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

Mira Murati, a former executive at OpenAI, is taking significant strides with her startup, Thinking Machines Lab, after securing a landmark multi-billion-dollar agreement with Google Cloud. This deal, estimated to be in the single-digit billions, enables the lab to leverage Google's state-of-the-art AI infrastructure, which is powered by Nvidia's latest GPUs. As part of this collaboration, Thinking Machines will gain access to cutting-edge AI systems built on Nvidia's new GB300 chips. This agreement also encompasses various infrastructure services tailored to enhance model training and deployment, marking a pivotal moment for the young company. Google has been proactively forming partnerships with AI developers to integrate its cloud offerings with complementary services like storage solutions and its Kubernetes engine, as well as Spanner, its database system. Recently, Anthropic, another AI entity, inked a deal with Google and Broadcom to secure substantial tensor processing unit (TPU) capacity—Google's custom AI chips designed for demanding machine learning tasks. However, the competitive landscape is intense, as Anthropic has also secured a new deal with Amazon for additional training capacity. This partnership is particularly significant for Thinking Machines, as it is their first collaboration with a cloud service provider, although it is not exclusive. This suggests that the startup may explore multiple cloud options in the future, highlighting Google's strategy to engage with emerging AI labs early on. Founded by Murati in February 2025 after her departure from OpenAI, Thinking Machines quickly garnered attention, raising $2 billion in seed funding at a valuation of $12 billion. Despite its initial secrecy, the company unveiled its first product, Tinker, in October. Tinker is designed to streamline the creation of custom frontier AI models, and the recent deal sheds light on the lab's ambitious projects. According to Google, the collaboration will facilitate Thinking Machines' reinforcement learning workloads, a cornerstone of their architecture. This training method has been pivotal for recent advancements at renowned labs such as DeepMind and OpenAI. The scale of this Google Cloud agreement underscores the computational demands associated with these innovative endeavors. Thinking Machines is among the first clients to utilize Google Cloud's GB300-powered systems, which promise to double the speed of training and serving compared to older GPU generations. Myle Ott, a founding researcher at Thinking Machines, expressed enthusiasm about the partnership, stating, "Google Cloud got us running at record speed with the reliability we demand."