menu_open Columnists
We use cookies to provide some features and experiences in QOSHE

More information  .  Close

DeepSeek Doesn’t Signal an AI Space Race

2 1
29.01.2025

Welcome to Foreign Policy’s China Brief.

The highlights this week: Chinese AI start-up DeepSeek disrupts U.S. markets with a new large language model, U.S. President Donald Trump threatens major tariffs on Taiwanese semiconductors, and senior doctors in China raise the alarm about the efficacy of Chinese generic drugs.

Welcome to Foreign Policy’s China Brief.

The highlights this week: Chinese AI start-up DeepSeek disrupts U.S. markets with a new large language model, U.S. President Donald Trump threatens major tariffs on Taiwanese semiconductors, and senior doctors in China raise the alarm about the efficacy of Chinese generic drugs.

By submitting your email, you agree to the Privacy Policy and Terms of Use and to receive email correspondence from us. You may opt out at any time.

✓ Signed Up

On Monday, the news of a powerful large language model created by Chinese artificial intelligence firm DeepSeek wiped $1 trillion off the U.S. Nasdaq 100 index in a single day, reversing weeks of gains in a heated market driven by belief in an AI-dominated future.

Chipmaker Nvidia was the worst off, losing nearly $600 billion in value on Monday. DeepSeek’s model was reportedly trained on Nvidia’s cheaper, older chips and not its cutting-edge products, which are sanctioned in China. Chinese stock markets are closed for Lunar New Year but will likely see a rally upon reopening this week—though DeepSeek isn’t publicly traded.

Some of the U.S. media discussion around DeepSeek is overblown, such as the claim that its AI model only cost $5.5 million to develop. As DeepSeek’s own statements make clear, that was the cost of the model’s final training run—not including the research, equipment, salaries, and other costs involved.

The rush by analysts to declare that chip sanctions aren’t working is also misplaced. DeepSeek was trained on Nvidia’s H800 chips, which, as a savvy ChinaTalk article points out, were designed to evade the U.S. chip sanctions put in place in October 2022. The newest round of U.S. sanctions only kicked in at the end of 2023—too late to affect DeepSeek’s model.

Nevertheless, U.S. officials and AI analysts will likely use DeepSeek to justify expanding sanctions, with Nvidia’s H200—which is very popular with Chinese buyers—a likely target.

Of course, DeepSeek operates........

© Foreign Policy


Get it on Google Play