Chinese AI labs open US offices to recruit from OpenAI and Anthropic

Chinese AI companies are hiring engineers and researchers in the United States, close to the staff of OpenAI and Anthropic. MiniMax, the Hong Kong-listed startup, lists about 16 open roles in San Francisco, Rest of World reports. They include a research lead for its core model and a lead recruiter for top AI talent in Silicon Valley and Seattle. Moonshot AI, which makes the Kimi model, is looking for US-based solutions architects to help enterprise customers build its models into their businesses.

Until now, Chinese labs have mostly relied on people trained and hired at home. Experts told Rest of World that hiring in the US is a strategic shift: it gives the labs access to talent and keeps them close to one of their most lucrative markets. The data so far suggest the labs will find it hard.

Where the researchers are

A September study by Carnegie China found that more than half of the world's top AI researchers did their undergraduate studies in China. The study based this on the educational backgrounds of people who published papers at NeurIPS. It also found that the US remains the main destination. In 2025 the US had a net gain of 2,145 researchers. China had a net loss of 1,729.

Chinese companies have tried to bring some of these people back. Tencent hired former OpenAI researcher Yao Shunyu to lead its AI research. A headhunter recruiting for Chinese labs in the Bay Area told Rest of World that clients want "little geniuses" who have worked at OpenAI or Anthropic. The headhunter said these efforts have mostly failed on salary.

"Wanting to establish research teams in Silicon Valley makes perfect sense for Chinese AI companies seeking access to talent and the latest developments in the industry," said Cheryl Wu, who researches the economics of AI at Yale University and the think tank GovAI.

The pay gap

Wu's explanation is direct. "It's money. Chinese companies simply don't have enough of it to offer competitive wages for talent," she said. With the research institute Epoch AI, she published a report on how Chinese AI companies are financed.

Company filings show the size of the gap. MiniMax spent $84.3 million on employee pay in 2025. With 415 staff at year end, that is about $203,000 per person. Z.AI, another Hong Kong-listed lab, had annualised pay costs of $140,000 per employee in the first half of 2026. Both figures include share-based compensation and staff in every department. Neither company breaks out pay for researchers.

OpenAI advertises base salaries of $250,000 to $445,000 for research engineers in San Francisco, before equity. Three Chinese researchers working in Silicon Valley told Rest of World that total packages above $1 million a year are common among their peers.

The gap in pay follows a gap in revenue. According to Wu's research, as of September China's six leading AI firms together made about 10% of the combined AI-related revenue of OpenAI and Anthropic.

Some Chinese companies are now bidding harder. A person familiar with ByteDance's frontier model team, Seed, said that when a researcher holds an offer from OpenAI or Anthropic, Seed will try to beat it by 1.5 times. Seed hired Wu Yonghui, who spent 17 years at Google, as head of research in early 2025. It currently lists 71 positions in San Jose, San Diego and Seattle.

Prestige and working hours

Money is not the only factor. Wu said senior Meta engineers may accept a lower-ranking job to join Anthropic because of its frontier work. "That makes it even harder to persuade top talent to join a lab whose models are just not at the cutting edge," she said.

Three Chinese researchers in the US told Rest of World that even with equal pay, they prefer American companies. They gave two reasons: better work-life balance and the chance to work on frontier models. Damien Ma, director of Carnegie China, thinks working culture matters less. He said Silicon Valley is "the only other place where '996' isn't really frowned upon". 996 is the Chinese tech industry habit of working from 9 a.m. to 9 p.m., six days a week.

Not every Chinese lab is looking abroad. DeepSeek still advertises research and engineering roles in China and recently announced about 150 positions for engineers working on agentic systems and back-end infrastructure. In a 2024 interview, founder Liang Wenfeng said he hires mostly Chinese-educated people and believes the company can train elite researchers itself.

National security

Washington has spent years restricting Chinese access to advanced Nvidia chips. It is also scrutinising Chinese use of overseas data centres and cloud services. Nothing comparable stops Chinese AI companies from opening US offices and hiring there.

Both governments are tightening control over advanced AI. China has proposed export controls on some of its most advanced models, particularly those trained on what it calls "important data". US officials and American AI companies have accused Chinese labs of using unauthorised distillation to copy their models' capabilities. In semiconductors, the US already requires a licence before some American engineers can work in Chinese facilities.

Ma expects talent to become another point of conflict. "Given it is already so competitive, it doesn't take much to imagine U.S. companies at some point capitalizing on U.S.-China competition and making national security arguments that will make it more difficult for Chinese labs to operate or to attract the best talent," he said. He added that the same would apply to US labs operating in China. The talent pool in question is small: Ma put it at roughly the top 5 to 10 percent of people in the field.

There is a precedent. In 2019, after Washington put Huawei on a trade blacklist, the company cut more than 600 jobs at Futurewei, its US research arm.

What it means for work

The contest is over a very small group: the researchers who build frontier models. For them, a Chinese lab in San Francisco is one more bidder in a market already led by OpenAI and Anthropic. For now that bidder pays less and offers less prestige, and its offices could be exposed to future policy changes. The roles Chinese labs can fill more easily are lower down the chain, such as Moonshot's solutions architects and MiniMax's specialists in finance, law, life sciences and cybersecurity. These jobs sell and adapt models more than they build them. It is still unknown whether higher offers like Seed's can outweigh the pull of frontier work, and whether either government will treat hiring across borders the way it already treats chips. "They may have plans to hire, but that doesn't necessarily mean they'll be able to find the people," Wu said.