Microsoft has begun integrating its own AI into its applications. Why is this important?

Microsoft, a software developer and cloud service provider, is beginning to replace OpenAI and Anthropic with its own AI models in software products such as Excel and Outlook in an effort to cut artificial intelligence costs, Bloomberg reported. According to a source at the news agency, tens of thousands of AI queries in widely used spreadsheet and email applications are now being processed by internal models from the Microsoft AI (MAI) family.

These models are also available for use in GitHub Copilot—the company's service for developing software using AI.

Bloomberg notes that the extent to which MAI is used in Microsoft's commercial products had not been disclosed previously. A company spokesperson declined to comment to the news agency on this matter.

Microsoft shares rose 2% during trading on July 7, but their gains slowed toward the close, ending the session up 0.5%.

Although MAI currently accounts for only a small fraction of the tech giant’s overall AI usage, this move shows that Microsoft is making progress in its efforts to create competitive AI models at a lower cost, Bloomberg notes. A month ago, Microsoft AI head Mustafa Suleyman stated that the company is trying to reduce spending on Anthropic by making more active use of its own models.

Microsoft uses massive amounts of AI tokens—units of computational resources consumed by artificial intelligence—in products such as its Copilot work assistant. For now, the company receives a significant portion of this technology at a discount thanks to its long-standing partnership with OpenAI and its status as a major investor in the developer of ChatGPT. But this arrangement won’t last forever, and Suleiman’s team is working to ensure that Microsoft doesn’t end up in a position where it has to pay the rates set by leading AI labs, the agency explains.

In June, at its annual Build conference for developers, Microsoft unveiled seven new AI models. One of them, according to the tech giant, is capable of matching the programming capabilities of Anthropic’s previous-generation Opus 4.6 model—which remains very popular—at a lower cost. “We pay Anthropic a lot of money, so our goal is to reduce those costs and, ultimately, eliminate them entirely,” Suleiman said at the time.

According to him, the speech decoding model developed by Microsoft will begin to be used in the Teams video conferencing app and other products in the coming months.

This article was AI-translated and verified by a human editor