In a bid to enhance the performance of its Maia AI server chip and potentially reduce reliance on NVIDIA, Microsoft is reportedly in the development phase of a new network card. Spearheaded by Pradeep Sindhu, co-founder of Juniper Networks and current CEO of Fungible (a server chip company acquired by Microsoft last year), the custom network card project aims to boost server performance and cut costs for the software giant.
Microsoft’s upcoming network card bears similarities to Nvidia’s ConnectX-7 card, commonly sold with its GPUs. The extensive development process for the network card is estimated to span over a year. If successful, the new chip will have the potential to significantly enhance the efficiency of Microsoft’s servers and reduce the cost associated with training OpenAI’s models.
The positive impact of the network card’s development extends to Microsoft Azure, particularly when integrated with the capabilities of Azure Boost, a system designed to offload virtualization processes.
Microsoft unveiled the Maia chip in November 2023, specifically designed for running large language models (LLMs) and supporting AI computing. The advancement of this new network card holds the promise of solidifying Microsoft’s standing in the competitive AI market by bolstering the performance of its Maia AI server chip. This proactive investment in AI technology highlights Microsoft’s commitment to maintaining a leadership position in this rapidly evolving field, showcasing a dedication to continuous innovation.
Read next: Equinix and NVIDIA form alliance powering AI supercomputing via private cloud services
