Tech giant plans multi-gigawatt deployments as in-house silicon becomes its primary inference system.
Microsoft has announced plans to scale its Azure Maia family of custom artificial intelligence chips to multi-gigawatt deployments starting in 2027, establishing Maia as the primary inference system across its product ecosystem.
The multigenerational Azure Maia program forms a foundational pillar of Microsoft’s long-term AI infrastructure strategy. Currently, the Maia 200 processor is powering production workloads including Microsoft Copilot, alongside supporting the development of next-generation internal models. Deployments for GitHub Copilot and Microsoft Foundry are scheduled to follow by the end of the year.
Maia 200 is being rolled out across data centres in the United States, Europe, and Australia. It runs models from OpenAI, Microsoft’s proprietary MAI team, and leading open-source families, delivering up to 1.7 times more tokens per watt for integrated AI features within Microsoft Teams, Copilot, and Cowork compared to leading GPU infrastructure.
Looking ahead to 2027, Microsoft will introduce the Maia 300, a complete intelligent system engineered for advanced workloads such as long-context reasoning, latency-sensitive conversational interfaces, and sustained agentic tasks. Built using TSMC’s N2P process node on a chiplet architecture, Maia 300 features an inference-first design that balances compute power, memory bandwidth, and power efficiency.
Microsoft projects that a fully configured 96-accelerator Maia 300 setup will deliver up to seven times the rack-scale token throughput of Maia 200, achieving 40 per cent more tokens per megawatt than leading GPU-based systems.
