Microsoft Plans to Unveil Maia 300 AI Chip in September…Report

The Information reported that Microsoft plans to unveil the Maia 300 AI chip this coming September. This is part of a strategy to reduce dependence on Nvidia GPUs and expand its own AI semiconductor ecosystem.

Microsoft has ordered over 300,000 Maia 300 units from TSMC, with products expected to be supplied starting in 2027. This is interpreted as a long-term investment to increase the use of its own chips in Azure AI infrastructure.

The Maia 300 is the successor to the first-generation Maia unveiled in November 2023. The first-generation chip was optimized for LLM training and inference based on a 5nm process and 105 billion transistors, but production scaling lagged behind initial targets and competitors.

Microsoft is persuading major cloud customers, including Anthropic, to adopt the Maia 300. If Anthropic migrates some AI workloads, it could shift the Nvidia-centric market landscape, but migrating models from the CUDA ecosystem to a new architecture requires significant engineering effort.

The Maia 300 is expected to integrate with Microsoft's Azure AI infrastructure and support major frameworks such as PyTorch and TensorFlow. It targets both training and inference, focusing on the efficiency of large Transformer models, but specific benchmarks have not been released.

As Google TPUs and Amazon Trainium spread and Nvidia GPU shortages continue, Microsoft's large-scale investment in its own chips is expected to further intensify the AI semiconductor competition among cloud companies.