Microsoft unveils Maia 300, its most powerful AI chip yet - and it might make using Teams, Copilot and Cowork much better at last (www.techradar.com)

Microsoft has announced the Maia 300, its most powerful AI chip to date, designed to enhance its Azure infrastructure and set to begin deployment in 2027. This "inference-first" architecture promises to improve efficiency and power, delivering 40% more tokens per megawatt compared to leading GPU systems. With a rack of 96 Maia 300 accelerators capable of achieving seven times the token throughput of the previous Maia 200, the new chip aims to elevate applications across Microsoft’s suite, including Teams, Copilot, and GitHub, enabling more effective long-context reasoning and multi-step agent tasks. Key technical innovations of Maia 300 include its use of eight stacks of HBM4 memory, achieving peak bandwidth of 18TB/s, and a multi-chiplet architecture that optimizes compute, memory, and I/O functions. Its design aims to minimize data movement through software-defined dataflow architecture and integrates high-bandwidth connections for tensor parallelism, allowing for scalable deployments from rack to data center levels. The Maia 300 is positioned to significantly enhance AI capabilities within Microsoft’s existing and future products, paving the way for improved user experiences and larger workloads.