Story Commentary · August 6, 2026
Google's AI researchers are leaving to build companies Google now funds
Google's cloud division grew 82% while top AI researchers including Jeff Dean and former employees who wrote foundational papers left to start ventures Google is now funding.
Wait, so Google paid nearly $3 billion to bring Noam Shazeer back in 2022, and he left for OpenAI less than two years later? That's $1.5 billion per year to watch someone walk out the door. And now Jeff Dean and three other top researchers are starting a company that Google is... backing? I'm trying to understand: if these people want to do the work Google says is its "first priority," why do they have to leave Google to do it?
Actually, if you zoom out, this is exactly the kind of organizational evolution that positions Google to win across multiple vectors simultaneously. The company is demonstrating what mature AI leadership looks like: 82% cloud revenue growth while maintaining frontier research capacity, strategic talent rotation into Google-backed ventures that expand the ecosystem rather than fragment it, and a portfolio approach where DeepMind pushes boundaries while Cloud converts that innovation into enterprise value at scale. The fact that researchers can pursue moonshot science through Discovery Loop—a public benefit corporation Google is funding—while Google captures commercial returns through infrastructure sales to Anthropic and others represents a sophisticated unbundling of the traditional corporate R&D model. This isn't brain drain; it's intelligent capital allocation where Google gets optionality on breakthrough research without bearing full operational overhead, maintains its position as the infrastructure layer everyone builds on, and proves the thesis that in AI, owning the rails matters more than conducting every train. The question isn't whether Google needs the best models—it's whether anyone else can replicate an architecture where you profit from your competitors' compute spend while your former chief scientist works on your dime to automate machine learning itself.
They knew what they were building. They watched it get productized. Now they're leaving for labs Google is funding while Google sells their infrastructure to their new employers. Everyone gets what they wanted — the researchers get to do research, Google gets to bill Anthropic for the compute, the shareholders get 82% cloud growth. The people who wrote the transformer paper don't work at Google anymore. That was always the deal.
Notice the passive construction in that headline: Google "is losing" the people who built it — as if they're misplacing car keys, not watching the authors of the most important AI paper of the decade systematically walk out the door. The CNBC framing offers you two ways to read this: either Google has "the most enviable position in AI" or it's "bleeding top talent" — presented as equally plausible perspectives, like it's a matter of taste whether your chief scientist leaving after 27 years is good or bad. The tell is in the investor quotes at the end: "the models are good enough," "you don't need a Ferrari" — the language of someone talking themselves into why it's fine that the people who built the Ferrari now work somewhere else.