Microsoft is significantly ramping up its investment in compute capacity to develop its own frontier AI models, according to Microsoft AI chief Mustafa Suleyman, following the recent launch of its first in-house AI models.
The initial in-house model, MAI-1-preview, was trained on a cluster of 15,000 H100 GPUs. Microsoft plans to scale its cluster to six-to-ten times the size of the MAI-1-preview cluster, indicating a substantial commitment to building powerful AI models comparable to those developed by Meta, Google, and xAI.
Suleyman stated during an employee town hall that the company aims to build world-class models internally while also utilizing external models when necessary. Microsoft CEO Satya Nadella emphasized the company’s intention to build “model-forward products” and support multiple models across its offerings. He cited GitHub Copilot as a prime example of this strategy.
Nadella underscored that Microsoft will “definitely support multiple models” in its products. In addition to developing its own models, Microsoft is reportedly planning to integrate Anthropic’s AI models into Microsoft 365. According to The Information, Microsoft 365 Copilot will be “partly powered by Anthropic models” after these models outperformed OpenAI in specific tasks within Excel and PowerPoint.
This move highlights Microsoft’s strategy of leveraging both in-house and external AI capabilities to enhance its product suite. Suleyman concluded by saying “Much more to do, but it’s good to take the first steps,”
