Microsoft is preparing to make its homegrown Maia chips the main system for running AI services such as Copilot while continuing to build Azure data centers around Nvidia hardware. The two commitments place Microsoft on both sides of the AI chip market, seeking to lower the cost of its own products without relinquishing the GPU infrastructure demanded by cloud customers.
The split became clearer on Monday, when Microsoft introduced Maia 300 and said it would scale the Maia processor family to multi-gigawatt deployments beginning in 2027. It described Maia as the future primary inference system across Microsoft products and services, a designation that concerns running models rather than necessarily training them.
Copilot already uses Maia hardware. Microsoft said Maia 200 is serving production workloads for Microsoft Copilot and its internal model team. GitHub Copilot and Microsoft Foundry are expected to follow by the end of 2026.
Sponsored · Selkirk Copper Mines
Microsoft’s separate investment in Nvidia is continuing. In January, Nvidia said Microsoft would deploy its Vera Rubin NVL72 systems at future AI data centers, including Fairwater sites. Microsoft subsequently described Azure infrastructure specifically engineered for Rubin’s cooling, networking, and power requirements.
At its July fiscal Q4 earnings call, Microsoft reiterated that it intended to be among the first cloud providers using Nvidia Vera Rubin and AMD Helios systems. CEO Satya Nadella said the company was modernizing its fleet with internally designed silicon alongside chips from both suppliers.
Azure revenue exceeded $100 billion in fiscal 2026, growing 41%, giving Microsoft a substantial cloud business to serve beyond its own software applications.
The collaboration extends outside data centers. On October 7, Microsoft detailed its Nvidia-powered Surface Laptop Ultra, built around the RTX Spark platform for local AI workloads. Five days later, it announced Maia 300’s expanded role inside its own services.
Microsoft says each Maia 300 rack will contain 96 accelerators and operate within a 167-kilowatt power envelope. It expects 40% more AI tokens per megawatt than leading GPU-based systems, without identifying those comparison systems.
Reuters reported that Microsoft expects to spend more than $175 billion in capital expenditures during calendar 2026. The tech giant has not disclosed how much of that spending supports Maia, how many Nvidia systems it may replace, or what the transition would save.
