Dario Amodei, the CEO of the AI company Anthropic, has responded to the current hysteria in his industry and the financial markets around a new and surprisingly advanced Chinese AI model called DeepSeek by saying it proves the United States needs export controls on chips to China in order to ensure China doesn’t “take a commanding lead on the global stage, not just for AI but for everything.”
As I wrote earlier this week, Amodei believes that DeepSeek’s current advantages over American AI companies are overstated and temporary. The true cost of the DeepSeek R1 is not entirely clear and almost certainly much higher than DeepSeek’s paper claims because it is building on previous research published by American companies and DeepSeek’s own previously released V3 model. Additionally, Amodei argues that American companies will be able to recreate the same efficiencies in their model training soon, if they haven’t already, and then gain the lead again when those efficiencies are paired with American companies’ much greater access to more and better. The US already has export controls on chips to China, and Amodei argues that DeepSeek shows that they are “more existentially important than they were a week ago.”
At the same time, Amodei believes that “making AI that is smarter than almost all humans at almost all things will require millions of chips, tens of billions of dollars (at least), and is most likely to happen in 2026-2027.” Multiple American companies, Amodei says, will definitely have the money and chips this requires. The important question, and the reason the US needs export controls on chips, is whether China will be able to get millions of chips in order to do this as well.
In one of his footnotes, Amodei expands on this: “To be clear, the goal here is not to deny China or any other authoritarian country the immense benefits in science, medicine, quality of life, etc that come from very powerful AI systems,” he said. “Everyone should be able to benefit from AI. The goal is to prevent them from gaining military dominance.”
To state the obvious here, it’s not just China that can direct “talent, capital, and focus to military applications of the technology.” OpenAI, arguably the leading AI company in the United States and the world, has already partnered with American military defense technology company Anduril to “deploy advanced artificial intelligence (AI) solutions for national security missions.” the US Military is already purchasing OpenAI software for war, and companies like Amazon, Google, and Microsoft are always competing for US military contracts. AI could have a lot of uses but the military is definitely one of them for US companies. That’s not something only China is doing.
Overall, Amodei piece is pretty diplomatic. It doesn’t vilify DeepSeek and Chinese researchers and respects their contributions to computer science. It acknowledges that societies deserve the benefits of technology even if we disagree with their governments. But the ultimatum Amodei says we are facing is: Do we want to live in a world in which an all powerful US owned AI is dominating the world or do we want to live in a world in which an all powerful China-owned AI is dominating the world.
If I had to choose, I guess I would choose the US AI dystopia over the Chinese AI dystopia. But those aren’t really the only choices available to us. Even if we just accept the assumption that AI will be as powerful as Amodei and other AI company CEOs tell us they are, are we really unable to even imagine a world in which we choose not to weaponize and militarize them in ways that brings humanity to the brink? Would preventing our own homegrown AI companies from doing exactly that not be a good place to start?
