Microsoft Will Scale Azure Maia System To Multi-Gigawatt Deployments In 2027

Microsoft has unveiled plans to expand its Azure Maia AI infrastructure to multi-gigawatt deployments by next calendar year. This strategic move positions Microsoft to compete directly with NVIDIA in the AI infrastructure space. Satya Nadella, Microsoft's CEO, announced this ambitious bet on the company's own silicon on X, stating that Maia will serve as the primary inference system for Microsoft's products and services.

The company has already deployed Maia 200 across the US, Europe, and Australia, achieving up to 1.7 times more tokens per watt compared to leading GPU-based systems for applications like Teams, Copilot, and coworking tools. Looking ahead, Microsoft aims to build on the efficiency gains of Maia 200 with the introduction of Maia 300, the third generation of the Azure Maia program.

Maia 300 is engineered with an inference-first architecture, optimized for complex AI workloads such as long-context, memory-intensive, and decode-heavy tasks. Key features of Maia 300 include a multi-chiplet architecture based on TSMC's N2P process node, improved memory bandwidth per watt, and scalable networking at a datacenter scale. These advancements aim to enhance data movement efficiency, network performance, and overall AI throughput across Microsoft's cloud infrastructure.