Nvidia stock is down this week after DeepSeek's AI model announcement

Nvidia (NVDA) stock plummeted nearly 17% on Monday, marking its worst single-day decline since March 2020 and wiping out a record $589 billion in market value. The massive sell-off was triggered by growing concerns over Chinese AI startup DeepSeek's latest breakthrough, which threatens to upend assumptions about AI infrastructure spending.

DeepSeek's Efficient Innovation

DeepSeek, a Chinese artificial intelligence company, released a new AI model on January 20 that has caught the attention of industry veterans. Marc Andreessen, a prominent venture capitalist, called it "one of the most amazing and impressive breakthroughs I've ever seen." What's particularly notable is DeepSeek's claim that their latest AI model cost just $5.6 million to train - a fraction of the $100+ million reportedly spent on OpenAI's GPT-4 (and an even higher, unknown amount, for OpenAI's o1 and o3 models).

The announcement has raised questions about the necessity of extensive chip purchases that have driven Nvidia's valuation skyward over the past year. Raymond James semiconductor analyst Srini Pajjuri noted that if DeepSeek's innovations are widely adopted, "model training costs could come down significantly even at U.S. hyperscalers, potentially raising questions about the need for 1-million XPU/GPU clusters."

Ripple Effects Across Tech

The sell-off wasn't limited to Nvidia. Other semiconductor companies faced significant pressure, with Broadcom (AVGO) dropping over 17%, Micron (MU) falling nearly 12%, and Advanced Micro Devices (AMD) declining more than 6%. The tech-heavy Nasdaq Composite fell 3% as investors reassessed valuations across the AI sector.

Analysts Urge Calm

However, some analysts believe the market reaction may be overblown. Bernstein analyst Stacy Rasgon characterized the $5.6 million training cost claim as incomplete, noting it "does not include all the other costs associated with prior research and experiments on architectures, algorithms, or data."

A picture of Daniel Newman

Futurum's chief strategist Daniel Newman suggested the market might be misinterpreting the implications of more efficient AI models. Citing the Jevons Paradox, Newman argued that increased efficiency could actually expand AI usage: "If we can use compute more efficiently... companies will be able to build their models cheaper. They'll be able to create solutions with less overhead expense."

Regulatory Considerations

The development comes amid tightened U.S. export restrictions on AI chips to China, announced in the final days of President Biden's administration. These rules limit China's ability to purchase Nvidia's AI chips through resellers and access chips in remote data centers, potentially complicating DeepSeek's future development efforts.

For its part, Nvidia appeared unfazed by the DeepSeek announcement, calling it "an excellent AI advancement" in a statement. However, the dramatic market reaction highlights growing investor sensitivity to potential disruptions in the AI chip market, where Nvidia has maintained a commanding lead.

The sell-off marks a significant moment in Nvidia's recent history, eclipsing its previous record drop of $279 billion in September. As the market digests these developments, questions remain about the future of AI infrastructure spending and the sustainability of current valuations in the sector.