Microsoft is reportedly developing its own advanced AI reasoning models, known internally as ‘MAI,’ to decrease reliance on OpenAI and enhance its AI capabilities. The Information reports that these models are designed for complex reasoning and problem-solving, with the aim of matching or exceeding the performance of leading models from OpenAI.
Microsoft’s MAI models are aimed to rival OpenAI
The company is experimenting with integrating the MAI models into products such as Microsoft 365 Copilot and is considering releasing them as an application programming interface (API) for external developers later this year. This could enable developers to incorporate Microsoft’s AI functionalities into their own applications, potentially offering cost savings and increased flexibility compared to solely using OpenAI’s technology.
This initiative follows reports from December 2024, suggesting that Microsoft, despite investing $13.75 billion in OpenAI since 2019, is exploring alternatives to ChatGPT’s AI model, particularly for its 365 Copilot. Initially, the Microsoft 365 Copilot—launched in 2023—heavily featured OpenAI’s GPT-4 model. However, Microsoft’s strategy appears to be shifting for several reasons: to avoid monopolistic dependency, reduce costs, and improve operational speed.
NASA’s new tool uses Microsoft AI to unlock data for all
Microsoft’s AI division, under the leadership of Mustafa Suleyman, has completed training a family of models that reportedly perform comparably to leading models from OpenAI and Anthropic on common benchmarks. The team is specifically focusing on reasoning models utilizing chain-of-thought techniques, which allow for intermediate reasoning when addressing complex problems.
In addition to the MAI models, Microsoft is testing AI models from xAI, Meta, and DeepSeek as potential replacements for OpenAI’s technology in Copilot. This development coincides with the recent introduction of AI features in Microsoft 365 for Personal and Family subscription plans, which were previously available only through a separate Copilot Pro subscription requiring an additional fee of $20 per month.
Earlier in 2024, Microsoft adjusted its partnership with OpenAI, moving away from exclusively providing cloud services. The revised accord maintains a ‘right of first refusal’ (ROFR) for Microsoft, allowing it the first opportunity to supply cloud resources to OpenAI, while giving OpenAI the liberty to explore services from other cloud providers like Oracle.
This transition in the Microsoft-OpenAI partnership aligns with the announcement of the Stargate Project, a collaborative effort including OpenAI, Oracle, SoftBank, and Nvidia, which is expected to invest up to $500 billion in infrastructure development for next-generation AI models in the United States. Concurrently, Microsoft plans to invest $80 billion in AI-focused data centers and cloud initiatives in the fiscal year 2025, along with $3 billion allocated to projects in India.
Suleyman’s team has been progressing with the development of the MAI models, which are larger than an earlier series of Microsoft models called Phi. These models are designed to potentially replace OpenAI’s models within Copilot.
Microsoft is contemplating the release of the MAI models as an API later this year, allowing outside developers to incorporate these capabilities into their applications.
Featured image credit: Kerem Gülen/Ideogram
