Microsoft is trying to reduce its dependency on OpenAI by pursuing self sufficiency with internal systems expected to debut this year. In a recent interview with Financial Times, Microsoft AI Chief Mustafa Suleyman said that the company is seeking “true AI self sufficiency” by developing its own frontier-level foundation models.
A Partnership Quietly Coming Apart
For years, Microsoft and OpenAI had one of the most respected tech partnerships. Microsoft funded the research, OpenAI powered Copilot, Bing and GitHub with both sides benefiting significantly.
In October 2025, Microsoft revised its deal with OpenAI. The new deal now allows Microsoft to develop its own frontier models and eventually pursue its own AGI (artificial general intelligence), while keeping the 27% stake in OpenAI and retaining long-term access to OpenAI’s models and IP through 2032.
In return, OpenAI gained the freedom to seek compute elsewhere. This isn’t a full break from their partnership but a diversified approach towards long-term profitability.
Wall Street Sounded the Alarm
The first signal came during Microsoft’s Q2 FY2026 earnings call when Jefferies analyst Brent Thill revealed that 45% of Microsoft’s $625 billion revenue backlog ties directly to OpenAI. This was particularly alarming because OpenAI is projected to lose $14 billion in 2026 alone.
As a result, investors reacted fast and Microsoft’s stock dropped 12% in a single session, erasing over $440 billion in market value. Bloomberg noted that a partnership that was once celebrated by all is now considered a huge problem.
Therefore, the message to the CEO Satya Nadella and AI Chief Suleyman is unmistakable. A billion dollar dependency on a cash-burning partner is a liability, not a strategy.
How Microsoft Builds Its Way Out
Rather than finding new partners, Microsoft is constructing its own complete AI stack across three fronts.
Firstly, Its MAI-1 model being trained on approximately 15,000 Nvidia H100 GPUs targets the eventual replacement of OpenAI systems inside Copilot.
Secondly, Microsoft’s newly launched Maia 200 chip built on TSMC’s 3nm process aims to slash inference cost across Microsoft’s data centers.
Thirdly, its Fairwater data center network built on a 315-acre flagship site in Wisconsin is housing millions of Nvidia GB200 and GB300 GPUs which gives Microsoft the raw compute power to train its frontier models independently.
Meanwhile, Microsoft already hosts models from Anthropic, Meta’s Llama, Mistral, and xAI on Azure, spreading its bets further.
OpenAI’s Uncomfortable Reality
This development puts OpenAI in a really tight spot. Despite generating roughly $20 billion in revenue and counts 92% of Fortune 500 companies as enterprise users, OpenAI is blowing through its cash reserves at an alarming rate and it now has to deal with its oldest partner becoming a competitive rival.
In addition, the October 2025 deal locks OpenAI onto Azure through 2032 and hands Microsoft 20% of OpenAI’s total revenues over that same period.
In short, OpenAI’s most powerful backer has quietly become its most motivated competitor.
The Breakup is Coming, Just Not Today
The AI Chief Suleyman said it plainly, “We have to build our own foundation models at the absolute frontier”. Still, Microsoft communications chief Frank X. Shaw acknowledged that OpenAI “has a huge role to play” for now. Financially, the two companies remain deeply entangled so a clean break is impossible in the short term.
But strategically, Microsoft’s plan is very clear. The company is already laying the foundation to walk away. How everything unfolds between the two tech giants remains to be seen.
