Exclusive: Google deepens Thinking Machines Lab ties with new multi-billion-dollar deal - TechCrunch

Google's latest multi-billion-dollar cloud deal with Mira Murati's Thinking Machines Lab shows that when you can't build the next big thing yourself, you can always rent the shovels to those who claim they can. The former OpenAI executive's startup, which has been operating in near-stealth mode since its founding last year, will now get access to Google Cloud's AI infrastructure powered by Nvidia's new GB300 chips—the same chips that Google is using for its own Gemini models. It's a classic case of the infrastructure provider hedging its bets: if Murati's team actually builds something revolutionary, Google gets a front-row seat; if they don't, Google still pockets billions in cloud revenue.

This 'single-digit billions' arrangement comes just weeks after Core Automation nerdsniped researchers from Anthropic and DeepMind, suggesting the talent wars have evolved into infrastructure wars where the real winners are the cloud providers selling GPUs by the truckload. Murati, who left OpenAI after reportedly clashing with Sam Altman over safety priorities, now gets to build her vision on Google's dime—or rather, on the dime of Google's investors, who are funding this arms race with cheerful abandon. The fact that this deal was signed while Thinking Machines Lab has yet to publicly demonstrate any working models speaks volumes about the current AI investment climate: promise is enough, delivery optional.

What's particularly delicious is the timing—coming right after Anthropic's Mythos model demonstrated it could find hundreds of Firefox vulnerabilities while itself being vulnerable to basic URL guessing. Google, which has its own security headaches with Gemini, is now providing the compute for another potential security nightmare. The cloud giant gets to position itself as the responsible adult in the room while simultaneously fueling the very frenzy that makes responsible development increasingly difficult. It's the perfect corporate double-play: profit from the chaos while maintaining plausible deniability.