Thinking Machines Lab, co-founded by former OpenAI executive Mira Murati, has entered into a significant multi-billion-dollar agreement with Google to enhance its utilization of Google Cloud's artificial intelligence infrastructure. This partnership will leverage advanced systems powered by Nvidia's cutting-edge GPUs, as reported by TechCrunch.
This deal, valued in the single-digit billions, grants Thinking Machines access to Google's latest AI technologies, including those built on Nvidia's new GB300 chips. It also encompasses essential infrastructure services to facilitate model training and deployment.
Google is actively pursuing collaborations with AI developers to integrate its cloud offerings with various services, such as storage solutions, a Kubernetes engine, and its database product, Spanner. Recently, Anthropic also forged a partnership with Google and Broadcom to secure substantial tensor processing unit (TPU) capacity, showcasing the competitive landscape in AI cloud services.
Despite the fierce competition, this agreement marks the first collaboration between Thinking Machines and a cloud services provider, although it is not exclusive. This flexibility allows Thinking Machines to explore partnerships with multiple cloud providers in the future, indicating Google's strategy to engage with rapidly evolving AI labs early on.
Founded in February 2025, Thinking Machines quickly gained attention by raising a $2 billion seed round, reaching a valuation of $12 billion shortly thereafter. The company has maintained a level of secrecy around its developments but recently launched its inaugural product, Tinker. This innovative tool automates the creation of custom frontier AI models.
The recent agreement with Google sheds light on Thinking Machines' focus areas. According to Google, the partnership will support the startup's reinforcement learning workloads, a critical aspect of Tinker's architecture. Reinforcement learning has been pivotal in driving breakthroughs at leading AI labs, including DeepMind and OpenAI, and the scale of this collaboration reflects the computational demands of such advanced work.
Thinking Machines is among the first to utilize Google Cloud's GB300-powered systems, which promise a twofold increase in training and serving speeds compared to previous GPU generations. Myle Ott, a founding researcher at Thinking Machines, expressed enthusiasm about the collaboration, stating, "Google Cloud got us running at record speed with the reliability we demand."