Microsoft MAI Grabs Thousands of Office Prompts From OpenAI

Microsoft has already moved part of its Office AI workload onto Microsoft MAI models, turning an internal cost-control strategy into a live production shift inside Excel and Outlook.

The sharper question is this: if Office prompts can move quietly from OpenAI and Anthropic to Microsoft’s own models, how much of Copilot is really a partner-powered product versus a Microsoft-controlled AI stack? The shift was reported Tuesday, July 7, with Bloomberg citing unnamed sources, according to PYMNTS.

Is Microsoft MAI already answering Office prompts?

Yes, according to the report. Microsoft has begun using its own MAI artificial intelligence models to complete prompts in Excel and Outlook, rather than relying only on models from OpenAI and Anthropic.

That matters because this is not just a research roadmap or a demo-stage model family. PYMNTS, citing Bloomberg, reported that Microsoft’s MAI is now handling tens of thousands of prompts in those applications each week.

The company did not immediately reply to PYMNTS’ request for comment.

The source also says the scale of MAI usage in those apps had not previously been known. Prior reports had already pointed to Microsoft’s plan to move more work onto its own models, but this report describes actual workload routing inside live productivity applications.

The practical read: this is Microsoft testing whether its own AI can handle specific Office tasks at production volume. It does not mean OpenAI or Anthropic are gone from Microsoft products. It does mean Microsoft is no longer treating external models as the automatic default for every prompt.

That fits a pattern readers may recognize from Microsoft’s Copilot sprawl. XOOMAR has separately tracked how Microsoft’s AI product strategy has become harder to parse in Microsoft Copilot App Merger Exposes Its AI Sprawl Problem. The Office prompt shift adds another layer: the app surface may stay the same while the model underneath changes.

Why is Microsoft reducing dependence on OpenAI and Anthropic inside Office?

The reported driver is cost and control. According to the Bloomberg report cited by PYMNTS, Microsoft is shifting to its own models to reduce AI costs and avoid being reliant on other AI labs.

That is the commercial core of the story. Every high-volume AI feature creates a recurring inference bill, a pressure also shaping the broader frontier AI price fight. If Microsoft can route routine Excel and Outlook tasks to Microsoft MAI models, it can reduce payments to outside model providers while keeping the user inside Microsoft 365.

The report also points to Mustafa Suleyman, CEO of Microsoft AI. Previous Bloomberg reports cited by PYMNTS said Suleyman had stated that Microsoft aims to reduce its spending on Anthropic models by using its own. Those reports also said that within months, a Microsoft model will start performing transcriptions in Teams and other products.

Microsoft’s public position has been that it uses a mix of models. In March 2025, after reporting that Microsoft was testing alternatives to OpenAI models in Copilot, a Microsoft spokesperson told PYMNTS:

“As we’ve said, we are using a mix of models, which includes models from OpenAI as part of our partnership, as well as Microsoft AI and open-source models.”

That quote now reads less like a hedge and more like the operating model.

PYMNTS also reported in January that Microsoft had become one of Anthropic’s top customers and was on track to spend around $500 million per year to use Anthropic’s AI models in Microsoft products. That figure explains why even a partial shift to Microsoft MAI models carries weight.

Will Office users notice if Excel or Outlook switches models?

Most users probably won’t know which model answered a spreadsheet or email prompt unless Microsoft decides to tell them. The report describes model routing behind Excel and Outlook, not a new user-facing product label.

For enterprise buyers, though, invisible does not mean irrelevant.

The questions will be concrete:

• Performance: Does a prompt in Excel or Outlook produce the same quality of answer when MAI handles it?
• Reliability: Are responses consistent across Microsoft 365 workflows if different models sit behind different features?
• Security: Does model routing change how customers think about data handling and internal review?
• Contracts: Do Copilot commitments depend on a specific model provider, or only on Microsoft’s product-level promises?
• Auditability: Can customers understand which AI system produced a given output if they need to review it later?

Those questions are analysis, not claims from the source. The source does not say Microsoft has changed Copilot terms, compliance guarantees, or customer controls. It says Microsoft is now using its own MAI models for some Excel and Outlook prompt completions.

Still, the implications are obvious for CIOs and procurement teams. If Microsoft can swap the model under a familiar Office feature, buyers will want to know whether the change affects performance, support, or governance.

The timing also lands against Microsoft’s broader AI spending debate. PYMNTS reported in April that Microsoft delivered double-digit growth across core segments, driven by demand for AI and cloud services, while facing investor concerns over elevated capital spending. XOOMAR has also covered Microsoft’s AI spending pressure in Microsoft Layoffs Hit Xbox as AI Spending Wins Again, a separate story that shows how closely Microsoft’s AI investments are being watched.

Does this mean Copilot will lean harder on Microsoft-built AI next?

That is the question that won’t be answered in one Bloomberg report.

The next signal is whether Microsoft expands Microsoft MAI models beyond select Excel and Outlook prompts into more Copilot features across Word, Teams, PowerPoint, and other business apps. PYMNTS’ source material already points to Teams transcription as an area where a Microsoft model is expected to play a larger role.

Microsoft’s public comments will matter. Watch for any detail on model routing, customer transparency, performance benchmarks, and how the company describes OpenAI’s role as its own models take on more Office work.

The strategic read is narrow but important: Microsoft is not breaking with OpenAI. It is building leverage inside the products where AI is supposed to generate returns. If Microsoft MAI models prove good enough for routine Office tasks, the company can keep premium outside models for harder workloads and pull the rest in-house.

The Bottom Line

• Microsoft is moving real Office AI workloads onto its own models, not just testing them in demos.
• The shift could reduce Microsoft’s dependence on OpenAI and Anthropic for Copilot features.
• Users may see Copilot become more of a Microsoft-controlled AI stack over time.