I’ve left the People's Republic of California and have now settled back into my home in Texas. You might wonder why I traded the Mediterranean climate of the Bay Area for the searing Texas heat. Ultimately, we were away from home for a long time, and we wanted some downtime before the school year began.
Most faculty in my position don’t take their sabbaticals—and if they do, they usually stay home. I get it. Moving a wife and kids across the country is a lot of work. It was a whirlwind year—high cost, high benefit. Let me share my biggest observation from the year.
China’s Superior AI Talent
For all of California’s political mismanagement, the Bay Area remains a special place. The entire digital revolution began there—from the first semiconductors to the transistor, and now to neural networks—and everything in between. The concentration of talent and capital is unreal, and I don’t see anyplace in the U.S. displacing it anytime soon. America is lucky it all began here—though it may not end here.
Up close, I saw the growing strength of China’s AI talent. You might assume it’s all in Hangzhou with DeepSeek, but there’s a fast-growing population of Chinese nationals dominating the AI software space here in the U.S.—and no one really talks about it.
The conventional wisdom in American media is that the U.S. has a commanding lead in AI, given that foundational models (OpenAI, Anthropic, Google, and Meta) are here. But the truth is more nuanced. Jensen Huang is one of the few public voices who is concerned that 50% of AI researchers are Chinese. From what I’ve saw, it’s closer to 90%+.
Growing up on the East Coast, I always heard that Americans were the innovators while the Chinese were the copycats. But not so in AI. China has a cultural affinity for math and science, a vast talent pool, and a relentless work ethic. And to be clear—I don’t think these students have any sinister motives or CCP ties. At Bay Area machine learning conferences, I found way more Chinese researchers than Americans, by factor of five. I guess this is what you get in our culture obsessed with Taylor Swift and Kim Kardashian. America’s fixation on sports, entertainment, woke politics, and government spending is finally showing its ugly head.
I dove deep into the open source AI research community, which is largely focused on AI inference. It started by chance—I met a Stanford grad named Kobe who founded an inference startup last year. His team, which primarily communicates on Slack in Mandarin, is focused on how to serve LLM requests at scale using limited GPU resources. I hadn’t even realized that was a problem until I began attending half a dozen inference-focused meetups in the Bay Area.
The more time I spent with Kobe and his friends, the more I respected them. They’re sharp, hardworking, collaborative, and most importantly, committed to open source software and research. It’s somewhat ironic that this community of Chinese AI researchers share all their models and progress, while American companies like OpenAI hide behind closed walls and disclose nothing.
America still leads in AI hardware (Nvidia > Huawei), and possibly still in AI platforms. But the direction of change at the talent level is in China’s favor. Zuckerburg and Musk both know this, as their heads of their AI labs are both Chinese.
But so what? One country has an edge over another in one specific domain. Well, it’s gets more complicated…
American Taxpayers are funding China’s AI Talent
At Berkeley and Stanford, nearly all computer science PhD students are Chinese nationals. While computer science has long been international, I was surprised by the sheer number of Chinese students above all other nationalities. Here in Texas, graduate programs are a bit more diverse, with Indian and Chinese students neck and neck. But not so at Berkeley and Stanford. I asked faculty point blank how this happened. The answer was simple: the best AI students are Chinese. There are strong pipelines from Chinese universities into top U.S. academic labs.
Berkeley’s Skylab, for example, routinely recruits from China’s top universities. The top graduating PhD student will call his friends in Tsinghua to recruit the next incoming PhD student. These students aren’t just peons at big tech companies; many launch their own startups — some in America, some back in China. Skylab has built a flywheel: publish a paper, release an open-source implementation, launch a startup, get VC funding. This is the story behind vLLM, the leading open-source framework for serving large language models.
Unlike the undergraduate and master’s level, PhD education is fully funded. Faculty write grants whose purpose is to fund their labs. In AI, the main expense is labor —- PhD students. And almost exclusively, PhD funding for computer science comes from the National Science Foundation (NSF), which itself is funded through tax revenue. And so you, dear reader, are paying for Chinese AI talent.
Reasonable people can disagree on the size and scope of the NSF. But this gross mismatch between costs (American taxpayers) and benefits (Chinese nationals) in AI research makes no sense, even if AI wasn’t the most critical industry ever. I doubt Vannever Bush, dean of MIT Engineering and founder of the NSF, would have envisioned this outcome.
Like fiat money and Wall Street, this symbiotic dependence between universities, the NSF, and Chinese technical talent has grown slowly over decades to the current extreme situation. Universities have built their balance sheets and faculty their careers around this co-dependence, and no one wants to rock the boat. The broad-based cuts from DOGE put some pressure on university administrators, but they don’t address this issue. A more specific solution would be to limit NSF funding to US citizens, which would reduce the fiscal footprint of the NSF anyway. And then build out the American AI talent pipeline in partnership with industry.
I’m a fiscal conservative, not a xenophobe. Taxes are not to be taken lightly — impose them with care, and ensure that the benefits accrue to the taxpayers. That is not happening right now with AI research.