Microsoft has begun routing a portion of user prompts inside Excel and Word to its own MAI models rather than to software from OpenAI and Anthropic, according to a Bloomberg report published July 8, 2026. The shift, first flagged in Microsoft's two most widely used productivity programs, marks a visible step toward reducing the company's dependence on the two labs whose models it has long marketed as the engine behind Office 365. Microsoft told TechCrunch it had nothing further to share.
The move is a cost play. AI inference bills have climbed sharply across the industry, and Microsoft has been building out an internal model lineup that can absorb the cheaper, higher-volume queries where a frontier model is overkill. Third-party models from OpenAI and Anthropic still power large portions of Office 365, but the mix is changing.
At its annual Build conference in June 2026, Microsoft announced seven new MAI models, including an agentic coder and a text-to-image generator. That launch broadened the internal roster from a small research effort into a production-grade family the company can actually route real user traffic through. Excel and Word appear to be the first at-scale deployment where MAI is handling a meaningful share of prompts.
Key facts
- 01Microsoft is routing a share of Excel and Word AI prompts to its in-house MAI models instead of OpenAI and Anthropic, per a Bloomberg report on July 8, 2026.
- 02At its Build conference in June 2026, Microsoft launched seven new MAI models, including an agentic coder and a text-to-image generator.
- 03Amazon, Uber, Meta, and Accenture have all reportedly moved to curb AI spending in the last few months of 2026.
- 04Microsoft told TechCrunch it had nothing further to share on the reported shift.
- 05Some Silicon Valley firms are reportedly evaluating Chinese models for cheaper agentic workloads despite security concerns.
The choice of Excel and Word is deliberate. Both applications generate an enormous volume of relatively short, structured prompts — formula help, summarization, rewrites, formatting — that do not require the full capability of a top-tier frontier model. A cheaper in-house model that hits acceptable quality on those tasks can meaningfully reduce per-query cost when multiplied across the Office 365 user base.
Microsoft is not alone in tightening the belt. Amazon, Uber, Meta, and Accenture have all reportedly made moves in recent months to curb AI spending after a brief period earlier in 2026 that some in the industry described as tokenmaxxing — a race to pump as many tokens as possible through frontier APIs regardless of unit economics. That phase appears to be ending.
The sticker shock has pushed some Silicon Valley firms to evaluate Chinese models, including offerings that have emerged as cheaper substitutes for agentic workloads, despite concerns about security and data handling. Microsoft's response has been the opposite: build in-house rather than outsource further down the cost curve to a foreign provider.
The dynamic complicates Microsoft's relationship with OpenAI, its largest AI partner and an investee. Microsoft has spent years positioning OpenAI models as the core intelligence layer inside Office, and the two companies remain deeply intertwined commercially. Quietly shifting prompt volume to MAI without a formal announcement is the kind of move that signals divergence without forcing a confrontation.
Anthropic sits in a similar spot. Microsoft integrated Anthropic's Claude models into parts of Office 365 earlier this year as a way to diversify beyond OpenAI, and any drop in Microsoft-driven Claude usage would show up in Anthropic's numbers. Neither lab has commented publicly on the reported reallocation.
There are risks. In-house models still trail frontier systems on the hardest reasoning and long-context tasks, and Microsoft has not published benchmark comparisons showing how MAI stacks up against the GPT and Claude models it is partially replacing. If quality slips on user-visible features inside Excel and Word, the savings evaporate quickly against the reputational cost. Microsoft has not disclosed what percentage of prompts MAI is now handling, so the scale of the shift remains unclear.
For the broader AI market, Microsoft's move is a signal that even the deepest-pocketed buyer of frontier inference is now segmenting its workloads by cost. The frontier labs still own the hardest queries, but the middle of the market — the enormous volume of routine assistant tasks inside enterprise software — is now contestable by cheaper in-house alternatives. That pressure lands directly on the revenue lines of OpenAI and Anthropic, both of which count Microsoft-driven Office 365 usage as a material share of their commercial API traffic.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
