AI chip stocks saw a blood sell-off on Monday, along with other tech giants, after China's DeepSeek introduced its cost-effective AI reasoning R1 model. The model offers comparable capabilities to OpenAI's GPT-1 but requires less computing power, raising doubts about the billions of dollars tech giants have invested in AI chips

Nvidia and TSMC tumbled 11%, while Broadcom dropped 13%. Tech giants that have heavily invested in AI chips, such as Microsoft, saw a decline of over 6%, and Google and Tesla dropped 4%. 

DeepSeek-R1-Zero, a model developed by a Chinese startup, is trained using large-scale reinforcement learning (RL) without the need for supervised fine-tuning (SFT) as a preliminary step. It performs reasoning tasks at the same level as OpenAI's GPT-1. However, the R1 model runs queries at just $0.14 per million tokens, compared to OpenAI's $7.50, making it roughly 98% cheaper.

DeepSeek researchers wrote in a paper last month that their latest AI model, DeepSeek-V3, used Nvidia's H800 chips for training, spending less than $6 million. The H800 is a Chinese-only version of Nvidia's chip, launched in March 2023 as a replacement for the H100. Due to tightened export controls under the Biden administration, the H800 has significantly lower training efficiency and computing power compared to the H100. Moreover, the H800 is prohibited from being sold in China after late 2023.

Wall Street was surprised by DeepSeek's achievement, which demonstrates that with much less computing capacity, they have produced a model that is competitive, and in some areas, even superior. The model soon hit No. 1 on the Apple App Store's Top Free Apps chart. It is free and open-source, allowing researchers to study and build upon the algorithm. Published under an MIT license, the model can be freely reused without the need to show training data.

DeepSeek clearly doesn't have access to as much compute as U.S. hyperscalers, and yet they managed to develop a model that appears highly competitive, said Srini Pajjuri, an analyst at Raymond James, in a note on Monday.

Microsoft, Meta, Google, and Amazon spent a combined $125 billion on investing in and running AI data centers between January and August 2024, according to a JPMorgan report citing New Street Research. Meta CEO Zuckerberg has said the company plans to spend up to $65 billion this year to power its AI goals.

DeepSeek's performance will also intensify the U.S.-China AI competition. It suggests that even with inferior AI chips, China's AI sector is challenging U.S. dominance and raising doubts about the future of U.S. leadership in AI.