put, but the broader trend is clear: China's domestic AI industry now offers credible alternatives to American labs. MacroPolo data shows only 42 percent of top AI researchers worked abroad in 2022, down 13 percentage points from 2019. Chinese universities now produce nearly half the world's elite AI talent.
And then there's the Stanford HAI 2026 AI Index, released yesterday, which concluded that China has effectively erased the US performance gap in AI. The report notes that over 90 percent of notable AI models now come from private companies, and that both US and Chinese labs have stopped disclosing training details, making external comparison harder. The talent migration and the benchmark convergence are feeding each other.
So what does this actually mean
Silicon Valley still has venture capital, startup culture, and a density of experienced operators that no Chinese city can match. Lu Zhang, a VC based in the Valley, argues the US ecosystem remains unmatched for rapidly scaling ideas. She's probably right, for now. But the talent pipeline that fed American AI dominance was always partly a Chinese talent pipeline, and that pipe is narrowing.
The Chinese labs are also changing strategy. A year ago, the story was DeepSeek and open-source models. Now Alibaba's Qwen3.6-Plus, ByteDance's Seedance 2.0, and Tencent's Hunyuan are all closed-source commercial offerings. You open-source when you can't charge for frontier work. You close when you can. Wu Yonghui's hiring and the shift to proprietary models are the same story from two angles: the Chinese labs now think they have the talent density to compete as commercial operations, not just research projects.
Shenzhen's next major robotics fair opens April 24, with 180 robot companies and 300 international buyers expected. The US has no equivalent event on the calendar.
