API
< Back to all clusters
[TECHNOLOGY] · United States · 3 sources

started · updated

Microsoft unveils Maia 300 AI accelerator for Azure

Microsoft has unveiled the Maia 300, its next-generation AI inference accelerator designed to improve the efficiency of its AI services. The hardware is intended to power complex reasoning tasks, conversational applications, and continuous AI agent workflows within Azure.

Built using a multi-chiplet design on TSMC’s N2P process, each Maia 300 accelerator features 288 GB of HBM4 memory and 18 TB/s of peak bandwidth. Microsoft reports that the chip provides 2.5 times the memory bandwidth and 33% more capacity than the previous Maia 200. The company claims the Maia 300 could generate “40% de tokens supplémentaires par mégawatt par rapport aux principaux systèmes basés sur des GPU” (40% more tokens per megawatt compared to leading GPU-based systems).

Large-scale deployment of the Maia 300 is expected to begin in 2027, targeting multi-gigawatt deployments. This development places Microsoft in direct competition with Nvidia’s Vera Rubin architecture as both companies race to optimize data center power limits and meet the growing demand for generative AI processing.

Entities

Azure · Microsoft · Nvidia · TSMC