Microsoft is reportedly developing a new network card that could improve the performance of its Maia chip, dedicated to artificial intelligence.
According to the sources of the website The Information, which reported the news, the move by the American giant would be due to the desire to reduce its dependence on Nvidia, the main manufacturer of GPUs in the world, i.e. graphics processing cards that also allow training artificial intelligence models.
It is no coincidence that in January last year, Microsoft hired Pradeep Sindhu, co-founder of Juniper Networks, a company that develops network equipment, to lead the project of creating internal hardware.
Furthermore, in 2023, Microsoft had acquired Fungible, a startup that deals with server chips, founded by Pradeep Sindhu himself.
According to The Information, Redmond's new network card would be similar to Nvidia's ConnectX-7, which the manufacturer sells alongside its graphics processing units.
The equipment could take more than a year to develop and, if successful, could reduce the time it takes OpenAI to train its models on Microsoft's servers, as well as make the process less expensive.
Microsoft has invested billions of dollars in OpenAI, parent company of Gpt Chat, on whose Gpt model Copilot's AI is also based.
Last November, Microsoft unveiled Maia, a chip designed to run large language models and support AI-based computing.
Reproduction reserved © Copyright ANSA