Nvidia calls DeepSeek R1 model ‘an excellent AI advancement’ despite losing $600 billion in market cap

Nvidia praised Chinese AI startup DeepSeek’s new R1 model on Monday, even as Nvidia faced its largest single-day loss in history, losing nearly $600 billion in market value. The launch of R1 triggered a wave of uncertainty among investors, raising questions about the sustainability of spending on AI chips and wiping out an estimated $1 trillion in U.S. tech market capitalization.

“DeepSeek is an excellent AI advancement and a perfect example of Test Time Scaling,” an Nvidia spokesperson told CNBC. “DeepSeek’s work illustrates how new models can be created using that technique, leveraging widely-available models and compute that is fully export control compliant.”

The R1 model, released last week, is an open-source reasoning system that reportedly surpasses the best efforts of U.S. firms like OpenAI. What’s catching the industry’s attention is its reported training cost of under $6 million—a fraction of the billions being spent by Silicon Valley giants on AI development, CNBC reported.

Nvidia appears to view DeepSeek’s innovation as a win for its GPU business. “Inference requires significant numbers of NVIDIA GPUs and high-performance networking,” the spokesperson added, highlighting how these models depend heavily on Nvidia’s technology.

DeepSeek’s use of Nvidia GPUs has not been without controversy. Scale AI CEO Alexandr Wang recently suggested that DeepSeek utilized GPUs banned in mainland China. Nvidia, however, countered this claim, stating that the GPUs used were export-compliant versions tailored for the Chinese market.

Industry Impact and AI Investment Concerns

DeepSeek’s cost-effective breakthrough has analysts questioning the massive investments in AI infrastructure by companies like Microsoft, Google, and Meta. Microsoft recently announced plans to spend $80 billion on AI infrastructure in 2025, while Meta CEO Mark Zuckerberg revealed an expected $60-65 billion in capital expenditures for the same year, focusing on AI.

“If model training costs prove to be significantly lower, we would expect a near-term cost benefit for industries like advertising, travel, and other consumer apps that use cloud AI services,” noted Justin Post, an analyst at BofA Securities. However, he cautioned that such advancements could ultimately reduce long-term revenue and costs tied to hyperscale AI infrastructure.

A Shift in AI Development

Nvidia’s remarks also reflect a broader shift in how AI advancements are being pursued. Much of the demand for GPUs has been driven by the “scaling law,” a concept introduced by OpenAI researchers in 2020. The idea posits that better AI models are achieved by significantly increasing computational power and data during training, creating a need for more GPUs.

Since November, industry leaders like Nvidia CEO Jensen Huang and OpenAI CEO Sam Altman have been discussing a new approach called “test-time scaling.” This method suggests that already-trained models can deliver better results by using additional computational power during inference to enhance reasoning capabilities.

DeepSeek’s R1 model exemplifies this principle, using test-time scaling to achieve performance levels that rival or surpass models developed with far greater investment. This approach, also present in some of OpenAI’s systems, highlights a potential path forward for the industry, where efficiency and cost-effectiveness play a larger role in AI innovation.

As the market processes these developments, the future of AI infrastructure investment may hinge on the balance between performance breakthroughs and the escalating costs of scaling computation.