The release of the Chinese DeepSeek artificial intelligence (AI) model hit Nvidia (NASDAQ:NVDA) stock hard, knocking $600m off the market capitalisation.
And even expectations-busting fourth-quarter results on Wednesday (26 February) didn’t reverse the decline. At close the following day, Nvidia was down 22% from its all-time high.
Cracking quarter
Q4 revenue rose 78% to $39.3bn, with earnings per share (EPS) up 82%. Data centre revenue nearly doubled to $35.6bn.
Early shipments of Blackwell chips alone contributed $11bn, driven mainly by demand from major cloud service providers. The next-gen Blackwell Ultra is expected to be released in March.
Nvidia predicted first-quarter sales for 2025 of $43bn, plus or minus 2%.
In the company’s earnings call, CEO Jensen Huang said: “We’ve really only tapped consumer AI and search and some amount of consumer generative AI, advertising, recommenders … the next wave is coming.“
“Agentic AI for enterprise, physical AI for robotics, and sovereign AI … We can see great activity happening in all these different places,” he added.
DeepSeek threat
DeepSeek hit the headlines by using older and cheaper Nvidia technology as the new-generation Blackwell chips are restricted for sale to China. And it was, it’s claimed, trained up for just $6m.
Who needs to spend multiple billions on this stuff if Chinese developers can do it for so much less? Well, some are already casting doubt on those cost claims. And there’s a lot more to it than just cheap pre-training.
Missing the point
Huang suggests the market reaction to DeepSeek is wrong and investors are missing the point. The pre-training that comes before the release of an AI large-language model (LLM) is just the start, he says.
Maybe an AI Model can be pre-trained using cheaper chips. But continuing competition is going to be based on how well these things can learn and develop in time. And those using better chips will surely have a competitive advantaage.
Huang said: “Reasoning models that apply inference time scaling … can consume 100x more compute.“
Exports
Nvidia might see overseas sales restrained further by tightening chip exports to China. And that might help keep overseas competition from boosting Nvidia’s profits the way shareholders might like.
Competitor Alibaba has just revealed what’s described as its first reasoning AI model, QwQ-Max. And that, depending on the scale of US export limits, could further drive demand for top-end processors. These are clearly risks.
Chip competition
There’s one key threat a lot of investors fear. And it’s all about what the competition is doing. In particular, the rise of application-specific integrated circuits (ASICs) is causing ripples. Google‘s Gemini AI platform was trained using its own ASIC, for example.
And surely we can’t write off companies like Intel, Advanced Micro Devices, and the rest.
Would I buy Nvidia stock now? ‘Father of Value Investing’ Benjamin Graham pointed out that markets follow sentiment in the short term. But in the long term, they weigh up the fundamentals. Right now, I’d say we’re in the grip of sentiment. I’ll keep watching, with my eye on the fundamentals.
