Microsoft Is Using Its Own AI More in Excel and Word to Save Money

Microsoft has started routing some user requests in Excel and Word through its own AI models instead of sending them to OpenAI or Anthropic, according to Bloomberg reporting cited by TechCrunch on July 7, 2026. The shift is a cost-saving measure: Microsoft is moving some of the work onto computers it owns and controls, rather than paying external companies each time a user makes a request.

Microsoft declined to comment beyond what was reported.

A Change in Direction

Last September, Microsoft made a public point of saying that Excel and Word's AI features used models from both OpenAI and Anthropic — a decision that signaled the company was comfortable paying for top-tier AI from outside partners. Ten months later, some of that work has moved in-house.

The financial pressure behind this pivot had already surfaced. On June 4, 2026, Bloomberg reported that Microsoft's AI leadership said Anthropic's models had become too costly to run at the scale of Office 365, which has hundreds of millions of users.

Around the same time, at Microsoft's Build conference in early June, the company unveiled seven new AI models it had built in-house. One of them, called MAI-Thinking-1, is described as a "midsize" model — not designed to beat every top-tier model at every task, but built to be cheap and fast enough to handle the routine work that makes up most of what Excel and Word's AI actually does: generating formulas, summarizing documents, reformatting text.

The Longer Story

Microsoft's move to build its own models didn't happen overnight. Reuters reported back in May 2024 that the company was training a homegrown model competitive enough to challenge Google and OpenAI — long before cost became the public talking point. The path from "we're building our own model" to "we're advertising partnerships with OpenAI and Anthropic" to "we're quietly running more traffic through our own systems" tells the story of a company gradually maturing its internal capabilities over roughly two years.

Why This Makes Sense

Here's the economics in plain terms. Every time someone uses Copilot — that's Microsoft's name for its AI features — in Excel or Word, Microsoft pays OpenAI or Anthropic a small amount of money per request. Every request Microsoft can handle with its own models instead saves that cost. At the scale of hundreds of millions of Office users worldwide, even a small percentage shift adds up to real money.

Running your own computers to handle the work (what's already paid for as part of Azure, Microsoft's cloud service) is cheaper than paying an external partner per transaction, especially if the traffic is large and routine.

What We Don't Know

Microsoft has not said what percentage of requests now go through its own models, which specific features are affected, or whether the choice was purely about money or also about making sure users still get good results. That matters, because it is the difference between "users get the same experience for less cost to Microsoft" and "users might notice their AI features work slightly worse now." Without those details, the full picture remains unclear.

The broader context here involves more than just business efficiency. Microsoft holds a significant financial stake in OpenAI, so reducing the volume of work sent to OpenAI is also a statement about how Microsoft values in-house capability against the ongoing cost of the partnership. Any large company using AI now has to make this same calculation — paying for premium external models or investing in building competitive systems of their own. Microsoft, because it has both a massive user base and the research teams to build credible alternatives, is simply in a better position than most to make that choice visible.

The Broader Shift

What Microsoft is doing isn't unique. Large companies with enough scale are increasingly adopting a mixed approach: keeping expensive, top-tier models for the hard problems that really need them, while using cheaper, simpler systems for routine volume work. Microsoft's position stands out mainly because of how much traffic flows through its Office products — few other companies have both the user base creating cost pressure and the in-house research capacity to build a working replacement.

Whether Microsoft's in-house models will perform as well as the external ones over time is the question enterprises and analysts will be watching in the coming months.