Microsoft wants to move away from NVIDIA, so it will spend more than it paid for Activision to get it, according to Windows Central

Have you ever wondered where all the questions you ask ChatGPT are processed? If you are not too immersed in the world of artificial intelligence, the answer might surprise you: it all happens on servers owned by Microsoft. This infrastructure, in turn, is primarily powered by the latest NVIDIA hardware, but we have reason to believe that things will change in the future.

Excessive dependencies are not recommended. In the business sector, they predispose scenarios of vulnerability, limited innovation and loss of control. One of the clearest examples that we can find of this in the technology industry is Apple's past dependence on Intel processors, something that Cupertino addressed from the M1 chip onwards.

Microsoft wants to reduce its dependence on NVIDIA

Regardless of the differences, part of Microsoft's present (and future) is tied to NVIDIA products. And, let's not fool ourselves, this has brought many benefits. The company run by Satya Nadella started building its relationship with OpenAI in a distant 2016 offering one of its most precious treasures, which was none other than its Microsoft Azure computing service.

Back then, the service was mainly supported by a huge number of GPUs NVIDIA Tesla K80 with Kepler architecture, although there was already talk of making a major update to the manufacturer's next GPUs based on its new Pascal architecture. OpenAI needed fast supercomputers to train its most advanced models, and Microsoft had them, thanks to NVIDIA hardware, of course.

Azure data centers reached another level when Microsoft made its first investment in OpenAI in 2019. put on the table 1 billion dollars The company promised to build a supercomputer with enormous power. The firm led by Sam Altman needed to train larger and larger models and, once again, Microsoft offered its solution.

This time, it acquired NVIDIA A100 GPU to make its cloud infrastructure more robust and powerful. This was one of the most important changes that Azure had undergone in recent years. In fact, Microsoft engineers had to innovate methods to make so many GPUs work together. But the AI ​​industry is moving fast, very fast.

That stupendous – and multimillion-dollar – investment by the Redmond giant soon began to be insufficient. The race to lead AI development required more changes. So Microsoft took out the portfolio two more times, in 2021 and in 2023, to fund OpenAIand again bought the most modern NVIDIA hardware, the NVIDIA H100, and then ordered the new NVIDIA H200.

As we can see, the link between OpenAI and Microsoft has been sustained based on the latter's powerful investments in combination with its wonderful computing infrastructure with an NVIDIA spirit. Both companies, however, are already working to reduce their dependence on the firm led by Jensen Huang, although their efforts will take time to bear first fruit.

According to Windows Central, Satya Nadella and Sam Altman's plan is tremendously ambitious. Specifically, it contemplates an investment of around 100 billion dollars in computing infrastructure for AI applications. We are talking about an investment greater than, for example, what Microsoft paid for Activision Blizzard. Let us remember that the operation was closed at 69,000 million dollars.

The plan, according to rumors, has several aspects. On the one hand, an AI supercomputer will be built for Azure known under the code name “Stargate” that would not have NVIDIA hardware. On the other hand, another somewhat smaller supercomputer would be prepared for the exclusive use of OpenAI. The first would be ready in 2028 while the second would start in 2026.

You may be wondering what alternatives to NVIDIA Microsoft has today. The answer is that the Redmond giant is also a hardware company, and for some time it has been dedicating budget to develop its own AI chips. At last year's Microsoft Ignite event we learned of the existence of two of their new products.

Azure Maia AI is a data center chip optimized for generative AI and general AI. Azure Cobalt, based on ARM, has been designed to power general-purpose computing. “We are reimagining every aspect of our data centers to meet the needs of our customers,” said Scott Guthrie, head of AI and cloud services, during the event.

Images | Microsoft

In Xataka | NVIDIA sells more chips for artificial intelligence than anyone else, although another company dominates this market

In Xataka | Companies from the US and China join hands for a good reason: to create the first global standard for generative AI