Microsoft Announces Maia 300 AI Chip

Microsoft has unveiled the third generation of its proprietary AI silicon, the Maia 300. The development of this in-house chip is part of a strategic initiative to decrease the company's dependency on hardware procured from Nvidia.

The Maia 300 processor is slated for use within the technology giant’s multi-gigawatt infrastructure deployments in 2027, serving as the primary inference system for Microsoft’s suite of products and services. The software firm anticipates potentially deploying over one million of these chips within the next calendar year.

According to a communication from EVP Scott Guthrie, Microsoft expects the Maia 300 to provide 40% more tokens per megawatt compared to prominent GPU-based systems, though the company did not explicitly name Nvidia. The Maia 300 is engineered as a comprehensive system, where the software, silicon, physical rack architecture, networking, and data center environments are co-designed to function as a unified platform.

Guthrie observed that a 96-chip rack configuration provides seven times the rack-scale token throughput of the previous Maia 200 model. He stated that the new processor features an inference-centric architecture optimized for long-context, memory-intensive, and decode-heavy artificial intelligence workloads.