Microsoft plans to reduce its reliance on Nvidia and AMD by making its own Maia chips the primary hardware for running AI models across its services.
Its newest processor, the Maia 300, is expected to enter production in 2027, promising more processing capacity while using less electricity.
The move comes as major cloud providers invest in custom hardware to manage growing AI demand and gain more control over infrastructure costs.
Why It Matters: Microsoft’s investment in custom chips could influence the cost, performance and availability of enterprise AI services. Greater control over its hardware may help the company expand capacity while managing operating expenses, with potential implications for Azure and Microsoft 365 Copilot customers. It also adds another factor to cloud infrastructure planning as providers develop different approaches to supporting AI workloads.
- Microsoft Plans to Expand Its Use of Maia: The company wants its own processors to handle most AI inference workloads, which involve running trained models to generate responses. Maia 200 already supports Microsoft 365 Copilot, OpenAI models and other production workloads. Microsoft eventually plans to deploy multiple gigawatts of computing capacity across its Maia chip family, although it has not provided a timeline for reaching that goal.
- Maia 300 Promises Greater Power Efficiency: According to Microsoft, the new processor can handle 40% more AI tokens per megawatt than leading GPU-based systems. Each rack will contain 96 chips, twice the number used with Maia 200, and deliver up to 7 times the token throughput of its predecessor. With electricity availability limiting data center expansion, these improvements could help Microsoft support more AI workloads without a comparable increase in power consumption.
- Custom Chips Could Help Reduce Operating Costs: CFO Amy Hood credited Microsoft’s use of internally developed processors with contributing to margin improvements over the past decade, while CEO Satya Nadella said Maia 200 delivers 30% better performance per dollar than the latest hardware in its existing fleet.
- Nvidia and AMD Remain Important Suppliers: Microsoft will continue using third-party processors, particularly for training AI models. Different workloads require different hardware, and the company wants to maintain flexibility in choosing what works best. Evercore ISI analysts estimate Maia could account for roughly 10% of Microsoft’s AI chip spending by the first quarter of 2027, suggesting Nvidia and AMD will remain important suppliers even as Microsoft expands its own hardware.
- Cloud Providers Are Taking Different Approaches: Google and Amazon have also developed custom AI chips, but unlike Microsoft, they allow customers to rent access to certain processors through their cloud platforms. Maia remains reserved for Microsoft’s internal operations, meaning Azure customers cannot select it directly. For enterprises comparing cloud services, those differences could influence workload compatibility, operating costs and how much control customers have over the hardware running their AI applications.
Go Deeper -> Microsoft to rely more on homegrown AI chips, reducing dependence on Nvidia – CNBC
Microsoft reveals Maia 300 AI chip as it works to reduce dependence on Nvidia, AMD – Yahoo!Finance
