Mustafa Suleyman, CEO of Microsoft AI, stated in a recent interview that superintelligence is near but it will not take people's jobs. He spoke about the company's evolving relationship with OpenAI, the announcement of seven new models at Microsoft Build, and his perspective on AI's current capabilities.
Microsoft's new AI structure
Suleyman explained that Microsoft has restructured its AI operations over the past 15 to 18 months, culminating in a new contract with OpenAI signed in October 2025. The agreement allowed Microsoft to independently pursue superintelligence while continuing to license OpenAI models. Since then, he has assembled a Superintelligence team focused on training frontier models.
"I've been assembling the Superintelligence team, building clusters of sufficient scale to train frontier models, and hiring a team focused on superintelligence," Suleyman said. He noted that this shift enabled him to concentrate entirely on the superintelligence mission.
At Microsoft Build, the company announced seven new models across multiple modalities. Suleyman described this as a culmination of long planning and a relief to be pursuing the frontier.
Relationship with OpenAI
Suleyman addressed the evolution of Microsoft's partnership with OpenAI, which has seen ups and downs. OpenAI has expanded from a research lab into a full-stack company with consumer products, data centers, and chips. Microsoft, meanwhile, decided to build its own models to ensure long-term independence.
"Superintelligence is coming. I think it's just around the corner," Suleyman said. "There's no way that long-term we could be structurally dependent on a third party for providing that IP for all eternity."
He acknowledged that the relationship has been successful for both companies but said partnerships evolve. Microsoft will continue to use OpenAI models while developing its own.
Decision-making framework
Suleyman described Microsoft AI's operating rhythm: six-to-eight-week cycles with a one-week in-person meetup after each cycle. Teams work in squads with a directly responsible individual (DRI) who executes missions. He separates the DRI role from the manager role to allow rotation and flexibility.
Superintelligence and benchmarks
Stay updated
Get the day's AI and automation news in your inbox. No spam, unsubscribe anytime.
Suleyman clarified that he did not claim superintelligence is here, only that it is coming. He pointed to log-linear scaling across modalities, where more compute and data yield consistent improvements on benchmarks. He believes applying more orders of magnitude of compute will lead to models that can invent new knowledge and self-improve.
When asked about measuring AI performance in domains like law versus coding, Suleyman said that chat has reached human-level performance by many measures. He cited real-world use cases such as emotional support, counseling, and coaching as evidence of human-level capability.
"I think many people are having long, meaningful conversations with AIs at human-level performance," Suleyman said. "The quality is exceptionally good. It has very good emotional intelligence."
MAI-Thinking-1 reasoning model
At Build, Microsoft announced MAI-Thinking-1, its first flagship reasoning model. Suleyman highlighted that the model was trained from scratch without distilling from existing models. It scored 97 percent on AIME, a reasoning benchmark, and is on par with Opus 4.6 on coding benchmarks.
Suleyman attributed the model's success to proprietary data curation, stable training runs with few crashes, and high model FLOPS utilization. He emphasized that Microsoft is building a hill-climbing machine that systematically improves against chosen objectives.
"We started right from the top of the stack; we have basically paid for and acquired an extremely high-quality, very conservative set of data," he said. "The methods that you do for that, I think, are actually quite proprietary."
Suleyman explained that distillation from superior models is a shortcut that can fit the model to that distribution but may limit its long-term potential. Microsoft chose to build from scratch to ensure full control and innovation.
Outlook
Suleyman remains optimistic about AI's trajectory. He believes that continued scaling of compute, data, and user interaction will drive further progress. He dismissed concerns about AI replacing jobs, stating that superintelligence will augment human capabilities rather than eliminate them.
Related on Neura Market: