Breaking away from OpenAI dependency
Microsoft has started public testing of its own foundation AI model, MAI-1-preview, signaling a strategic shift toward reducing its reliance on OpenAI. Though the two companies remain closely tied through a $13 billion partnership, Microsoft is laying the groundwork for more autonomy in its AI stack.
The software giant announced Thursday that MAI-1-preview is now live on LMArena, a platform where users can interact with and evaluate language models. Microsoft says it will begin integrating MAI-1 into certain Copilot use cases in the coming weeks to gather user feedback and fine-tune performance. A request form has also been released for developers seeking early access.
A growing rivalry with OpenAI
Microsoft and OpenAI’s partnership remains strong — with OpenAI using Microsoft’s Azure cloud infrastructure to run its models — but competition is mounting. OpenAI has started working with other cloud providers like CoreWeave, Google, and Oracle to meet growing demand. Additionally, Microsoft now lists OpenAI among its competitors, alongside Amazon, Google, Apple, and Meta.
OpenAI’s ChatGPT assistant reaches 700 million users weekly, and while Microsoft relies on its models for Bing and Windows 11 features, it’s clear the tech giant wants more control over its future AI roadmap.
MAI-1-preview: Microsoft’s first full-scale model
MAI-1-preview is Microsoft’s first end-to-end foundation model developed entirely in-house. Previously, the company had only built smaller open-source models under the Phi name. Mustafa Suleyman, head of Microsoft’s AI division, emphasized the company’s ambition to reach billions through its own models. The new model was trained using approximately 15,000 Nvidia H100 GPUs, and Microsoft is now running it on a cluster that includes Nvidia’s high-end GB200 chips.
Despite the high ambitions, the model ranked 13th for text performance on LMArena, trailing competitors like OpenAI, Anthropic, Mistral, DeepSeek, and xAI. Still, the foundation is now set for Microsoft to build proprietary AI into Copilot, Office, Windows, and other products.
A team built for AI independence
Microsoft’s AI division has expanded rapidly. Led by Suleyman — a co-founder of DeepMind and former CEO of Inflection — the team includes several former DeepMind engineers and members from Inflection. This group is tasked with building Microsoft’s internal AI capacity and reducing reliance on third parties.
While Microsoft continues to benefit from its OpenAI collaboration, the launch of MAI-1-preview represents a long-term pivot: from strategic partner to parallel competitor in the AI space. As demand for enterprise-grade AI grows, Microsoft appears determined to build not just infrastructure, but also the brainpower behind its future tools.
