Microsoft is working on a new network card that could improve the functionality of its Maia AI server chip and lessen the company's dependency on Nvidia, a chip manufacturer.

Pradeep Sindhu, a co-founder of Juniper Networks, a company that develops networking hardware, has been appointed by Microsoft CEO Satya Nadella to lead the network card initiative. Last year, Microsoft purchased Fungible, Sindhu's server chip firm.

If the equipment is effective, it might reduce the time and cost required for OpenAI to train its models on Microsoft servers. However, the development of the equipment could take more than a year.

Microsoft has a competitive advantage in the artificial intelligence software market because to its billion-dollar investment in ChatGPT maker OpenAI and integration of its technology into a broad range of products.

In November, the company unveiled Maia, a chip designed to facilitate AI computing and run massive language models.

Named after a bright blue star, Microsoft's Maia 100 AI accelerator is intended to handle tasks related to cloud AI, such as large-scale language model training and inference. Some of the company's biggest AI workloads on Azure will be powered by it, including a portion of the multibillion-dollar collaboration with OpenAI, in which Microsoft powers all of OpenAI's workloads. The software behemoth and OpenAI have been working together on Maia's design and testing stages.

In order to maximize performance, power, and cost, Microsoft has completely redesigned its whole cloud server stack, including the new Azure Maia AI processor and Azure Cobalt CPU.

With 105 billion transistors, Maia is made using a 5-nanometer TSMC technology, which is about 30 percent fewer than the 153 billion transistors found on AMD's own Nvidia rival, the MI300X AI GPU.

Microsoft is a member of the group that is standardizing the upcoming data formats for artificial intelligence models, together with AMD, Arm, Intel, Meta, Nvidia, and Qualcomm. Microsoft is leveraging the Open Compute Project's (OCP) collaborative and open work to modify entire systems to meet the demands of artificial intelligence.