For years Microsoft’s public AI story stayed tightly coupled to OpenAI. In June 2026, VentureBeat reported an interview with Microsoft AI CEO Mustafa Suleyman at Build. He said a contractual change about six months earlier formally allowed his division to pursue “superintelligence” with Microsoft researchers, data pipelines, and custom silicon, while shipping a family of in-house models.
How to read the claim
- Treat “superintelligence” as strategic language, not a shipping SKU name.
- Ask which model family is default for Copilot versus platform offerings.
- Ask about logging and training options for customer prompts.
- Ask how safety evaluation practices apply to Microsoft-trained models.
- Ask exit costs if you later prefer Anthropic, Google, or open-weight hosts.
Multi-sourcing is rational. A single frontier lab concentrates outage, price, and policy risk. Microsoft shipping more of its own models (if sustained) gives Azure customers another path. You still need to compare quality and region controls per workload. Related market pressure: Nadella’s industry-hollowing essay.




