Microsoft’s new AI chip makes it less reliant on Nvidia’s GPUs

Microsoft has announced its third-generation Maia AI chip, Maia 300, designed to become the primary AI inference system for Microsoft's products and services. Set for production rollout in 2027, the chip promises 40% more tokens per megawatt than leading GPU-based systems and up to 7x the rack-scale token throughput of Maia 200 in a full 96-accelerator configuration. This efficiency boost will allow Microsoft to scale its Azure Maia family to multi-gigawatt deployments and rely mainly on its in-house chips instead of Nvidia and AMD GPUs. Maia 300 uses a new chiplet architecture on TSMC's N2P process, separating compute, memory controller, and I/O functions. It targets emerging AI workloads like long-context reasoning and agentic tasks, and is central to Microsoft's goal of catching up with Google DeepMind, OpenAI, and Anthropic.

(Source：The Verge)