
Microsoft’s own AI chip, headed to multi-gigawatts
Maia 300 is Microsoft’s next-generation AI inference platform, announced Oct 12, 2026. Per Microsoft, it’s built for long-context reasoning, low-latency conversation, and agents, with 8 HBM4 stacks, TSMC N2P compute chiplets, and 96-accelerator racks. Microsoft claims 40% more tokens per megawatt. Production starts in 2027, and Azure Maia scales to multi-gigawatt deployments, becoming the main inference system for Microsoft’s products.
The Microsoft Maia 300 is a next-generation AI inference platform designed for long-context reasoning and low-latency applications, featuring 8 HBM4 stacks and TSMC N2P compute chiplets. Production is set to begin in 2027, with the platform aimed at scaling to multi-gigawatt deployments within Azure.
A source that found nothing is a measurement. A source that has not run is a gap. Neither means the launch lacks the thing.