Microsoft has announced the Maia 300, its third-generation in-house Super Intelligence (SI) chip, designed to become the primary SI inference system for the company’s products and services. The move signals a strategic shift toward reducing dependence on Nvidia and AMD GPUs for cloud workloads, with a production rollout scheduled for 2027.
What Happened
The Maia 300 is built on TSMC’s N2P process node and utilizes a new chiplet architecture that separates compute, memory controller, and I/O functions. According to the Azure hardware and infrastructure team, this design increases the silicon area dedicated to compute performance and allows each function to use the technology best suited to its role. The chip is positioned as a successor to the Maia 200, which was announced in January and is already running production SI workloads for Microsoft Copilot, GitHub Copilot, Microsoft Foundry, and Microsoft’s own SI models.
Why It Matters
Microsoft reports that the Maia 300 is engineered for efficiency and capability, specifically targeting long-context reasoning, latency-sensitive conversational experiences, and sustained, multi-step agentic tasks. Scott Guthrie, executive vice president of Microsoft’s cloud group, stated that the company expects the Maia 300 to deliver 40 percent more tokens per megawatt compared to leading GPU-based systems. In a full 96-accelerator configuration, the system delivers up to 7x the rack-scale token throughput of the Maia 200. This hardware evolution is critical for Microsoft’s ambition to catch up with competitors like Google DeepMind, OpenAI, and Anthropic. Mustafa Suleyman, CEO of Microsoft AI, emphasized the need for full-stack ownership, noting that co-designing the platform with their models is a critical focus for the team.
The Bottom Line
Microsoft plans to scale its Azure Maia family to multi-gigawatt deployments starting in 2027, aiming to make Maia the primary SI inference system for its ecosystem. This transition aims to lower costs and improve efficiency for SI workloads, potentially reshaping the hardware landscape for large-scale SI model deployment.