Microsoft starts replacing Claude and ChatGPT models to cut its AI bill

Microsoft has reportedly begun replacing OpenAI and Anthropic models with its own AI in Excel and Outlook as it looks to cut soaring AI costs. The move suggests the company's growing push to build cheaper, in-house AI without relying heavily on external providers.

In Short

• Microsoft is using its MAI models for thousands of AI prompts in Excel and Outlook every week
• The company wants to reduce its dependence on expensive AI models from Anthropic and OpenAI
• Microsoft recently introduced seven new AI models designed to deliver strong performance at a lower cost

Microsoft appears to be taking a major step toward reducing the soaring cost of artificial intelligence by gradually replacing some OpenAI and Anthropic models with its own in-house AI systems across popular workplace apps. According to a Bloomberg report, the company has started using its internally developed MAI models to handle tens of thousands of AI prompts every week in Microsoft Excel and Outlook. These applications had previously relied more heavily on AI models from OpenAI and Anthropic for several intelligent features. Microsoft has not officially confirmed the report and declined to comment.

The change is still relatively small compared to Microsoft's overall AI operations, but it shows a change in strategy. As AI becomes deeply integrated into products such as Copilot, Microsoft is looking for ways to keep operating costs under control while reducing its dependence on outside AI providers.

Microsoft wants cheaper AI without compromising performance

AI models consume enormous computing resources, with companies paying for usage based on AI tokens — the units that measure how much processing an AI model performs. For a company operating AI services at Microsoft's scale, these costs can quickly add up. While Microsoft currently benefits from discounted access to OpenAI's technology through its long-term partnership, the company is also preparing for a future where relying heavily on external AI models could become much more expensive.

Microsoft AI chief Mustafa Suleyman has made the company's intentions clear. Speaking in June, he said, "We pay a lot of money to Anthropic—so our goal is to reduce and ultimately eliminate that cost." He also noted that "many, many people in our organization are spending millions of dollars" on AI tokens, highlighting just how expensive large-scale AI deployment has become.

The company's MAI models are designed to deliver similar capabilities while requiring fewer computing resources. At its annual Build developer conference in June, Microsoft introduced seven new AI models, including MAI-Thinking-1, its first reasoning model. The company said the model focuses on delivering strong performance at a lower token cost.

Microsoft also claimed one of its new coding models can match the programming capabilities of Anthropic's Claude Opus 4.6 while operating at a lower cost. Beyond Excel and Outlook, Microsoft's in-house models are already available inside GitHub Copilot. Suleyman has also said that Microsoft-built transcription models will soon power Teams and other products in the coming months.

The move comes at a time when AI companies across the industry are racing to build more efficient models instead of simply making them larger. Chinese AI startup DeepSeek drew global attention earlier this year by offering significantly lower pricing than premium AI providers, putting additional pressure on established players to reduce costs.

- Ends