DeepSeek’s R1 reasoning model, developed for under US$6 million, has sent shockwaves through the AI industry, triggering a significant 12% drop in Nvidia’s stock during pre-market trading.

DeepSeek trained its R1 model using just 2,000 Nvidia chips, while major AI companies like Microsoft, OpenAI, and Meta typically use tens of thousands for their large-scale AI models.

This stark difference in resource usage highlights DeepSeek’s efficiency and cost-effectiveness, further fueling questions about the future of AI infrastructure investments.

For instance, an estimated US$100 billion of the US$500 billion budget allocated to the Stargate Project, a major initiative by leading AI companies, is reportedly dedicated to purchasing Nvidia chips.