Maia 300: How Microsoft Is Cutting Its Dependence on Nvidia GPUs With Its Own AI Chip

Microsoft

Unveils Maia 300, Its Third In-House AI Accelerator

Microsoft has officially announced Maia 300, the third generation of its in-house artificial intelligence accelerator. The new chip is positioned as the primary AI inference system for Microsoft's product ecosystem and cloud services. Maia 300 doesn't just cut costs and boost efficiency for AI workloads — it also handles modern reasoning models and agentic tasks better.

The chip will officially enter production workloads starting in 2027. Microsoft designed it with maximum energy efficiency in mind: Maia 300 can deliver a 40% increase in output tokens per megawatt compared to leading GPU systems. In a full configuration of 96 accelerators, the chip achieves a 7x increase in rack-scale token throughput over its predecessor, Maia 200.

New Architecture and Process Node

Microsoft built Maia 300 on an innovative chiplet architecture manufactured on TSMC's N2P process. The key feature is splitting functions across three types of components — compute blocks, a memory controller, and I/O functions. This approach increases the share of silicon dedicated directly to computation and lets each functional block use the technology best suited to its specific role.

Where Maia 200 Is Already Running, and What's Next

Maia 200, the previous generation, is already deployed in production and delivers strong results across real-world AI service metrics. It currently powers:

• Microsoft Copilot • GitHub Copilot • Microsoft Foundry • Microsoft's own AI models

Maia 300 was built specifically for new types of AI workloads: long-context reasoning, latency-sensitive conversational experiences, and sustained multi-step agentic tasks.

Microsoft plans to scale the Maia family to multi-gigawatt deployments starting in 2027, which will make Maia the core AI infrastructure for all of Microsoft's products and cloud services. This move will reduce the company's dependence on graphics processors from Nvidia and AMD.

Integration With Microsoft's Own AI Models

Maia 300 is a critical piece of Microsoft's strategy for developing its own AI models to compete with offerings from Google DeepMind, OpenAI, and Anthropic. According to Mustafa Suleyman, head of Microsoft AI, the company needs control over the full stack — from algorithms down to silicon. "At this scale, we need a platform that we own and co-design alongside our models. That's why launching Microsoft's own AI models on the Maia chip is a critical priority for our team," Suleyman explained.