Microsoft plans to shift most of the artificial intelligence workloads powering its products, including Copilot and Teams, to its own Maia chips starting next year, a move analysts see as a potential blow to Nvidia.
Microsoft CEO Satya Nadella announced that the company will expand Maia chip deployments to multi-gigawatt capacity beginning in 2027, making them the primary system for running AI models across Microsoft's products and services. He described the initiative as the company's biggest bet yet on its own silicon.
Microsoft also introduced Maia 300, the third generation of its custom AI chip. According to the company, it delivers 40% more tokens per watt than leading GPU-based systems. Production deployment is scheduled to begin in 2027.
The previous generation, Maia 200, is already operating in the United States, Europe, and Australia. According to Microsoft, it delivers up to 1.7 times more tokens per watt in Teams and Copilot.
Maia 200 currently runs Copilot, OpenAI models, and Microsoft's own MAI models, with GitHub Copilot and the Microsoft Foundry platform expected to join by the end of 2026.
Microsoft's chips are designed for AI inference, which refers to the everyday operation of AI models when responding to user requests, rather than the process of training those models.
However, Microsoft is not completely abandoning other chip manufacturers. The company will continue using Nvidia and AMD infrastructure alongside its own chips in its data centers.
Analysts believe that if Maia proves cost-effective, Microsoft could redirect some of its future investments away from Nvidia, potentially weakening the chipmaker's pricing power over time.
Microsoft is one of Nvidia's largest chip customers, while other technology giants, including Google and Amazon, are also developing their own custom AI chips.
Discussion(0)