Microsoft Moving to Internally Developed AI Models in Office Apps

• By John K. Waters
• 07/14/26

Microsoft is reportedly using its own in-house artificial intelligence models to handle some workloads in Excel and Outlook, offering new evidence that the company is moving its AI strategy beyond model development and into large-scale cost reduction.

According to Bloomberg (paywalled), Microsoft has begun replacing some OpenAI and Anthropic models in Microsoft 365 applications with its MAI models for selected tasks. Tens of thousands of prompts each week are now being processed by the in-house models, although they still account for only a small portion of Microsoft's overall AI usage. A Microsoft spokesperson declined to comment.

The reported deployment is notable not because Microsoft appears to be moving away from OpenAI or Anthropic, but because it offers new evidence that the company's AI strategy is evolving beyond the pursuit of frontier models. ("Microsoft Bets Enterprise AI's Next Battle is Deployment, Not Models").

Over the past several months, Microsoft executives have increasingly emphasized that the next phase of enterprise AI competition will be defined as much by deployment, economics, and operational efficiency as by raw model capability.

From Frontier Models to Frontier Economics

That message became clearer at Microsoft's annual Build developer conference in June.

Microsoft AI Chief Executive Officer Mustafa Suleyman introduced seven new MAI models spanning reasoning, coding, transcription, image generation, and other workloads. Among them was MAI-Code-1, which Microsoft said delivers coding performance comparable to Anthropic's earlier Opus 4.6 model at lower operating cost. During the presentation, Suleyman said Microsoft wanted to reduce, and ultimately eliminate, spending on Anthropic models.

The Bloomberg report suggests those efforts are beginning to move from strategy into production.

The development also reinforces a broader theme Microsoft Chief Executive Officer Satya Nadella has articulated publicly in recent months: that long-term AI leadership will depend not only on building powerful models, but also on creating the infrastructure, deployment capabilities, and ecosystems needed to deliver them efficiently. Bloomberg's reporting suggests Microsoft is beginning to execute that strategy inside its own products.

For Microsoft, every interaction with Copilot consumes computing resources, including inference tokens, GPU capacity, networking, memory, storage, and safety systems. As enterprise adoption grows, even small reductions in per-request costs can translate into substantial operational savings.

A Portfolio of Models

The reported changes also illustrate an architectural shift that is becoming increasingly common across enterprise AI platforms.

Rather than relying exclusively on one foundation model, vendors are assembling portfolios of models optimized for different kinds of work.

Complex reasoning tasks may still require the most capable frontier models from OpenAI or Anthropic. Routine activities, including e-mail assistance, spreadsheet analysis, transcription, summarization, or document generation, can often be handled by smaller, less expensive models without a noticeable difference for end users.

Under that approach, AI platforms increasingly determine which model to use for each request, routing simpler tasks to lower-cost models while reserving more capable models for complex reasoning.

For users, the distinction may become largely invisible.

Cost Is Becoming a Competitive Advantage

The strategy also reflects a broader shift taking place across the AI industry.

For much of the past three years, the AI industry's competitive narrative has focused on which company could build the smartest model. Increasingly, companies are asking a different question: Which model can deliver the required performance at the lowest cost?

Operating costs, particularly inference costs, are becoming a strategic consideration alongside model quality as organizations deploy AI across millions of users.

That trend mirrors Microsoft's own public messaging. In recent months, company executives have argued that enterprise AI success will increasingly depend on deployment efficiency rather than model leadership alone.

The Bloomberg report suggests Microsoft is beginning to apply that philosophy inside its own products.

A Broader Industry Trend

Microsoft has already made MAI models available through GitHub Copilot, and Suleyman has said Microsoft-developed transcription models will begin powering Microsoft Teams and other products in the coming months.

At the same time, Microsoft continues to maintain its close commercial relationship with OpenAI while giving customers access to both Microsoft-developed and third-party models through Azure AI Foundry.

That combination suggests Microsoft is positioning itself less as a consumer of a single AI supplier and more as a platform capable of orchestrating multiple models based on capability, latency, and cost.

For enterprise customers, the implications extend beyond Microsoft.

As AI deployments continue to scale, organizations may increasingly evaluate platforms not simply on which model achieves the highest benchmark score, but on which platform can select the right model for the right workload at a sustainable price.