Microsoft just poached 3 top AI scientists to ditch OpenAI

Microsoft just pulled off one of the biggest talent raids in AI history. Three of the most influential researchers at the Allen Institute for AI (Ai2) — the nonprofit that built one of the most important open-source AI models in existence — are leaving for Microsoft's secretive Superintelligence team.

The message is hard to miss: Microsoft is done relying on OpenAI alone.

Who Microsoft just hired — and why it matters

The three researchers joining Mustafa Suleyman's team at Microsoft AI are not replaceable hires:

All three will keep their positions at the University of Washington — but their day-to-day research is moving inside Microsoft.

The real story: Microsoft is building AI to replace OpenAI

Microsoft invested over $13 billion in OpenAI. It bet its entire product line on ChatGPT's technology — Copilot in Windows, Office, Bing, GitHub, and Azure all run on OpenAI's models.

But that dependency has become a problem. OpenAI is pushing to restructure as a for-profit company, raising questions about whether Microsoft's deal will hold. And with competitors like Anthropic, Google, and Meta building their own models, Microsoft can't afford to have a single supplier.

Enter Mustafa Suleyman — co-founder of Google DeepMind, now leading Microsoft's in-house AI effort with a team literally called "Superintelligence." These three Ai2 hires are the clearest signal yet that Microsoft is building its own frontier AI models.

The cost to open-source AI

Here's the part that worries researchers: Ai2 isn't just any lab. It was founded by the late Paul Allen — Microsoft's own co-founder — specifically to keep AI research open and accessible to everyone.

Its flagship model, OLMo 3, comes in sizes up to 32 billion parameters and is one of the few AI models where you can see everything — the training data, the code, the evaluation tools. In a world where OpenAI, Google, and Anthropic keep their training recipes secret, OLMo is a rare exception.

Now the people who built it are leaving for a corporate lab with the word "Superintelligence" in the name. The $3.1 billion foundation that funds Ai2 (the Fund for Science and Technology, created from Allen's estate) is also shifting priorities — moving from funding open-source foundation models toward "real-world AI applications."

What this means if you use AI at work

If you use Microsoft 365 Copilot, Windows, or Bing — the AI behind those products may soon change. Microsoft currently licenses GPT-4 and GPT-5 from OpenAI. If Suleyman's team succeeds, Microsoft could swap in its own models without you noticing.

For developers and startups building on OLMo or other Ai2 tools, this is a warning sign. The lead researcher behind the project is now splitting her time with a corporate employer. Open-source AI needs funding and talent — both are being pulled away.

For everyone watching AI: the pattern is clear. Nonprofits and universities develop breakthrough AI. Then big tech hires the researchers, takes the knowledge, and builds proprietary products. It happened with Google hiring Geoffrey Hinton's team from Toronto. It's happening again.

The bigger picture

This hire is part of a broader trend. Microsoft, Google, Meta, and Amazon are all vacuuming up AI talent from academia and nonprofits at unprecedented rates. The result: fewer independent researchers, less open-source progress, and more concentration of AI power in a handful of trillion-dollar companies.

The irony? Paul Allen created Ai2 precisely to prevent this kind of consolidation. His own co-founded company just accelerated it.

Related Content — Get Started with Easy Claude Code | Free Learning Guides | More AI News