Microsoft AI introduced its first in-house models last month, a move that complicates its partnership with OpenAI. Now, Microsoft AI chief Mustafa Suleyman says the company is investing heavily in compute capacity to support future frontier models.
“We should have the capacity to build world-class frontier models in-house at every scale,” Suleyman told employees during a Thursday town hall. “But we should be pragmatic and use other models when necessary. We’re also investing in our own cluster.”
Currently, the MAI-1-preview model was trained on 15,000 H100 GPUs, which Suleyman described as small by industry standards. He hinted that Microsoft plans to train frontier models on clusters six to ten times larger than the MAI-1-preview effort. “Much more to do, but it’s good to take the first steps,” he added.
Microsoft CEO Satya Nadella echoed this ambition. He said he is eager to see Microsoft build strong model capabilities to deliver “model-forward products.” Nadella also emphasized the company’s commitment to supporting multiple models across its ecosystem. GitHub Copilot, he noted, is the clearest example of that approach.
With growing investments and broader partnerships, Microsoft is signaling a future where its AI ecosystem is not tied to a single provider.