This post may contain links from our sponsors and affiliates, and Flywheel Publishing may receive compensation for actions taken through them.
Live coverage has ended. The full story is below.
Nvidia (NASDAQ:NVDA | NVDA Price Prediction) dropped 10% at the market open this morning, on news over the weekend about the release from China’s artificial intelligence lab DeepSeek of its R1 large language model (LLM).
The AI chipmaker was not alone as the tech sector trembled over the development of a low-cost, powerful, efficient AI model. Broadcom (NASDAQ:AVGO) tumbled 11%, Marvell Technology (NASDAQ:MRVL) fell 13%, and Advanced Micro Devices (NASDAQ:AMD) was down almost 6%.
Nvidia, though, was sharply hit because it could be one of the biggest losers. Using underpowered Nvidia H800 graphics processing units (GPUs), DeepSeek was able to develop hacks and workarounds to create an AI model that was superior in performance to OpenAI’s latest iteration of ChatGPT o1, while using far fewer processors and consuming less time.
Why is DeepSeek so worrisome?
R1 was able to be quickly trained using only 2,048 Nvidia GPUs at a cost of less than $6 million. Moreover, it comes only a month after DeepSeek had delivered V3, an LLM that took only two months to train and was found equal to OpenAI’s GPT-4o and Anthropic‘s Claude Sonnet 3.5. It was also able to outperform Meta Platforms (NASDAQ:META) Llama 3 as well as Alibaba’s (NYSE:BABA) Qwen2.5.
That was especially notable because it took Meta’s Lllama 3.1 the equivalent of 30.8 million GPU hours using 16,384 full-powered H100 GPUs. DeepSeek took the equivalent of about 2.8 million GPU hours. R1 is also reportedly 27 times cheaper than ChatGPT.
What’s it mean for NVDA stock?
Nvidia has been premium-priced for sustained chip sales and earnings growth. A bargain-basement workaround that is open-source to allow users to manipulate and fine-tune the algorithm jeopardizes the need for the premium Nvidia charges for its chips.
Certainly, the AI chipmaker’s business isn’t going to dry up. The recently announced $500 billion Stargate project, for example, will still be replete with Nvidia advanced Blackwell accelerators, but there is likely a strong headwind undercutting NVDA’s premium valuation.
Contact [email protected] for any questions or corrections.