Walk into any AI lab in San Francisco or Palo Alto and you'll find engineers who grew up in Shenzhen, studied in Beijing, and now build the models that everyone from Sacramento to Washington is scrambling to regulate. That's not a footnote to the American AI boom. It's the boom. And this week that arrangement cracked open into one of the nastiest ideological brawls California's tech industry has seen in years, with some of the loudest voices in venture capital and AI policy demanding the door slam shut on Chinese researchers and Chinese-made models alike.
Related: THE GREAT DIAMOND HOAX THAT FOOLED SAN FRANCISCO'S FINEST IN 1872 · THE TRAITOROUS EIGHT: HOW A 1957 WALKOUT IN MOUNTAIN VIEW BUILT SILICON VALLEY
The fight isn't playing out in a courtroom or a congressional hearing room. Not yet. It's happening on X, in Substack posts, at closed-door dinners in Atherton and Hillsborough, where the people who built the modern AI industry are now arguing about whether it can survive without the talent pipeline some of them want to torch.
On one side: a bloc of investors and policymakers, some with direct ties to the Trump administration's AI posture, pushing to restrict Chinese nationals from sensitive AI research roles and to keep Chinese-developed models like DeepSeek out of U.S. government and enterprise systems. On the other: founders and engineers, many immigrants themselves, warning that California's AI dominance was built by exactly the kind of global recruiting the restrictionists want to choke off.
The DeepSeek Hangover
None of this can be separated from what happened last year, when a relatively obscure Chinese lab called DeepSeek dropped a model that matched or beat American frontier systems at a fraction of the training cost. Widely reported market reactions briefly wiped hundreds of billions of dollars off U.S. tech stocks in a single session. That shock never really faded. It just went underground, resurfacing every time a new export control proposal or visa restriction gets floated in Washington.
For the restrictionist camp, DeepSeek is Exhibit A: the U.S. can't keep training its geopolitical rival's engineers on American soil and then watch some of them carry that knowledge home. For the other side, DeepSeek proves the restrictions don't even work. A determined Chinese lab built a competitive model anyway, chip embargoes and visa politics notwithstanding.
Who's Actually in the Room
OpenAI's research arm, Anthropic in San Francisco, Google DeepMind in Mountain View. California's AI labs have spent a decade recruiting the best machine learning talent on Earth, and a significant share of that talent, by every available account, trained at Chinese universities before landing in the Bay Area on student and work visas. Multiple published studies on AI research authorship have found Chinese-born researchers make up a large share of top-tier AI paper contributors at U.S. institutions.
That's the math the restrictionists have to reckon with. Figures aligned with the administration's AI policy shop, including voices close to White House AI adviser David Sacks, have pushed for tighter vetting of visa applicants tied to Chinese state-linked universities and research programs. Critics inside the same tech ecosystem, including prominent venture capitalists who've spent careers courting global engineering talent, argue that kind of blanket suspicion will just send the next generation of AI researchers to labs in Toronto, London, or Singapore instead of Menlo Park.
Neither side is wrong about the stakes. They're just weighing them differently.
The Chip Wall Nobody Can Fully Build
Underneath the talent fight sits the older, thornier fight over hardware. Export controls on advanced Nvidia chips bound for China have been tightened, loosened, and tightened again over the past two years, with California-based Nvidia caught in the crossfire every time. The company's leadership has repeatedly warned that overly aggressive restrictions only accelerate China's push to build its own chip supply chain, undercutting the very dominance the controls are supposed to protect.
That argument has picked up unlikely allies in Silicon Valley's more libertarian-leaning wing, who see chip and visa restrictions as two branches of the same mistake: trying to wall off an industry that runs on open collaboration and global capital flows. The counterargument, gaining traction in Washington and among some AI safety researchers, is that the walls exist for a reason, and that a five-year head start matters when the technology touches military applications, surveillance systems, and critical infrastructure.
What Happens in California, Happens Everywhere
The reason this fight matters beyond the Bay Area is simple. California isn't just where American AI policy gets debated. It's where it gets built. Immigration rules that reshape who can work at OpenAI or Anthropic don't stay theoretical. They show up in hiring freezes, relocated research teams, and slower model releases. Decisions about which foreign-made AI tools get greenlit for U.S. government use ripple straight through California's enterprise software economy, since so much of it is built to interoperate with whatever the federal government blesses.
Nobody in this fight is proposing a tidy compromise. The restrictionists want tighter vetting now, arguing every year of delay hands China more ground. The open-borders camp wants the pipeline left alone, arguing the real threat is an America that trains the world's best AI engineers and then tells them to go build for somebody else.
Both camps agree on exactly one thing: whoever wins this argument in the next year effectively decides who builds the next generation of artificial intelligence, and where they build it. California doesn't get to sit this one out.