There is much to read between the lines whenever tech leaders make sweeping public pronouncements. Anthropic’s CEO Dario Amodei published a blog post this week categorically denying that the company had ever backed a ban on so-called open-weight models, appearing at first glance to defend himself against attacks from the US government’s recently departed AI czar David Sacks.
But read a little deeper, and Amodei is joining a recent trend of deflecting White House ire towards someone else. In this case, it is Nvidia.
His post, ‘Our position on open-weights models,’ argues that there are better ways to counter China’s AI efforts than banning its models that freely publish their core software components so anyone can use and modify them. These have looked like bad news for Anthropic’s business and upcoming public offer.
Sacks seized on that to point out Anthropic was alone in refusing to sign an open letter supporting open-weight AI; Nvidia CEO Jensen Huang had promoted the letter with his first-ever post on X. The industry needed to watch Anthropic like “like a hawk,” Sacks said.
Mark Zuckerberg piled on with more veiled criticism with many seemingly coordinated statements in the press which warned against curbs on Chinese AI—a position that also protects Meta’s own work on open-weight models—and publicly challenged Anthropic and OpenAI.
Such rules risked “regulatory capture” and “concentrating power,” he said, with no hint of irony from a billionaire whose statements were aimed at shaping regulation and whose social media platforms reach roughly 3.5 billion people each day—about 43% of the world’s population.
Amodei denies that framing. A ban “would protect US AI companies from competition,” he writes, “but that has never been my goal.” His primary worry is an authoritarian state like China building models more powerful than anything in the US and using them to achieve military supremacy or to repress their citizens.
You do not need open-weight models for that, he adds, and the best way to stop it from happening is to block China’s ability to build AI in the first place, through chips.
Huang probably got heart palpitations reading that. Last year, the Trump administration reversed Biden-era restrictions on Nvidia’s H20 chip exports to China, thus reopening what Huang described as a potential $50 billion market. Amodei has implied that ban shouldn’t have been lifted.
The Anthropic CEO is right on one point. The popularity of Chinese models like Qwen and DeepSeek has become unstoppable among US businesses, especially startups. So the toothpaste is effectively out of the tube.
“Open source has to win,” says Uljan Sharka, CEO of Italian AI company Domyn, arguing that companies do not want to depend on two or three AI firms for the technology they will run their businesses on. “It will win. It’s inevitable.”
Banning such models also shoots US entrepreneurialism in the foot, since new business owners benefit from using cheap AI agents to spin up a website or do their taxes. And a US ban would not stop those models from becoming globally popular.
But it is unclear that Nvidia should be the company that bears the economic cost of containing China. Amodei cites a six-year-old research paper on AI’s scaling laws in his reasoning for focusing on chips.
The paper says, in a nutshell, that an increase in computing power produces more capable AI. That still broadly holds true, but is no longer the whole story. Beijing now takes a different view to the US on AI supremacy being more about wide adoption than the smartest bot—this explains why the rise of its open-weight models matters so much.
Export controls also have not prevented China from making major AI breakthroughs over the past year. Chinese companies have compensated through efficient methods and algorithmic improvements. Export controls might slow China, but they cannot stop it.
The debate over free AI has for years been one about safety, since closed systems have guardrails that can stop bad actors from using them for harm. Now it is becoming entangled in business interests.
For the wider population, cheaper models hold genuine promise, potentially cutting cost burdens for businesses and startup founders. That is surely a good thing if AI is meant to boost economic productivity; but it could also severely disrupt the current bellwethers of a buzzy new industry that is propping up the US stock market.
As the White House works out how to contain China’s technology advancements, it is also figuring out which US companies can be weaponized in that fight. It is clear from this week’s public sparring between Amodei, Zuckerberg and Huang that Silicon Valley would rather have someone else conscripted. ©Bloomberg
The author is a Bloomberg Opinion columnist covering technology.