Why Microsoft is now building its own AI models
What's the story
Microsoft is embarking on a new journey in the artificial intelligence (AI) space. The tech giant is exploring fresh research paths to make its AI models competitive with those of its rivals. This comes after the company signed a revised agreement with ChatGPT parent OpenAI in 2025. The move marks a major shift in Microsoft's strategy, as it looks to become more self-sufficient within this rapidly evolving industry.
Strategic partnership
Initial investment and the exclusivity agreement
Microsoft's first investment in OpenAI back in 2019 was more than just a financial deal. The agreement made Azure, Microsoft's cloud unit, the exclusive cloud provider for OpenAI. In return, Microsoft promised not to compete with OpenAI in developing superintelligence: machines capable of outperforming humans at knowledge work. However, as ChatGPT gained popularity in 2023 and 2024, both companies began reassessing their partnership due to concerns over dependency on a single vendor for such critical technology.
Deal revision
The relaxed agreement and potential dilution of Microsoft's stake
Last fall, as part of an OpenAI restructuring, the original agreement was relaxed. This allowed OpenAI to use other cloud providers and gave Microsoft the freedom to pursue its own version of superintelligence. At that time, Microsoft still owned about 27% of OpenAI. However, this stake may have been diluted after OpenAI's recent $122 billion fundraising round closed.
New initiative
New frontier models and superintelligence unit
Microsoft has launched a new program to create "frontier" models that can compete with ChatGPT, Anthropic's Claude, Google's Gemini, and some Chinese models. This was announced by Mustafa Suleyman, CEO of Microsoft AI. He also revealed a new superintelligence unit at Microsoft in November 2025. The move is part of Microsoft's strategy to own more of its AI technology stack and achieve true self-sufficiency in this space.
Product focus
Microsoft's vision of superintelligence and organizational changes
Microsoft's definition of superintelligence is more product-oriented, focusing on domain-specific systems. Suleyman said in a November blog post that they envision "AI that is carefully calibrated, contextualized, within limits." In March 2026, CEO Satya Nadella reorganized the effort and Suleyman handed over AI services responsibilities. This was done so he could focus on new frontier models.
Resource allocation
Internal resource allocation challenges at Microsoft
An internal AI resource battle came to light during Microsoft's second-quarter earnings call. CFO Amy Hood said Azure sales could have beaten expectations but the company chose to reserve data center capacity for its own services and research. This decision disappointed investors, resulting in a 10% drop in stock value the next day. The trade-off between renting out AI servers and using them internally has also affected Suleyman's project.