Former OpenAI executive Mira Murati's startup, Thinking Machines Lab, has inked a multibillion-dollar agreement with Google Cloud to expand its use of cutting-edge AI infrastructure. The deal includes access to Google’s latest systems built on Nvidia’s GB300 chips and other infrastructure services for model training.

Google has been aggressively pursuing cloud deals with AI developers as it integrates its AI computing offerings with storage, Kubernetes and Spanner. Earlier this month, Anthropic secured a similar agreement with Amazon and Broadcom to train models using tensor processing units.

This deal marks Thinking Machines Lab’s first venture into the cloud services realm, indicating Google's proactive strategy in locking in fast-growing frontier labs early on. Murati founded the company in February 2025 following her role at OpenAI, and it has since raised a $2 billion seed round to launch its proprietary AI tool, Tinker.

The scale of this deal reflects the significant computational resources required for reinforcement learning workloads, a key area that Thinking Machines is exploring. The GB300-powered systems from Google Cloud offer a 2x improvement in training and serving speed compared to previous-generation GPUs, according to Murati.