Microsoft Develops Its Own AI Reasoning Models

In a notable strategic transition that indicates a shift from its singular dependence on OpenAI, Microsoft is now in the process of creating its own artificial intelligence (AI) reasoning models, termed ‘MAI’ internally. This initiative, initially highlighted by The Information, aims to bolster Microsoft’s AI capabilities while lessening its reliance on the creator of ChatGPT.

Exploring AI Alternatives for Microsoft 365 Copilot

Microsoft has started experimenting with AI models from xAI, Meta, and DeepSeek as potential substitutes for OpenAI’s technology in its leading AI product, Microsoft 365 Copilot. This diversification strategy comes despite Microsoft’s substantial investment of $13.75 billion in OpenAI since 2019, emphasizing its broader objectives in the AI sector.

The Creation of MAI Models

The MAI models are specifically developed to address intricate reasoning and problem-solving challenges, achieving performance levels that are reported to be on par with those of OpenAI and Anthropic. Mustafa Suleyman, who leads the AI division at Microsoft, has managed the completion of a series of these models. As per The Information, Microsoft is employing chain-of-thought techniques, an advanced AI reasoning method that produces responses with intermediate reasoning steps, to train its models.

Significantly, Microsoft is already trialing these models within Copilot, substituting OpenAI’s technology in certain cases. The company is also contemplating the release of MAI as an application programming interface (API) later in the year, enabling external developers to integrate Microsoft’s AI functionalities into their applications.

Minimising Dependence on OpenAI

Microsoft’s initiative to create internal AI solutions aligns with a broader strategy to diversify its AI infrastructure. At the launch of Microsoft 365 Copilot in 2023, the product was prominently advertised as utilising OpenAI’s GPT-4 model. However, over time, Microsoft has sought out alternative AI models for three key reasons: to eliminate exclusive dependency, to lower costs, and to increase processing speeds.

This shift follows earlier revelations from Reuters in December 2024, indicating that Microsoft was combining both internal and external AI models within Copilot. While Microsoft continues to maintain a significant partnership with OpenAI, the dynamics of their collaboration have transformed. A noteworthy change occurred when Microsoft ended its exclusive status as OpenAI’s cloud provider, transitioning instead to a ‘right of first refusal’ agreement, which enables OpenAI to explore cloud services from other vendors, including Oracle.

As Microsoft advances its AI capabilities, the launch of MAI could have a substantial effect on the wider AI landscape. Should Microsoft manage to implement its models across its range of products and offer them to developers, it could alter the competitive dynamics in the AI sector, challenging OpenAI’s supremacy while providing businesses with greater flexibility in integrating AI-driven solutions.

Neither Microsoft nor OpenAI has responded to media inquiries regarding these developments, as reported by Reuters.