Microsoft Shifts More AI Work to In-House Models

Microsoft is reportedly reducing its dependence on outside AI models from companies such as OpenAI and Anthropic as the cost of running artificial intelligence tools continues to rise across the tech industry.

The company has started using more of its own in-house MAI models inside major productivity products, according to recent reports. The shift does not mean Microsoft is ending its relationship with OpenAI or Anthropic, but it shows that the company wants more control over AI costs, performance and long-term product strategy.

Microsoft Moves More AI Work In-House

Microsoft has reportedly begun routing a portion of AI prompts in Office products through its own MAI models instead of relying only on third-party models. TechCrunch reported that Microsoft is using its homemade MAI models for some prompts in widely used programs such as Excel and Word, while Times of India, citing Bloomberg’s reporting, said tens of thousands of prompts in Excel and Outlook are now being handled weekly by Microsoft’s internal models.

This is a notable change because Microsoft has heavily promoted its AI products through partnerships with OpenAI and, in some areas, Anthropic. Microsoft Copilot, Microsoft 365 AI features, GitHub Copilot and other tools have been closely tied to advanced external models. Now, the company appears to be building a more balanced AI stack where its own models take on more work.

The Main Reason: AI Is Becoming Expensive to Run

AI tools are not cheap to operate. Every prompt, answer, summary, email draft or spreadsheet analysis uses computing resources. For a company like Microsoft, which serves enterprise customers and millions of workplace users, those costs can become very large.

The move toward in-house models gives Microsoft a way to reduce external model fees and improve margins. It also helps the company avoid being fully dependent on pricing decisions made by outside AI labs.

Microsoft’s MAI Model Family Is Growing

This shift follows Microsoft’s June announcement of seven new in-house AI models developed by Microsoft AI. The company said the MAI model family covers reasoning, coding, image generation, transcription and voice. Microsoft described the models as designed for real-world tasks and direct integration into products people already use.

The model family includes MAI-Thinking-1 for reasoning, MAI-Code-1-Flash for coding, MAI-Image-2.5 for image creation and editing, MAI-Transcribe-1.5 for transcription and MAI-Voice-2 for speech generation. Microsoft also said its MAI-Code-1-Flash model is built into GitHub Copilot and VS Code, showing that the company is already placing internal models inside developer tools.

Cost Cutting Is Becoming a Wider AI Trend

Microsoft is not alone. Across the AI industry, companies are facing pressure to reduce the cost of model training, inference, cloud infrastructure and data center expansion. TechCrunch noted that companies including Amazon, Uber, Meta and Accenture have also been linked to efforts to control AI spending.

Reuters Breakingviews also warned that major AI players are spending heavily on chips, data centers and energy infrastructure. It reported that large hyperscalers including Alphabet, Amazon, Meta, Microsoft and Oracle are expected to spend heavily on AI infrastructure between 2026 and 2030, increasing pressure on profit margins and cash flow.

This makes Microsoft’s strategy easier to understand. The company is still investing aggressively in AI, but it is also trying to make each AI feature cheaper to deliver.

OpenAI Partnership Still Matters

Microsoft’s move should not be read as a complete break from OpenAI. The company still has a major partnership with OpenAI, and advanced models from external labs remain important for many high-end AI tasks. TechCrunch also noted that Microsoft still relies on third-party models even as it expands use of its own systems.

The more realistic picture is that Microsoft is building a mixed-model strategy. Expensive, frontier-level models can be used where they are truly needed, while smaller or more efficient in-house models can handle routine tasks such as document help, spreadsheet prompts, coding assistance, transcription or simple workplace automation.

A Strategic Shift, Not Just a Budget Move

The decision is about more than saving money. By developing its own models, Microsoft gains more control over product design, speed, data handling, model tuning and enterprise deployment.

Microsoft has also promoted “Frontier Tuning,” a system where models can adapt to specific workflows. The company said its tuned MAI model for Excel can match GPT-5.4 while being up to 10 times more efficient, according to Microsoft’s own claims.

That efficiency claim is important because enterprise AI adoption depends not only on model quality, but also on cost. If Microsoft can deliver strong performance at lower cost, it can make Copilot and other AI tools more profitable over time.

The Bigger Picture

Microsoft’s reported move toward its own AI models shows how the AI market is entering a more practical phase. The first phase was about launching powerful tools quickly. The next phase is about making those tools affordable, scalable and profitable.

For users, the change may not be immediately visible. A Copilot answer in Excel or Outlook may still feel like the same AI assistant. Behind the scenes, however, Microsoft may increasingly choose which model handles each task based on cost, speed and complexity.

For the wider AI industry, the message is clear: owning the model stack is becoming a major advantage. Companies that can build, tune and deploy their own AI systems may be better positioned than those that depend entirely on external model providers.