Bloomberg reported on July 7, 2026 that Microsoft is starting to route some Excel and Outlook AI prompts from OpenAI and Anthropic models to its own MAI models. A The Fly item on TipRanks, citing Bloomberg, says tens of thousands of weekly prompts in those apps are already being completed with Microsoft-built models. For practitioners, the signal is not a full partner break; it is production model routing for cost, latency, and product fit inside Microsoft 365. Enterprise AI teams should expect more mixed-model portfolios behind familiar copilots.
The important signal is not that Microsoft has stopped using partner models; the evidence points to a more practical shift toward workload-level routing. In high-volume productivity apps, the model behind a feature can change for cost, latency, availability, or product-fit reasons while the user still sees the same Copilot surface.
What happened
Bloomberg, in a report republished by Yahoo Finance, said Microsoft is starting to replace OpenAI and Anthropic models with its own models in some software products, including Excel and Outlook. A The Fly item on TipRanks, also citing Bloomberg, says tens of thousands of weekly prompts in those applications are already being completed with internally built MAI models, according to a person familiar with the work. The reporting frames the move as part of Microsoft's effort to reduce AI costs.
Technical context
Microsoft had already signaled this direction at Build 2026. Official Microsoft AI and Build materials describe a broader MAI model family across reasoning, coding, image generation, voice, and transcription. Those models are positioned less as one general-purpose replacement for every frontier workload and more as a portfolio of systems tuned for Microsoft product workflows. That is the architecture implied by the July 7 report: use first-party models where they are good enough and cheaper, and reserve partner models for tasks that need them.
For practitioners
Enterprise AI teams should treat productivity copilots as routed model portfolios, not static wrappers around one provider. That has procurement and evaluation consequences. A feature that performed one way last quarter may shift as Microsoft changes the underlying model, and internal AI governance needs to test outputs, latency, cost, compliance posture, and data handling at the feature level. The same pattern is likely to spread beyond Microsoft as vendors try to control inference costs.
What to watch
The next useful signal is whether Microsoft documents which workloads use MAI models, exposes tenant-level controls, or changes pricing and performance guarantees around Microsoft 365 Copilot features. Clearer model-disclosure and evaluation hooks would help customers distinguish normal optimization from changes that affect regulated workflows, auditability, or vendor-risk reviews.
Key Points #
- 1Bloomberg says Microsoft is routing some Excel and Outlook AI prompts to in-house MAI models to reduce inference costs.
- 2Microsoft's Build materials show MAI has become a broader model family across reasoning, image, voice, transcription, and coding.
- 3Enterprise teams should treat copilots as model portfolios where cost, latency, governance, and vendor dependency can change by task.
Scoring Rationale #
This is a notable enterprise-AI shift because Microsoft 365 is a high-volume production surface where model-routing decisions affect cost, reliability, and vendor dependency. The score stays below major-policy territory because the core claim is still sourced to Bloomberg reporting and appears limited to some app workloads.
Sources #
Public references used for this report. Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
[Active Search Campaigns by BudgetEasy](/problems/sql/active-search-campaigns-by-budget)
[High CPC Clicks & Poor Landing PagesMedium](/problems/sql/high-cpc-clicks-poor-landing-page)
[Campaign ROAS by Attribution ModelHard](/problems/sql/campaign-roas-by-attribution-model)
250 free problems · No credit card