Sources: Microsoft, looking to reduce AI costs, is starting to replace models from OpenAI and Anthropic with its MAI models in products like Excel and Outlook
Context & Ripple Effects
Microsoft’s reported shift toward MAI in Excel and Outlook follows a multi-year effort to reduce the cost of serving AI features. Earlier coverage described smaller, distilled models for Bing-era features, then MAI-1 and plans to broaden the set of models available to 365 Copilot amid cost and speed concerns.
The company has also reportedly used Anthropic models for some Office 365 features when they performed better on particular tasks. That makes the current move less a wholesale model choice than an effort to control the economics of high-volume product workloads.
First-order effects
- Microsoft can move some Excel and Outlook AI inference from OpenAI and Anthropic to MAI, reducing its dependence on external model providers for those features.
- OpenAI and Anthropic face reduced model usage within the affected Microsoft product flows, while MAI becomes a more consequential internal production asset.
Second-order effects
- Microsoft’s product teams gain greater leverage to route tasks among in-house and third-party models based on cost, speed, and task performance rather than relying on one provider.
- The move raises pressure on external model suppliers to demonstrate differentiated performance in enterprise applications where Microsoft can substitute its own models.
Third-order effects
- If Microsoft continues deploying MAI across productivity software, frontier-model partnerships may increasingly evolve into multi-model supply arrangements rather than exclusive platform dependencies.
- The broader competitive advantage in enterprise AI may shift toward operators that can tailor and run models cheaply at product scale, alongside raw model capability.
The trend: This is one data point in the shift from single-provider AI integrations toward cost-optimized, multi-model stacks controlled by major software platforms.