Microsoft is reportedly expanding the use of its in-house developed artificial intelligence (AI) models to reduce reliance on OpenAI and Anthropic.
Bloomberg News, citing sources familiar with the matter, reported on July 7 that Microsoft has begun utilizing its 'MAI' model in key productivity software such as Excel and Outlook. According to the sources, tens of thousands of AI prompts are processed weekly through the MAI model in these programs, which previously relied more heavily on OpenAI and Anthropic models.
While the MAI model currently represents a small portion of overall AI usage, Bloomberg noted that this move indicates progress in Microsoft's efforts to build a competitive AI model at a lower cost.
Microsoft is using a significant amount of AI tokens in its major products, including the AI assistant Copilot. Tokens are units that measure AI computational usage. Currently, Microsoft benefits from discounted access to a substantial portion of related technology through its long-term partnership with OpenAI, but it appears to be seeking to avoid excessive dependence on prices set by external AI research labs in the long run.
Mustafa Suleyman, head of Microsoft's AI models, stated in June that the company aims to increase the use of the MAI model to reduce spending on Anthropic. He remarked, "We are paying a lot of money to Anthropic," adding, "Our goal is to reduce that cost and ultimately eliminate it."
At its annual developer conference, Build, held in June, Microsoft announced seven new AI models. Some of these models reportedly match the coding capabilities of Anthropic's previous generation model, Claude 4.6, while being more cost-effective.
Microsoft's MAI model is also available in GitHub Copilot, an AI-based software development service. Suleyman indicated that the in-house developed model is expected to be applied to video conferencing app Teams and other products in the coming months.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.