TLDR
- NVDA stock has fallen 23% from its January high, partly due to concerns raised by DeepSeek’s AI models
- CEO Jensen Huang dismissed concerns, claiming AI compute needs are “100 times more” than anticipated last year
- Nvidia has shipped 3.6 million Blackwell GPUs to major cloud providers in 2025, up 170% from 1.3 million Hopper GPUs in 2024
- Nvidia announced its next-generation Rubin GPU, set for release in late 2027 with 3.3x more computing power than Blackwell
- Nvidia plans to invest “hundreds of billions” in US-made chips over the next four years, potentially around half a trillion dollars total
Nvidia’s stock has experienced a significant drop of 23% from its record high in early January. The decline began when AI models developed by Chinese startup DeepSeek raised questions about whether companies were overspending on AI infrastructure, particularly Nvidia’s graphics processing units (GPUs).
These concerns stemmed from DeepSeek’s development of AI models that performed comparably to those from US companies like OpenAI and Anthropic. The key difference was that DeepSeek reportedly used fewer and less powerful Nvidia GPUs to train its models.
Nvidia CEO Jensen Huang addressed these concerns during his keynote speech at the company’s annual GTC developer conference this week. He dismissed the idea that more efficient model training techniques would reduce demand for AI accelerators.
“This last year, this is where almost the entire world got it wrong,” Huang stated during his presentation. He claimed that the computing power needed now for advanced AI features is “easily 100 times more” than anticipated at this time last year.
Rather than seeing a decline, Huang suggested that training efficiencies would actually increase demand. His reasoning is that lower costs would enable more companies across various industries to adopt AI technologies.
To support his argument, Huang pointed to GPU shipment figures among the four largest cloud service providers. Nvidia has shipped 3.6 million Blackwell GPUs to these companies in 2025.
This represents a substantial increase of over 170% compared to the 1.3 million Hopper GPUs shipped in 2024. These providers include industry giants Amazon, Alphabet, Microsoft, and Oracle.
Rubin, the next-generation GPU
At the conference, Huang also unveiled the company’s next-generation GPU. Named Rubin, this new architecture will succeed the current Blackwell GPU lineup.
The Rubin GPU is scheduled for launch in the second half of 2027. According to Nvidia, it will feature 3.3 times more computing power than its predecessor.
Beyond addressing current market concerns, Huang emphasized Nvidia’s position in “physical AI.” This refers to technologies enabling autonomous vehicles and robots to navigate and interact with the real world.
“We’ve been working on self-driving cars now for over a decade,” Huang noted. “We build technology that almost every self-driving car company uses.”
Major companies like Tesla and Alphabet’s Waymo already leverage Nvidia GPUs. Tesla uses them in data centers to train neural networks for self-driving software, while Waymo employs them both for training and in their actual robotaxis.
In separate news reported by the Financial Times, Nvidia plans to invest heavily in US-based chip production. The company expects to spend “hundreds of billions of dollars” on US-made chips and electronics over the next four years.
The total investment could reach around half a trillion dollars during this period.
“I think we can easily see ourselves manufacturing several hundred billion of it here in the US,” Huang told the Financial Times.
Huang added that the Trump administration could help accelerate the expansion of the US AI industry. He mentioned that Nvidia can now manufacture its latest systems domestically through suppliers including TSMC and Foxconn.
The CEO also acknowledged the growing competitive threat from Chinese telecoms firm Huawei. However, he viewed TSMC’s US investments positively, saying: “TSMC investing in the US provides for a substantial step up in our supply chain resilience.”
Earlier at the GTC conference, Huang told analysts that current orders for 3.6 million Blackwell chips from four major cloud firms underestimate total demand. These figures exclude Meta Platforms, smaller cloud providers, and various startups.
Wall Street remains optimistic
Wall Street remains optimistic about Nvidia’s prospects despite recent stock volatility. The median target price among analysts is $175 per share, suggesting potential upside of 52% from the current price of around $115.
Analysts estimate the company’s earnings will increase by 51% in fiscal 2026, which ends January 2026. At current levels, the stock trades at 39 times earnings with a price-to-earnings-to-growth (PEG) ratio of 0.76, which traditionally suggests the stock may be undervalued.
Stay Ahead of the Market with Benzinga Pro!
Want to trade like a pro? Benzinga Pro gives you the edge you need in today's fast-paced markets. Get real-time news, exclusive insights, and powerful tools trusted by professional traders:
- Breaking market-moving stories before they hit mainstream media
- Live audio squawk for hands-free market updates
- Advanced stock scanner to spot promising trades
- Expert trade ideas and on-demand support
