A chatbot made by Chinese artificial intelligence startup DeepSeek has rocketed to the top of Apple’s App Store charts in the US this week, dethroning OpenAI’s ChatGPT as the most downloaded free app. The eponymous AI assistant is powered by DeepSeek’s open-source models, which the company says can be trained at a fraction of the cost using far fewer chips than the world’s leading models. Shares for Nvidia, the biggest global supplier of advanced AI chips, are currently down over 12 percent in pre-market training,
Downloads for the app exploded shortly after DeepSeek released its new R1 reasoning model on January 20th, which is designed for solving complex problems and reportedly performs as well as OpenAI’s o1 on certain benchmarks. R1 was built on the V3 LLM DeepSeek released in December, which the company claims is on par with GPT-4o and Anthropic’s Claude 3.5 Sonnet, and cost less than $6 million to develop. By contrast, OpenAI CEO Sam Altman has said GPT-4 cost over $100 million to train.
DeepSeek also claims to have needed only about 2,000 specialized chips from Nvidia to train V3, compared to the 16,000 or more required to train leading models, according to the New York Times. These unverified claims are leading developers and investors to question the compute-intensive approach favored by the world’s leading AI companies. And if true, it means that DeepSeek engineers had to get creative in the face of trade restrictions meant to ensure US domination of AI.
Nvidia, Microsoft, OpenAI, and Meta are investing billions into AI data centers — $500 billion alone for the Stargate Project, of which $100 billion is thought to be earmarked for Nvidia. Investors and analysts are now wondering if that’s money well spent, with Nvidia, Microsoft, and other companies with substantial stakes in maintaining the AI status quo all trending downward in pre-market trading.