Microsoft is developing AI server gear to reduce NVIDIA dependence
Microsoft is working on a new network card to boost the performance of its Maia AI server chip, as reported by The Information. This will potentially reduce the company's dependence on chip designer NVIDIA. Led by Pradeep Sindhu, co-founder of Juniper Networks, the project follows Microsoft's acquisition of Sindhu's server chip start-up, Fungible, in 2023. If successful, this technology could accelerate OpenAI's model training on Microsoft servers, making the process more cost-effective.
A new network card similar to NVIDIA's ConnectX-7
The network card in development is said to resemble NVIDIA's ConnectX-7 card, which is sold alongside its graphic processor units (GPUs). The project is anticipated to take over a year to complete. Microsoft has invested billions in OpenAI, the creator of ChatGPT, and has integrated its technology into numerous products, giving the company an edge in the race to retail artificial intelligence software.
Microsoft Azure to get a boost
Successful development of the network card will not only benefit OpenAI. It could also boost the prowess of Microsoft Azure, especially when the power of Azure Boost (a system that offloads virtualization processes) is combined with it.
What does Maia chip do?
Microsoft unveiled the Maia chip in November 2023, designed for running large language models (LLMs) and supporting AI computing. The new network card's development could further solidify Microsoft's position in the AI market by enhancing its Maia AI server chip's performance. However, Microsoft has not yet commented on the matter. The company's ongoing investment in AI technology showcases its dedication to remaining at the forefront of this rapidly evolving field.
