Microsoft is planning to expand its computing capacity for the training of its own in-house models.

As reported by Bloomberg, Microsoft's AI chief, Mustafa Suleyman, shared some of the company's plans with employees during a September 11 meeting.

According to Suleyman, Microsoft will make "significant investments" in its AI clusters to train Large Language Models (LLMs) that it hopes will compete with the likes of OpenAI and Anthropic.

At the start of 2025, Microsoft estimated that its capex for the year would reach $80bn.

Suleyman told employees that it is critical for the company to have the ability to be "self-sufficient" in AI.

He added that the company is "deepening" its ties with OpenAI, partnering with other model makers, and building its own.

Microsoft declined to comment to Bloomberg.

Suleyman was a co-founder of Google's DeepMind, but joined Microsoft in 2024. In August, the company released its first LLM under Suleyman's leadership, which was trained on 15,000 Nvidia H100 GPUs.

Historically, Microsoft has relied heavily on LLMs from OpenAI. CEO Satya Nadella has said it now plans to develop products with a multi-model approach. The Information reported that this will see it using some Anthropic models earlier this week.

Microsoft and OpenAI have had a long-standing relationship, and today signed a non-binding Memorandum of Understanding for "the next phase of our [their] partnership" as OpenAI has restructured its business. The two are still negotiating details of their contract.