Last week, like semiconductor stocks Nafidia (NASDAQ: NVDA), Advanced small devices (Nasdaq: AMD), and Micron technology (NASDAQ: MU) caused the news that a Chinese emerging company called Deepseek has discovered how to train artificial intelligence models (AI) for a small part of the cost of its American peers.
Investors were concerned that the innovative Depceik approach would lead to a breakdown in the demand for graphics processors (GPUS) and other data center components, which are the key to developing artificial intelligence. However, these fears may be exaggerated.
Definition platforms (Nasdaq: Meta) is a huge buyer of artificial intelligence chips from NVIDIA and AMD. On January 29, CEO Mark Zuckerberg gave a series of comments that should be music for investor ears who have the shares of artificial intelligence devices.
Photo source: Getty Images.
A successful Chinese hedge box has been used Amnesty International To build trading algorithms for years. Dibsic was established as a separate entity in 2023 to take advantage of the success of other artificial intelligence research companies, which were rapidly rising in value.
The stock market panic was running last week through the DEPSEK Grand Language Model (LLM), which matches the performance of the latest GPT-4O models from Airame AIRANT AI, Openai, through several criteria. This is not a source of concern for the nominal value, except for Deepseek that he spent $ 5.6 million only V3 training, while Openai has burned more than $ 20 one billion Since 2015 to reach its current stage.
To make things more important, Deepseek cannot reach the latest graphics processing units at the NVIDIA data center, because the United States government prevented them from selling them to Chinese companies. This means that young people had to use older generations such as H100 and H800, which suffers from a lack of strength, which indicates that it is It is possible to train artificial intelligence models Without the best devices.
To compensate for the lack of arithmetic performance, Deepseek on the side of the program by developing more efficient algorithms and data entry methods. In addition, it has adopted a technique called distillation, which includes the use of a successful model to train smaller models. This speeds up in the training process and requires a much lower computing capacity.
Investors are concerned that if other artificial intelligence companies adopt a Deepseek approach, they will not need to buy the largest number of graphics processing units from NVIDIA or AMD. This would crush the demand for the leading memory center in the industry in Micron.
NVIDIA graphics units are the most popular in the world to develop artificial intelligence models. The fiscal year of the company has just ended 2025 on January 31, and according to the directives of the administration, its revenues are likely to be more than twice to 128.6 billion dollars (official results will be issued on February 26). If the last quarters are anything you must go through, about 88 % of these revenues will come from their data center thanks to GPU sales.
This amazing growth is the reason why NVIDIA has added $ 2.5 trillion to the market value over the past two years. If the demand for chips slows down, it is likely that a lot of this value will evaporate.
AMD has become a competitor worth Nvidia in the data center. The company plans to launch the new GPU Mi350 later this year, which is expected to compete with the latest Blackweell chips in NVIDIA, which has become the golden standard for tackling the burdens of artificial intelligence work.
But AMD is also a leading resource for artificial intelligence chips for personal computers, which may become a major growth sector in the future. Since LLMS becomes cheaper and more efficient, it will eventually be able to run on smaller chips inside computers and devices, which reduces dependence on external data centers.
Finally, Micro is often overlooked as Amnesty International Frame, but it plays an important role in this industry. HBM3E (high -frequency domain memory) for the best data center in the separation is when it comes to capacity and energy efficiency, which is why NVIDIA uses it within the latest BlackWell graphics units. The memory stores the information in a ready condition, which allows GPU to receive it immediately when needed, and since the burdens of artificial intelligence work are very intense for data, it is an important part of the device’s puzzle.
Photo source: Getty Images.
The Meta platforms spent $ 39.2 billion on the infrastructure and data infrastructure during 2024, and plan to spend up to $ 65 billion this year. These investments help the company to increase the progress of Llama Llms, which are the most popular models in the world, with 600 million downloads. Llama 4 is scheduled to be launched this year, and CEO Mark Zuckerberg believes that he can be the most advanced in this industry, and outperforms the best closed source models.
On January 29, Meta made a phone call with analysts about the fourth quarter of 2024. When Zuckerberg was interrogated about the possible effect of Deepseek, he said it was too early to determine what capital investments in chips and data centers. However, he said even if it leads to less capacity requirements for the training burdens of artificial intelligence, this does not mean that companies will need fewer chips.
Instead, it is believed that the capacity can be transformed away from training and inference, a process in which artificial intelligence models deal with inputs of users and models responses. Many developers are transmitted from training models using countless quantities of data, and focus on “thinking” abilities instead. This is indicated as the scaling of the test time, and the model that takes an additional time for “thinking” includes before providing the output, which leads to high -quality responses.
Thinking requires more inference account, so Zuckerberg believes that companies will still need the best infrastructure for the data center to maintain a competition. In addition, most of the artificial intelligence programs have not yet achieved a major adoption yet, and Zuckerberg admits that the service of many users will also require an additional data center capacity over time.
So, although it is difficult to place accurate numbers on how to rebuild Deepseek innovations, Zuckerberg comments indicate that there is no reason for Ashdawia, AMD and Micron Stock Destors. In fact, there is even a bullish issue for these long -term stocks.
Have you ever felt that you missed the boat in buying the most successful stocks? Then you will want to hear this.
On rare occasions, our expert team issues an analyst A. “Double Permanent” stock A recommendation for companies they believe is about to pop. If you are worried that you have already missed your chance to invest, it is now the best time to buy before it is too late. And the numbers speak for themselves:
Nafidia: If you invest $ 1,000 when we doubled in 2009, You will have $ 307661!
apple: If you invest $ 1,000 when we doubled in 2008, You will have 44,088 dollars!
Netflix: If you invest $ 1,000 when we doubled in 2004, You will have 536,525 dollars!
Currently, we issue “Double Download” alerts for three incredible companies, and there may not be another chance like this any time soon.
*The stock consultant dates back from February 3, 2025
Randy Zuckerberg, former Director of Market Development and Speak for Facebook and Sister to Meta Platforms, Mark Zuckerberg, member of Motley Fool Board of Directors. Anthony de Bizio He has no position in any of the mentioned stocks. Motley Fool has positions in advanced advanced devices, identification platforms, and NVIDIA. Motley deception has Disclosure.
