Microsoft is pulling away from OpenAI. Mustafa Suleyman, head of AI at Redmond, confirmed to the Financial Times a major strategic shift toward “total self-sufficiency” in artificial intelligence.
This repositioning follows a restructuring of the ties between the two partners in October. Specifically, Microsoft will now develop its most advanced technologies in-house, rather than relying on an external partner.
“We must develop our own foundation models, at the cutting edge of technology, with gigawatt-scale computing power and some of the world’s best AI training teams,” says Mustafa Suleyman, the former cofounder of Google DeepMind who joined Microsoft in 2024.
Microsoft is investing heavily in assembling and organizing the vast data sets required to train advanced systems. According to Suleyman, the first internally developed models are expected to be launched “in the course of the year.”
A $140 Billion Bet
Until now, Microsoft relied on OpenAI’s models to power its AI tools, notably the Copilot software assistant. The deal negotiated last year granted Microsoft an approximate $135 billion stake in the ChatGPT creator and access to its most advanced models through 2032.
But this arrangement also gives OpenAI more freedom to seek out new investors and infrastructure partners, potentially turning it into a direct competitor.
To mitigate risk, Microsoft has diversified its investments in other model developers such as Anthropic and Mistral, while accelerating the development of its own solutions.
The stated ambition: conquer the enterprise market with a “professional-grade AGI”—AI tools powerful enough to perform the daily tasks of knowledge workers. “Most white-collar tasks in front of a computer—lawyers, accountants, project managers or marketers—will be fully automated by AI within 12 to 18 months,” Suleyman predicts.
Microsoft plans $140 billion in investments for its fiscal year ending in June, primarily to build the infrastructure required for AI.
Beyond the corporate sphere, Microsoft is targeting health with the aim of building a “medical superintelligence” capable of helping to address staffing shortages and wait times in overcrowded health systems. Last year, the publisher unveiled a diagnostic tool that would outperform doctors on certain tasks.
