In January, Chinese start-up DeepSeek introduced an artificial intelligence (AI) chatbot that quickly became the most downloaded free mobile application in the U.S. Importantly, the base large language model behind the DeepSeek application allegedly cost $6 million to train and reportedly outperforms top U.S. models on certain benchmarks.
That shook Wall Street because OpenAI spent over $100 million training its GPT-4 model and supposedly had access to more advanced chips than DeepSeek. Investors rushed to the conclusion that U.S. companies have been overspending on AI infrastructure, and Nvidia (NVDA +2.71%) shares fell 17% on the news, erasing $600 billion in market value. That is the largest single-day loss in U.S. history.
Nvidia shares are still down by about 9% since DeepSeek rattled the market, and the stock is currently 13% off its all-time high. But Nvidia shareholders just got some good news from Amazon (AMZN +0.20%) and Alphabet (GOOGL +2.81%). Read on to learn more.
NASDAQ: NVDA
Key Data Points
Some Wall Street analysts are skeptical about the DeepSeek narrative
Many analysts have praised DeepSeek for its engineering breakthroughs. The company reduced expenses by cutting down on data processing with innovative training techniques. However, several analysts also question whether DeepSeek has been completely forthcoming about its infrastructure and the associated costs.
For instance, Dan Ives at Wedbush Securities wrote, "Saying DeepSeek was built for $6 million with no Nvidia next generation hardware is likely a fictional story." Likewise, research company SemiAnalysis reports that DeepSeek not only had next-generation Nvidia GPUs but also incurred about $1.6 billion in expenses while training its model.
Importantly, cost efficiencies could actually boost demand for Nvidia graphics processing units (GPUs) by enabling faster diffusion of artificial intelligence (AI) through the economy. Put differently, if novel training methods reduce model development costs, more software companies will build AI products, increasing the total demand for Nvidia GPUs.
Image source: Getty Images.
Amazon and Google plan to invest even more aggressively in AI infrastructure
Last week, Amazon and Google parent Alphabet reported fourth-quarter financial results. Both companies said capital expenditures would increase substantially in 2025, as they are currently supply-constrained where AI infrastructure is concerned.
Amazon CEO Andy Jassy predicted costs associated with developing AI models will continue to fall. But he also said, "I think it will make it much easier for companies to be able to infuse all their applications with inference and with generative AI." And CFO Brian Olsavsky told analysts that capital spend may exceed $100 billion in 2025, up from $83 billion in 2024, due to demand for AI infrastructure.
Alphabet CEO Sundar Pichai provided similar insights. He said a greater percentage of capital expenses are shifting toward inference as training becomes less expensive, implying DeepSeek's breakthroughs would not slow AI spending. Indeed, CFO Ana Ashkenazi told analysts that capital expenditures would total $75 billion in 2025, up from $52 billion in 2024, due primarily to investments in data centers, servers, and networking.
The capital spending guidance from Amazon and Google parent Alphabet corroborates the narrative that demand for Nvidia GPUs could actually increase as training costs decline because companies will shift their resources toward inference. Andy Jassy compared the situation to how cloud computing sales have actually increased over time despite cloud services becoming less expensive.
"What happens is companies will spend a lot less per unit of infrastructure, and that is very, very useful for their businesses. But then they get excited about what else they could build that was always cost-prohibitive before, and they usually end up spending a lot more in total on technology once you make the per unit cost less. I think that is very much what's going to happen here in AI."
Wall Street still expects Nvidia's earnings to grow very quickly
Wall Street estimates Nvidia's adjusted earnings will grow at 52% annually through fiscal 2026, which ends in January 2026. That consensus makes the current valuation of 50 times adjusted earnings look cheap. Admittedly, analysts have set a very high bar, so the stock could fall sharply if Nvidia misses those expectations.
Here is the bottom line: Nvidia shareholders worried about the DeepSeek drama can take a deep breath. Many analysts think more efficient AI training techniques will lead to greater demand for Nvidia GPUs. And evidence from the largest hyperscale cloud companies supports that conclusion.
