Mira Murati's AI lab just secured its largest known deal yet. Thinking Machines Lab - founded by the former OpenAI CTO after her surprise September 2024 departure - has signed a multi-billion-dollar agreement with Google Cloud for AI infrastructure, according to a TechCrunch exclusive report.
The deal gives Thinking Machines Lab access to Nvidia's GB300 chips, the latest in Nvidia's Blackwell Ultra line, which are purpose-built for training and running large AI models at high speed. Google Cloud has been aggressively signing compute agreements with AI startups as part of its push to compete with AWS and Microsoft Azure.
The phrasing "deepens" in the announcement suggests Google already had an existing relationship with Thinking Machines Lab before this agreement was finalized - this is not a cold introduction.
For context, Thinking Machines Lab is less than a year old and already attracting infrastructure commitments at a scale that most startups never see. Multi-billion-dollar cloud agreements typically come with minimum spending commitments, which means Thinking Machines Lab is either burning significant compute to train proprietary models, or Google is betting heavily on its future commercial potential, or both.
Murati has kept quiet about what the company is actually building. The lab has hired researchers from OpenAI and other frontier AI companies, but hasn't shipped a public product. Infrastructure deals at this size suggest serious model training work is underway behind closed doors.
For the broader market, this is another data point showing how fierce competition has become to lock in relationships with the next generation of model developers. Google signing a multi-billion-dollar deal with a company that hasn't released a product yet says more about where the compute race is heading than it does about Thinking Machines Lab alone.