Microsoft Develops AI Server Gear to Ease Nvidia Dependence

On Tuesday, insiders claimed that Microsoft is working on a new network card to reduce reliance on Nvidia artificial intelligence (AI) chips.

CEO Satya Nadella appointed Pradeep Sindhu, co-founder of AI networking platform Juniper Networks, to head the project. Moreover, Sindhu co-founded Fungible, a data center software and hardware developer that Microsoft acquired in January 2023.

The new network card will replace the Nvidia ConenctX-7 card, which OpenAI uses to train its large language models (LLMs). However, the tech giant estimates that the new chip would take over a year to achieve company-wide deployment.

If successful, the unnamed network card will boost the performance of Microsoft’s Azure Maia 100 AI server chip. Furthermore, it will significantly reduce the training time for LLMs while decreasing processing costs.

Analysts said the project illustrates how integral OpenAI is to the tech giant’s strategy regarding the generative AI (GenAI) arms race. The Windows maker had bet $13.00 billion on the ChatGPT creator, starting with a $1.00 billion investment in 2019.

Microsoft released Maia in November to help alleviate the shortage of AI chips and accelerate the development of GenAI products. As a result, the AI research organization deployed several ChatGPT improvements, including a memory function.

On February 15, OpenAI revealed Sora, the world’s first text-to-video solution, which generates a 60-minute video from a line of text.

Nvidia Sales to Drop as Others Mimic Microsoft

Industry watchers cautioned Nvidia that more clients will mirror Microsoft in self-developing AI chips and GPUs. Other tech companies conducting research and development of AI chips include Google, Amazon, Meta, Tesla, and OpenAI.

In addition, OpenAI CEO Sam Altman pitched a $7.00 trillion megaproject to expand the world’s AI chip production capacity. It involves the construction of a global supply line that includes dozens of large-scale AI factories and data centers.

Lastly, the Chinese government has expanded its support for domestic chipmakers amid tightening sanctions from the US. Nvidia supplies nearly 80.00% of high-end AI chips used by Chinese cloud operators, a figure that may drop to 50.00% in the next five years.