Chief Executive Satya Nadella stated in a post on Monday that Maia is 40% more efficient than “mainstream GPU systems”.

Microsoft released a new generation of its own AI chips on Monday, named Maia 300, and announced that starting next year, it will deploy computing power on the scale of several gigawatts with this chip.

This move indicates that Microsoft aims to reduce its dependence on NVIDIA GPU. Currently, the chips used by Microsoft for running AI applications come predominantly from NVIDIA, and its own-developed chips are also expected to lower the operating costs of Copilot artificial intelligence software. CEO Satya Nadella stated in a blog post on Monday that the efficiency of Maia 300 is 40% higher than that of “mainstream GPU systems,” and Microsoft plans to rely primarily on this chip to power its own AI products.

Whether Microsoft can truly implement Maia chips on a large scale depends on the chip production capacity next year. As previously reported, Microsoft is in talks with TSMC, hoping to obtain a supply of over 300,000 chips. This quantity far exceeds Microsoft's previous total production of tens of thousands of custom chips; however, to support the deployment of several gigawatts of computing power, millions of chips will actually be required.

The Information also mentioned that Microsoft has been working to get large AI customers such as Anthropic to adopt its own chips, but so far, no major customer orders have been officially announced.