In the inconspicuous office buildings of Redmond, an AI project is being developed that could change the power dynamics of the entire industry.
Over the past five years, Microsoft has invested more than $13 billion in OpenAI, becoming the face of the AI revolution and positioning Copilot as the central product of its future strategy. But while the public eagerly awaits each new GPT release, a remarkable strategic shift is unfolding in the background: Microsoft is quietly building its own AI model family called MAI (Microsoft Artificial Intelligence)—a project that could be far more than just a backup plan.
The quiet paradigm shift
The key figure in this impending change is Mustafa Suleyman. The former DeepMind co-founder and former CEO of Inflection carries the burden of a $650 million acquisition and the expectations of a tech giant on his shoulders. Since March 2024, as CEO of Microsoft AI, he has led one of the most ambitious changes in recent corporate history.
"For a company that has invested over $14 billion in OpenAI and enjoys privileged access to groundbreaking models, you would assume Microsoft would be perfectly content to continue relying on OpenAI for its AI ambitions," explains an industry observer familiar with its internal workings. "But the reality is different."
This reality is manifested in a new project: MAI-1, an AI model with an estimated 500 billion parameters that can already compete with the leading models from OpenAI and Anthropic in internal tests. It embodies Microsoft's strategic goal of becoming more independent in the AI landscape and regaining complete control over its technological future.
Tensions in Silicon Valley
The relationship between Microsoft and OpenAI has suffered noticeably in recent months. Particularly telling was an incident in which OpenAI refused to share technical details about its o1 model with Microsoft engineers—despite the massive investment and close partnership between the companies.
The recent renegotiation of the partnership, which now allows OpenAI to move some workloads from Microsoft's Azure platform to other cloud providers, also signals a shift in the balance of power. Microsoft is clearly preparing for a future in which it no longer wants to rely exclusively on OpenAI.
"The tensions are real," confirms a person familiar with the negotiations. "It's like in a long-term relationship where both partners suddenly realize that they may have different goals after all."
The Suleyman Empire
Under Suleyman's leadership, Microsoft has begun building an impressive team. In an unusual move for the industry, several former Google DeepMind employees have been recruited for a new consumer health division under Microsoft AI, including Dominic King, who previously led Google DeepMind's health division.
"Suleyman isn't building a team—he's building an empire," notes a Silicon Valley insider. "It's a clear statement: Microsoft wants to be not just a buyer of AI technology, but a leading innovator."
The flagship of this new empire is MAI-1. With its 500 billion parameters, it may at first glance appear smaller than OpenAI's GPT-4 with over a trillion parameters, but internal tests show that it can certainly keep up in terms of reasoning skills—the ability to solve complex problems and demonstrate human-like analytical intelligence.
A portfolio approach
However, Microsoft isn't pursuing a binary replacement strategy. Instead, the company is experimenting with a diverse portfolio approach, testing technologies from Anthropic, Meta, DeepSeek, and even Elon Musk's xAI alongside its own MAI models.
The Redmond labs are currently testing various scenarios in which OpenAI's models in Microsoft Copilot could be replaced by these alternatives. It's a calculated chess game in which Microsoft is slowly but surely reducing its dependence on a single vendor.
"By focusing on advanced features like improved reasoning capabilities, Microsoft is demonstrating its determination to stay at the forefront both technologically and economically," explains one industry analyst. "It's no longer just a race for the most parameters, but for the most intelligent and useful models."
The path to public availability
Although Microsoft has not yet announced a specific timeline, reports indicate that the company plans to make its MAI models available as an API to external developers later this year. This would position Microsoft not only as a consumer but also as a provider in the AI technology market—a fundamental shift in its role within the AI ecosystem.
Such an opening could have far-reaching effects: from faster innovation cycles and more intense competition to potentially lower costs for AI technologies. It could also lead to democratization by giving companies access to more adaptable AI solutions.
The new balance of power
While Microsoft officially emphasizes that its relationship with OpenAI remains strong and collaborative, a more nuanced picture is emerging behind the scenes. It is a picture of a strategic repositioning in which Microsoft is diversifying and securing its future AI strategy.
"Microsoft is playing the long game," explains a former Microsoft executive. "They've learned the lessons of previous technological disruptions and don't want to be on the wrong side of history this time."
The development of MAI clearly signals the beginning of a new chapter in the history of artificial intelligence. In a world where AI is increasingly becoming the centerpiece of technological innovation, Microsoft is strategically positioning itself to be not only a consumer but also a leading producer of this transformative technology.
The coming months will show whether MAI truly has the potential to compete with the leading models from OpenAI and Anthropic. One thing is certain, however: the AI race is far from settled or linear, and Microsoft is determined not just to participate, but to win.