Microsoft starts swapping OpenAI and Anthropic models for its own in parts of Office and Outlook

Microsoft is beginning to replace OpenAI and Anthropic in select AI-powered features within Microsoft apps such as Excel and Outlook, as the company works to reduce the cost of using third-party artificial intelligence models, according to a report published Tuesday by Bloomberg and carried by Yahoo Finance.

The change, described as a move toward Microsoft’s own models in “some apps,” indicates a shift in how the company builds and delivers AI assistance in its widely used productivity software. Microsoft’s Office suite and email and calendar products are among its most important distribution channels for AI features, making the underlying model choice relevant both to user experience and to the economics of deployment.

The report frames the company’s decision primarily around AI cost control. Running large language model inference and serving responses at scale can be expensive, and Microsoft has been under pressure across the industry to justify AI spending as adoption grows. A migration from external model providers to internally managed models can also give Microsoft more flexibility over performance, latency, and product behavior.

For Microsoft, the practical effect is that some AI experiences inside familiar workflows may be powered by different “brains” even if the interface looks similar. In Excel and Outlook, that can include tasks that summarize content, draft or rewrite text, generate responses, or help interpret information. Microsoft did not provide additional detail in the reported account on exactly which features are switching, the timelines, or how fully each app will rely on Microsoft-made models versus third-party options.

The company’s broader AI strategy has already included building more of its stack in-house. Recent coverage around Microsoft’s AI roadmap has pointed to in-house model development and expanded hardware efforts designed to reduce dependency on external providers. Those steps, if they continue, would support the kind of gradual substitution described in the new report: keep the user experience consistent, but change the supply chain of model capacity behind the scenes.

Sector context matters because Microsoft is competing in the “enterprise AI assistant” market where large language model providers are both partners and rivals. OpenAI and Anthropic are major suppliers of frontier models, but they have also faced industry scrutiny over the economics of model access and the pricing of inference. Microsoft, which distributes AI through products used daily by businesses, has leverage to steer the architecture of its own assistant features over time.

Still, significant questions remain. The report does not specify whether the model swaps are limited to certain regions, languages, or subscription tiers. It also does not say whether the transition is all-or-nothing for a feature, whether users can expect different outputs as models change, or whether Microsoft will maintain multiple model providers in parallel for different tasks.

For now, the development is best read as an incremental but notable attempt to bring more of the AI supply chain under Microsoft’s control. What to watch next is whether the company expands the rollout from “some apps” to additional Office and productivity surfaces, and whether Microsoft provides clearer disclosure about which underlying models power specific Copilot-like capabilities over time.