A week after banning TikTok from app stores (no, it still isn’t downloadable) and one day after threatening 25% tariffs on Colombia, the United States took a major blow in its ongoing trade wars—overnight, Silicon Valley seemed to have lost its AI dominance.
The shakeup follows a release from fresh-faced Chinese AI startup DeepSeek, which on Jan. 20 updated its ChatGPT-like AI assistant with its open-source R1 reasoning model. According to DeepSeek’s testing, the R1 model matches OpenAI’s o1 reasoning model on several metrics, all while being much, much cheaper to develop.
The Wall Street Journal was the first to report on DeepSeek R1’s ultra-low development cost, citing the one-year-old company’s claims that it only took $5.6 million to develop the new model, vs. over $100 million from OpenAI for its equivalent.
It took a little bit of time for the news to get out there, but DeepSeek consequently rose to the top of the App Store, unseating ChatGPT as the most-downloaded free app. The sudden surge in attention was hard on U.S. stocks on Monday, sending the Dow down by about 0.22%, the S&P down by 2%, and the Nasdaq down by 3.6%. More specifically, Google parent company Alphabet was down 2.89%, with Meta, Oracle, and other tech giants also seeing significant declines. Notably, data center and graphics card company Nvidia, which supplies much of the hardware powering AI development, fell 11.64%.
Nvidia’s stock drop in particular likely had to do with claims from DeepSeek that it only needed roughly 2,000 specialized Nvidia chips to train its latest AI model, whereas leading U.S. models tend to use closer to 16,000 chips. Said claims are still awaiting verification, but if true, would poke holes in the U.S.’ recent policy efforts to restrict the amount of American chips Chinese developers can use.
On the consumer side of things, DeepSeek promises cheaper access to higher-tier models than ChatGPT, which puts basic access to its o1 model behind a $20/month ChatGPT Plus subscription and unlimited access to the o1 model behind its pricey $200/month ChatGPT Pro plan. For what it's worth, frequent OpenAI collaborator Microsoft has since integrated the o1 model into the free tier of Copilot, although it appears to still rolling out.
Unfortunately, the company seems to be suffering from success right now—servers appear to be overloaded, and I’m currently not able to sign up for an account for testing. Presumably, as more people get through and get their hands on these models, it'll be easier to verify just how scared of DeepSeek U.S. companies should be. However, when my colleague Jake Peterson was able to get his account up and running, he noticed several security loopholes, including chat logs that were left exposed online.
Still, the competition could prove to be a shot across the bow for U.S. AI developers, who, alongside President Trump, just announced the $500 billion “Stargate Project,” an initiative to build out U.S. AI infrastructure starting with a $100 billion plan to build out data centers in Texas.
Update 1/30/25 at 8:00 pm: Added further reporting from Jake Peterson, as well as acknowledged o1 compatibility with Copilot.
