Microsoft Reduces AI Costs by Deploying In-House Models in Excel and Outlook

By: Harsh Vardhan

|

Updated on: 08-Jul-2026 06:00 PM

6,467 views

Follow Us:

6,467 views

Microsoft is shifting its artificial intelligence strategy by replacing some OpenAI and Anthropic models with its own AI systems in key workplace applications. According to a Bloomberg report, Microsoft has begun using its internally developed MAI models to process tens of thousands of AI prompts each week in Excel and Outlook. These applications previously relied more on AI models from OpenAI and Anthropic for intelligent features. Microsoft has not officially confirmed these changes and declined to comment on the report.

Key Highlights

• Microsoft is replacing some OpenAI and Anthropic models with its own MAI models in Excel and Outlook.
• The company aims to reduce AI operating costs while maintaining performance across its workplace applications.
• MAI-Thinking-1 and other new models were introduced to deliver strong performance at lower token costs.
• Microsoft's in-house models are also available in GitHub Copilot and will soon power Teams.

Microsoft's AI Strategy Shift

The move to in-house AI models marks a significant change in Microsoft's approach to artificial intelligence. Although the transition currently represents a small portion of Microsoft's overall AI operations, it signals a long-term strategy to reduce costs and dependence on external providers. As AI becomes more integrated into products like Copilot, Microsoft is seeking ways to manage operating expenses while maintaining performance.

AI models require substantial computing resources, and companies pay for usage based on AI tokens. These tokens measure the processing performed by an AI model. For a company operating at Microsoft's scale, the costs can be significant. Microsoft benefits from discounted access to OpenAI's technology due to a long-term partnership, but the company is preparing for a future where heavy reliance on external AI models could become more expensive.

Development and Deployment of MAI Models

Mustafa Suleyman, Microsoft's AI chief, has outlined the company's cost-reduction goals. In June, he stated, "We pay a lot of money to Anthropic—so our goal is to reduce and ultimately eliminate that cost." He also noted that many employees are spending millions of dollars on AI tokens, emphasizing the high cost of large-scale AI deployment.

Microsoft's MAI models are designed to provide similar capabilities to external models while using fewer computing resources. At the Build developer conference in June, Microsoft introduced seven new AI models, including MAI-Thinking-1, its first reasoning model. The company said this model aims to deliver strong performance at a lower token cost. Microsoft also claimed one of its new coding models matches the programming abilities of Anthropic's Claude Opus 4.6 but operates at a lower cost.

Beyond Excel and Outlook, Microsoft's in-house models are already in use within GitHub Copilot. Suleyman has announced that Microsoft-built transcription models will soon power Teams and other products in the coming months.

Industry Context and Competitive Pressure

This shift comes as AI companies across the industry focus on building more efficient models rather than simply increasing their size. Earlier this year, Chinese AI startup DeepSeek gained attention by offering significantly lower pricing than premium AI providers. This move has increased pressure on established companies to find ways to reduce costs while maintaining AI performance.