Nvidia(Nasdaq: NVDA) The shares are currently down 12% compared to its top of all time. He suffered a strong sale in January after the start-up based in China, Deepseek, said that she had formed a competitive model of artificial intelligence (AI) using a fraction of the computing power that had been deployed by leading American developers such as Openai.
Investors feared that Deepseek techniques will be adopted by other AI developers, leading to a substantial drop in demand for high -end graphic processing units (GPU), which are the best material available to develop IA models. However, these concerns could have been exaggerated.
Google Parent Alphabet(Nasdaq: Goog)(Nasdaq: Googl) is a big buyer of the chips of the NVIDIA AI Data Center, and on February 4, his CEO, Sundar Pichai, made comments that should make Nvidia investors feel much better.
Image source: NVIDIA.
Deepseek was created in 2023 by a successful Chinese hedge fund called High-Flyer, which has used AI to build commercial algorithms for years. Deepseek published his V3 Great language model (LLM) in December 2024, followed by its R1 reasoning model in January, and their competitiveness with some of the latest Openai models and other start-ups has buzzed the technological sector.
Since Deepseek’s work is open source, industry quickly learned of certain important details. The start-up claims to have trained the V3 for only $ 5.6 million (not to mention $ 500 million in flea and infrastructure, according to semianalysis), which is a drop in the bucket compared to the tens of billions of dollars spent by Companies like OpenAi to reach their current development stage.
Deepseek has also used older generations of Nvidia GPU as the H100, because the US government has prohibited the flea manufacturer from selling its latest equipment to Chinese companies (to protect the management of the American AI).
It turns out that Deepseek has implemented unique innovations on the software side to compensate for the lack of computing power. He has developed very effective algorithms and data input methods, and also used a technique called distillation, which consists in using the knowledge of an already successful AI model to form a smaller model.
In fact, Openai accused Deepseek of using his GPT-4O models to train Deepseek R1, inviting the Chatgpt chatbpt on a large scale to “learn” from his outings. Distillation quickly accelerates the training process because the developer does not have to collect or process data mountains. Consequently, it also requires much less computing power, which means less GPU.
Naturally, investors fear that if all other AI developers have adopted this approach, this would trigger a collapse of the demand for Nvidia fleas.
On February 26, NVIDIA reported its financial results for its 2025 financial year, which ended on January 31. The company plans to have generated $ 128.6 billion in total income, which would represent an increase of 112% compared to the previous year. The recent quarterly results suggest that around 88% of these income will be attributable to its data center segment, thanks to the rowing of GPU sales.
According to consensual forecasts for Wall Street (provided by Yahoo), Nvidia could establish another record for its 2026 in progress, with $ 196 billion in potentially in the cards. Hitting this estimate will depend more GPU request for AI developersIt is therefore easy to understand why investors are nervous about Deepseek news.
While the H100 remains a hot product, the latest GB 200 GPU of Nvidia – which is based on its Blackwell architecture – can make an AI inference up to 30 times speed. Inference is the process by which an AI model absorbs live data (such as a chatbot prompt) and produces an output for the user. It usually comes after the initial training phase (more about this in a moment).
The GB200 is currently the OR stallion for AI data centers, and demand considerably exceeded the supply when it started to ship customers at the end of 2024.
Image source: alphabet.
Pichai kept a conference call with Wall Street analysts on February 4 to discuss the results of the fourth quarter of alphabet in 2024. In response to one of their questions, he said that there had been a notable change In the allocation of calculation power in the past three years, with an increasing amount towards inference compared to the training.
Pichai said that new reasoning models (such as Deepseek R1 models and alphabet flash thinking) will only accelerate this change. These models spend more time “reflect” before producing an answer, they therefore require much more computing power than their predecessors. The technical term for this is the testing of the time of testing, and it is a means for the models of AI to provide more precise information without performing a more pre-tiral scale (which involves to supply infinite models of new data).
Meta-platforms CEO Mark Zuckerberg has similar thoughts. He recently declared that a drop in training charges does not necessarily mean that developers need fewer chips, because the capacity simply moves to inference instead.
Finally, Alphabet told Wall Street that he planned to allocate $ 75 billion to capital expenses (CAPEX) in 2025, most of which will go to infrastructure and chips in the data center. This figure represents a significant increase compared to its CAPE 2024 of $ 52 billion, so the company certainly does not decrease.
Overall, it seems that the image of the demand for NVIDIA GPUs is still very intact. Since its actions are negotiated with an attractive assessment at the moment, the recent decline could even be an opportunity to purchase.
Before buying actions in Nvidia, consider this:
THE Motley Fool Stock Advisor The team of analysts has just identified what they believe 10 Best Actions So that investors are buying now … and Nvidia was not part of it. The 10 actions that cut could produce monster yields in the coming years.
Inquire Nvidia Make this list on April 15, 2005 … if you have invested $ 1,000 at the time of our recommendation, You would have $ 850,946! *
Now it’s worth notingStock advisorTotal average yield is959% – an outperformance that has made the market compared to178%For the S&P 500. Do not miss the last list of the best 10.
Randi Zuckerberg, former Director of Development of the Facebook and Sister of the CEO of Meta Platforms, Mark Zuckerberg, is a member of the board of directors of Motley Fool’s. Suzanne Frey, director of Alphabet, is a member of the board of directors of Motley Fool’s. Anthony Di Pizio Has no position in the actions mentioned. The Motley Fool has positions and recommends alphabet, meta and nvidia platforms. The Word’s madman has a Disclosure policy.
