Using the most advanced AI models at a corporate scale isn’t cheap, even for a company as massive as Microsoft. The cost of AI tokens, the unit used to measure how much computing work a model is doing, is getting so expensive that Microsoft is reportedly turning to its own models to save some cash.
Bloomberg reports, citing an unnamed source, that tens of thousands of AI prompts each week in Microsoft’s Excel and Outlook software are now being completed using Microsoft’s own MAI models. The spreadsheet and email programs previously relied more heavily on models from OpenAI and Anthropic to complete some tasks.
That is still only a small fraction of Microsoft’s overall AI usage, Bloomberg notes. For example, Copilot, the company’s workplace AI assistant, requires massive amounts of AI tokens.
Microsoft declined to comment on the report.
The news comes a little over a month after Microsoft announced seven new in-house models, including MAI-Thinking-1, Microsoft AI’s first reasoning model.
At the time, the company said MAI-Thinking-1 was built for “high efficiency and performance, but importantly, at a low-token cost.” Microsoft describes the model as a mid-sized, 35 billion active-parameter model with a 256K context window. According to a blind test, it matched the coding abilities of Anthropic’s popular Claude Opus 4.6.
Microsoft also rolled out new image, transcription, voice, and coding models.
The new models arrive as cheaper and more efficient AI models are getting more attention across the industry.
China’s DeepSeek made headlines earlier this year when it realeased new budget-friendly models. For instance, DeepSeek charges $0.435 per million input tokens and $0.87 per million output tokens for its V4-Pro model. While, Anthropic charges $10 per million input tokens and $50 per million output tokens for its most advanced Fable 5 model.
“Anthropic is extremely expensive and I think many people are urgently looking for alternatives,” Microsoft AI CEO Mustafa Suleyman told Bloomberg last month. “We pay a lot of money to Anthropic—so our goal is to reduce and ultimately eliminate that cost.”
OpenAI’s pricing is cheaper than Anthropic’s, at $5 per million input tokens and $30 per million output tokens for API use of GPT-5.5. Microsoft also enjoys a discount through its partnership with OpenAI. But the clock is ticking on the current deal, which ends in 2032.
Still, it looks like Microsoft is trying to get ahead of those pricey costs.
Suleyman told Bloomberg that “many, many people in our organization are spending millions of dollars” on AI tokens.