Microsoft is taking a major step toward AI independence. Starting in July 2026, the company will begin diverting some of its Microsoft 365 Copilot workloads away from OpenAI and toward its own in-house MAI models, according to a new report from Bloomberg. The pivot, which touches apps like Excel and other productivity mainstays, signals a strategic effort to reduce reliance on a single AI partner while sharpening its competitive edge in the enterprise.
The move doesn't mean Copilot is breaking up with OpenAI entirely—at least not yet—but it does mean that the familiar blend of generative AI inside Word, PowerPoint, and Excel is about to get a homegrown twist. For the millions of users who've come to depend on Copilot's suggestions, data analysis, and content generation, the transition could bring subtle yet meaningful changes in performance, privacy, and cost.
The Shift to Microsoft's Own AI Models
The core of this story is routing. Copilot in Microsoft 365 has, until now, leaned heavily on OpenAI's large language models to power everything from summarizing email threads to generating formulas in Excel. Under the new plan, reported by Bloomberg, Microsoft will route a subset of those requests through its own MAI—"Microsoft AI"—models. The initial focus is on Excel, where complex calculations and data interpretations are prime candidates for an internal AI engine.
Why MAI? Microsoft has been quietly building its own generative AI capabilities for years, leveraging its vast research arm and the immense computational resources of Azure. These homegrown models are designed to be more efficient for specific tasks, potentially faster, and easier to integrate deeply with Microsoft 365's under-the-hood infrastructure. Details are thin, but the implications are clear: Microsoft wants an AI stack it controls from top to bottom.
Bloomberg's report suggests the shift is already underway, with Excel workloads being the first to move. Other productivity apps like Word and PowerPoint are expected to follow. This isn't a sudden flip of a switch—it's a gradual rerouting that will likely see a hybrid state for months or even years. Users might not even notice which model is handling their request, but behind the scenes, the plumbing is being reworked.
What This Means for Your Daily Workflow
For the average home user or student using Copilot in Excel to clean up a budget spreadsheet, the transition will probably be invisible. The core experience—asking Copilot to generate a formula, visualize data, or highlight trends—should remain consistent. What might change are the subtleties: response speed, the accuracy of complex requests, and how the model handles natural language quirks. Microsoft has a strong incentive to ensure MAI-powered features perform at least as well as the OpenAI-powered ones, if not better.
Enterprise users, however, stand to gain the most. Data governance has been a sticking point for many organizations. Routing Copilot requests through Azure-based MAI models could keep sensitive data entirely within Microsoft's ecosystem, easing compliance headaches and addressing the concerns of IT administrators who've been wary of sending proprietary information to third-party AI providers. This shift could unlock wider Copilot adoption in regulated industries like finance, healthcare, and government.
Admins may also get new levers to control model routing, similar to how they manage update channels today. A future dashboard might let them decide whether a specific department's Copilot queries go to OpenAI or MAI—or even pick different models for different tasks. While no such tools have been announced, the trajectory points toward greater flexibility as Microsoft builds out its own AI portfolio.
For developers who have been building atop Copilot APIs, the change introduces a new variable. If Microsoft eventually exposes MAI models via Azure AI services, it could open up a fresh playground of custom AI integrations—but also require revalidation of existing apps. For now, the developer story remains unchanged, but it's a space to watch.
The Road to MAI: Microsoft's Evolving AI Strategy
Microsoft's relationship with OpenAI has been both symbiotic and, at times, strained. The multi-billion-dollar partnership catapulted Microsoft into the AI era, giving it early access to GPT-4 and a dominant position with Copilot. But relying so heavily on an external lab—one with its own ambitions and a growing list of enterprise clients—carries obvious risks. Pricing volatility, feature alignment, and the simple fact that competitors can also license OpenAI's models have pushed Microsoft to accelerate its internal efforts.
The company has been dropping hints for months. At its Build 2024 conference, executives emphasized the role of "small language models" (like Phi-3) and the importance of fine-tuning AI for specific tasks. MAI models fit that narrative perfectly: they are likely smaller, nimbler, and optimized for Microsoft 365 workloads. Instead of calling a massive general-purpose model for every Copilot request, Microsoft can use a targeted MAI model that's faster and cheaper to run.
Cost is another critical factor. Processing a Copilot query through OpenAI isn't free—Microsoft pays per token, and while exact figures are closely guarded, it's a line item that can balloon with scale. By shifting even a fraction of those queries to infrastructure it already owns, Microsoft can rein in operational expenses. Some of those savings might eventually trickle down to customers, though don't expect a Copilot price cut overnight.
The timing is also notable. July 2026 is far enough out that Microsoft can test, refine, and certify MAI models without rushing. It coincides roughly with the typical rhythm of major Windows and Microsoft 365 updates, which often land in the second half of the year. By then, the AI landscape will likely be even more competitive, with Apple, Google, and Amazon all angling for a piece of the productivity pie.
What Should You Do Now?
If you're a regular Copilot user, the short answer is: nothing. The transition will be handled server-side, with no updates required on your end. Your familiar Copilot pane won't suddenly morph into something unrecognizable. However, it's wise to keep an eye on official Microsoft 365 communications later this year, which may shed light on feature differences or new capabilities tied specifically to MAI models.
For IT administrators, the preparation is more proactive. Start reviewing your organization's data residency and compliance requirements to understand how a model switch could affect you. While Microsoft hasn't published documentation yet, you can set the stage by auditing which Copilot features your teams rely on most heavily. If Excel data analysis is a cornerstone, pay extra attention to performance benchmarks and model behavior once early adopters get their hands on the MAI-powered version.
There's also a conversation worth having with Microsoft representatives or your licensing partner. Inquire about the data flow for Copilot under the new routing architecture—specifically, will there be an opt-in or opt-out mechanism? Could organizations designate which model processes their most sensitive files? Getting those questions on the record early could help shape the rollout to meet your business's needs.
Developers building on Microsoft 365 should track the Azure AI and Microsoft 365 Developer blogs. If MAI models eventually become available as APIs, you'll want to experiment early. The faster you can prototype with in-house models, the better positioned you'll be to optimize your applications when the shift becomes mainstream.
What's Next for Copilot and OpenAI
This isn't the end of the Microsoft-OpenAI partnership—far from it. Copilot will continue to tap OpenAI's models for many scenarios, especially those requiring broad world knowledge or creative content generation. But the era of exclusivity is fading. Microsoft's MAI push suggests a future where the two companies compete as much as they collaborate, with each serving different slices of the AI pie.
Expect more official confirmation from Microsoft later this year, likely alongside Windows 11 version 26H2 or a dedicated Microsoft 365 event. The roadmap will be critical: which apps get MAI first, how performance compares, and whether the transition introduces any temporary hiccups. Early internal testing is already underway, according to Bloomberg, so leaks and insider reports may give us a clearer picture before the public rollout.
For now, the message is clear: Microsoft is building its own AI brain, and it's plugging it into the suite that 345 million people use daily. The next time you ask Copilot to crunch numbers in Excel, the answer might come from a model that's entirely Microsoft's own—and that's a shift worth watching.