Microsoft Unveils Azure Maia 300 and Plans Multi-Gigawatt Rollout Starting in 2027

Microsoft Bets on Maia for Inference at Scale

Microsoft is making a major commitment to its in-house AI silicon. The company says it will scale its Azure Maia family to multi-gigawatt deployments starting in 2027. That expansion is meant to make Maia the primary inference system behind Microsoft's products and services.

The move signals that Microsoft sees its custom hardware as central to how it will serve AI workloads across its portfolio in the years ahead.

Maia 200 Already in Production

Microsoft's current-generation chip is already doing real work. Maia 200 is now serving Microsoft production workloads, including Microsoft Copilot, and the company says more services are on the way.

• Current workloads: Microsoft Copilot is already running on Maia 200.
• Coming next: GitHub Copilot and Microsoft Foundry are slated to follow by the end of the year.
• Model support: Maia 200 runs OpenAI models, Microsoft's own MAI models and leading open-source model families.
• Geographic footprint: Deployments are underway across the United States, Europe and Australia.

What Maia 300 Is Built For

Microsoft describes Maia 300 as a complete intelligent system rather than just a chip. It is aimed at the kinds of AI workloads becoming more common in modern products:

• Long-context reasoning, where models work across large amounts of input.
• Latency-sensitive conversational experiences, where responsiveness matters to users.
• Multi-step agentic tasks, where AI systems carry out sequences of actions.

Performance and Efficiency Claims

Microsoft is making ambitious efficiency and throughput claims for its next-generation hardware. The company expects Maia 300 to deliver 40% more tokens per megawatt than leading GPU-based systems, a figure that matters as power becomes a key limit on data center growth.

Microsoft also claims a full 96-accelerator Maia 300 configuration will deliver up to seven times Maia 200's rack-scale token throughput.

These numbers are Microsoft's own projections, so independent benchmarks will be needed to see how Maia 300 performs in practice.

Under the Hood: Architecture and Networking

Further technical details about Maia 300 come from a single report and have not been widely confirmed. According to that report:

• Chiplet design: Maia 300 uses a multi-chiplet architecture with TSMC N2P compute chiplets.
• Networking: The system uses an Ethernet-based custom AI transport, with electrical connections over short distances and optical links between racks.
• Scale: Superpods can reach up to 128 racks and 12,288 accelerators.

If accurate, these details point to a system designed from the start for very large deployments, in line with Microsoft's multi-gigawatt ambitions.