Microsoft is increasing use of its own MAI systems in Excel and Outlook to lower AI inference costs as discounted access to outside models may not last and broader cost cuts deepen.
Microsoft has begun replacing some OpenAI and Anthropic model usage in Excel and Outlook with its internally built MAI systems as it tries to rein in a fast-growing artificial intelligence bill. Tens of thousands of prompts in the two applications are now being processed each week by Microsoft’s own models, though MAI still represents only a small share of the company’s overall AI activity. The shift reflects pressure to reduce inference costs across products such as Copilot, where Microsoft consumes large volumes of AI tokens and currently benefits from discounted computing through its long-standing OpenAI partnership. Mustafa Suleyman said in June that Microsoft was seeking to cut spending on Anthropic by relying more on internal systems, saying the company ultimately wants to eliminate that cost. The effort follows Microsoft’s mid-May 2026 move to wind down most internal Claude Code licenses and comes as the company pursues wider spending restraint alongside heavy AI investment, including workforce cuts affecting 2.1% of staff, or 4,800 jobs, with roughly 3,200 reductions in Xbox.