What is Microsoft's MAI shift in Excel and Outlook? It is Microsoft quietly replacing OpenAI and Anthropic models with its own in house MAI models inside Excel and Outlook to cut the cost of running Copilot, a change first reported by Bloomberg on July 7, 2026. For commercial real estate teams that live in spreadsheets and email, this Microsoft MAI models change matters because the AI engine behind your Copilot prompts is being swapped with no visible change to the interface. For the broader landscape of tools, see our guide to the best AI tools for commercial real estate.
Key Takeaways
- Microsoft is routing tens of thousands of weekly Copilot prompts in Excel and Outlook to its own MAI models instead of OpenAI and Anthropic to reduce inference costs.
- MAI is a family of seven models introduced at Build 2026, led by MAI-Thinking-1 for complex reasoning and optimized for productivity, latency, and data residency.
- Microsoft AI CEO Mustafa Suleyman said the goal is to reduce and ultimately eliminate spending on Anthropic, signaling a long term shift, not a one time test.
- Routine tasks like email summaries, spreadsheet analysis, and draft replies now run on MAI, while complex creative work still routes to OpenAI models.
- Because MAI runs on Microsoft's own infrastructure, prompt data stays inside Microsoft's environment, a plus for confidential deal information.
Microsoft's MAI Models in Excel and Outlook, Explained
The change is a model swap under the hood, not a new app. Microsoft has begun serving many Excel and Outlook Copilot prompts with MAI, its internally built large language models, rather than the OpenAI and Anthropic engines that have powered Copilot since launch. Bloomberg reported that tens of thousands of prompts per week in these two apps now run on MAI, though the in house models still handle only a small share of Microsoft's overall AI usage.
MAI is not a single model. Introduced at Microsoft Build 2026, the lineup spans seven models covering reasoning, coding, image generation, speech, and transcription, led by MAI-Thinking-1 for complex reasoning. Microsoft optimized them for productivity tasks, emphasizing cost efficiency, latency, and data residency. The company frames this as demoting the frontier labs from default infrastructure to a premium option reserved for the hardest work, a strategy that echoes the build versus buy debate we covered in the Microsoft Copilot shakeup.
Crucially, this is a hybrid split, not a wholesale swap. Bloomberg's reporting indicates that routine, high volume tasks such as summarizing an email thread, analyzing spreadsheet figures, or drafting a reply now run on MAI, while more complex, open ended work such as generating a long document from scratch still routes to OpenAI's models. For CRE users, that means the fast, repetitive prompts you fire dozens of times a day are the ones most likely already running on Microsoft's own engine, while a heavy drafting request may still reach a frontier model.
Why the Model Swap Matters for CRE Teams
It matters because Excel and Outlook are the daily operating system of commercial real estate, so any change to the model behind them touches underwriting, asset management, and investor relations. When a broker asks Copilot to summarize a lease abstract in Outlook, or an analyst asks Excel to extract totals from a trailing twelve month statement, a different model may now be answering than did last month. Industry advisors including CBRE report that CRE has moved decisively from AI pilots into daily use, which raises the stakes on any quiet change to the underlying engine.
The practical implication is that output can shift in tone, format, and occasionally accuracy, even when your prompt does not change. A model tuned for cost efficiency may summarize a rent roll differently than a frontier model would. This does not make MAI unfit for CRE work; it makes testing essential. Treat the swap as a reason to re validate any Copilot driven workflow that feeds a number into a model, the same discipline we stress in our piece on why CRE firms win with better data, not just bigger models.
The Upside: Data Residency and Cost Discipline
The clearest win for CRE is data residency. Because MAI models run entirely on Microsoft's own infrastructure, prompt data does not leave Microsoft's controlled environment to reach a third party lab. For firms handling confidential purchase and sale agreements, rent rolls, and investor data under nondisclosure agreements, keeping prompts inside the Microsoft 365 boundary reduces third party exposure.
- Tighter data boundary: Fewer external model calls means fewer places sensitive deal data travels.
- More predictable roadmap: Microsoft controlling its own models reduces the risk of an abrupt third party price hike or model suspension flowing through to Copilot.
- Coming model choice: Microsoft is expected to let admins choose between MAI, OpenAI, or customer supplied models per task, giving CRE IT teams real governance control.
These are the kinds of vendor governance questions The AI Consulting Network helps CRE firms work through before they standardize a tool across the organization.
What CRE Technology Leaders Should Do Now
The right response is measured, not reactive. The swap is gradual, Copilot still works, and for many routine tasks MAI will be indistinguishable or better. But a few concrete steps protect quality:
- Re test critical workflows: Re run your most important Copilot prompts in Excel and Outlook and compare outputs against a known good baseline.
- Keep humans on the numbers: Any output that becomes a cap rate, NOI, or DSCR input still needs analyst review, regardless of which model produced it.
- Track the roadmap: Watch Microsoft Ignite in November 2026 for the MAI roadmap and the promised model choice controls.
- Map your dependencies: Know which of your AI workflows sit inside Microsoft 365 versus a standalone tool, so a future change does not surprise you.
This is also a reminder that an enterprise knowledge layer matters as much as the model. We covered Microsoft's version of that in our look at Microsoft Work IQ, and the model comparison context in our AI model comparison for CRE investors.
Frequently Asked Questions
Q: What are Microsoft MAI models?
A: MAI models are large language models built in house by Microsoft's AI division, introduced at Build 2026. The lineup includes seven models across reasoning, coding, image, speech, and transcription, led by MAI-Thinking-1. Microsoft is now using them to answer many Copilot prompts in Excel and Outlook instead of OpenAI and Anthropic models.
Q: Will the Microsoft MAI change affect my Copilot subscription price?
A: Not directly and not yet. The shift is about Microsoft's own inference costs, not your subscription. Over time, lower internal costs could influence Copilot pricing and margins, but the immediate impact is on which model answers your prompts, not on your bill.
Q: Is my data safer with MAI models?
A: In one respect, yes. MAI models run on Microsoft's own infrastructure, so prompt data stays inside Microsoft's environment rather than being sent to a third party lab. That is a meaningful plus for CRE firms handling confidential deal data, though you should still confirm your specific Microsoft 365 data protection terms.
Q: Should CRE teams stop using Copilot because of this change?
A: No. The smarter move is to re test your most important Excel and Outlook workflows and keep a human reviewing any financial output. For hands on help evaluating how the MAI shift affects your stack, CRE investors can reach out to Avi Hacker, J.D. at The AI Consulting Network.