Microsoft(MSFT.US) recalibrates its relationship with OpenAI: shifting focus to AI "self-sufficiency" and advancing in-house cutting-edge models

Microsoft (MSFT.US) AI Business Head Mustafa Suleyman said the company is seeking to achieve “true self-sufficiency” in AI by developing high-performance models independently and gradually reducing reliance on OpenAI’s technology.

Suleyman revealed that this strategic shift began after the restructuring of Microsoft’s partnership with OpenAI in October 2025. Since then, Microsoft has started building cutting-edge technology independently rather than continuing to rely on external partners.

To date, Microsoft has invested over $13 billion in OpenAI and owns nearly 27% of the ChatGPT developer.

“We must develop our own foundational models—be at the absolute forefront, have gigawatt-level computing power, and a top-tier global AI training team,” Suleyman said. He is one of the co-founders of Google DeepMind and joined Microsoft in March 2024.

Suleyman stated that Microsoft is heavily investing in integrating and organizing large-scale datasets, which is key to training advanced systems. “This is our core mission to truly achieve self-sufficiency.”

It is understood that Microsoft has previously relied on OpenAI’s models to power its AI services like Copilot. Under a new agreement reached in October 2025, Microsoft’s rights to use intellectual property for products and models have been extended to 2032 and expanded to include related technologies beyond general artificial intelligence, with appropriate safety measures; meanwhile, OpenAI also gains greater freedom to bring in new investors and infrastructure partners.

In addition to OpenAI, Microsoft has invested in AI companies like Anthropic and France’s Mistral. Meanwhile, internal self-developed models are accelerating. Suleyman said that self-developed products are expected to be launched “within this year.”

Targeting “Professional-Grade AGI” for Enterprise Markets

Suleyman said Microsoft aims to capture more enterprise market share by creating “professional-grade general AI.” These AI tools can help knowledge workers such as lawyers, accountants, project managers, and marketers complete daily tasks.

“In the next 12 to 18 months, most tasks performed by white-collar workers at their computers will be fully automated by AI,” he said.

In the next two to three years, AI agents will be able to work more efficiently within large organizations’ workflows. These AI tools will also have continuous learning and evolving capabilities, autonomously executing more tasks.

“Creating new models will become as common as making podcasts or writing blogs,” Suleyman said. “In the future, every organization and individual on Earth will be able to design AI tailored to their needs.”

Intensified Competition in Enterprise Market, Capital Expenditure Under Pressure

However, Microsoft faces competitors in the enterprise market. Anthropic has established a clear first-mover advantage in AI programming tools, and OpenAI and Google are actively competing for enterprise AI contracts.

Microsoft’s capital expenditure for this fiscal year (ending June) is expected to reach $140 billion, mainly to boost AI infrastructure. But investors are concerned that such massive spending may create an AI “bubble” and weigh down the performance of large tech stocks. Over the past month, Microsoft’s stock has fallen about 13%.

“Undoubtedly, we are in an unprecedented era, and the market is trying to understand how the next five years will unfold,” Suleyman admitted. “But everyone is confident that these investments will eventually translate into revenue and profit.”

Healthcare “Superintelligence” and “Human-Centric” Vision

Suleyman said another focus for Microsoft is advancing AI applications in healthcare, aiming to build “medical superintelligence”—using AI to alleviate shortages of medical staff and long wait times. Last year, Microsoft launched an AI-assisted diagnostic tool that reportedly outperforms human doctors in certain tasks.

He also emphasized that Microsoft pursues a “human-centric superintelligence”—AI technology always under human control—to address industry concerns about some companies developing “difficult-to-regulate” advanced AI systems.

“We must redefine standards: only systems that are truly controllable and can operate in obedience to humans should be introduced into the real world,” Suleyman said. “Like any past technology, these tools are designed to enhance human well-being and serve humanity, not to surpass humans.”