Microsoft is preparing to deploy the latest version of its own-developed chips in data centers, with the goal of primarily relying on its own hardware to provide computational power support for its AI software.

Microsoft stated on Monday that it plans to deploy Maia chips with a total capacity of several gigawatts in its data centers, among which the latest generation of Maia 300 will begin deployment next year. These chips are specifically designed for AI inference, that is, to use AI models to perform tasks such as summarizing the content of teleconferences or answering research questions.

Microsoft, Amazon, and Google (a subsidiary of Alphabet) are all developing their own artificial intelligence accelerators, a move that may help reduce their dependence on NVIDIA's hardware. Amazon and Google started chip development earlier than Microsoft and are considered to be ahead of the world's largest software company in this field.

Microsoft bets that by customizing chips for its own data centers, it can run AI inference more efficiently, while also reducing some of the costs associated with purchasing NVIDIA's market-leading graphics processors (GPU). Currently, most of Microsoft's AI inference tasks are still powered by NVIDIA GPU.

Rani Borkar, an executive in charge of Microsoft's data center hardware business (including the chip design team), stated, "We are confident that we have created the best products for these workloads. With this confidence, we will begin deploying Maia at the gigawatt level starting next year."