Mira Murati, a former executive at OpenAI, has launched a new startup named Thinking Machines Lab, which has recently secured a multi-billion-dollar agreement to leverage Google Cloud’s advanced AI infrastructure. This deal, reportedly valued in the lower billions, grants Thinking Machines access to Google’s state-of-the-art AI technology, utilising Nvidia’s latest GB300 GPUs to enhance model training and deployment capabilities.
Google has been actively forming partnerships with AI developers to integrate its cloud services with other offerings, including storage solutions and database products. Notably, earlier this month, Anthropic, another AI company, entered a similar agreement with Google and Broadcom to gain substantial tensor processing unit (TPU) capacity for their machine learning tasks. However, competition remains intense as Anthropic also inked a partnership with Amazon to secure additional processing capacity for their AI model, Claude.
This marks the first time Thinking Machines has collaborated with a cloud services provider, following a prior partnership with Nvidia that included investment from the chipmaker. The new agreement is non-exclusive, allowing the startup to engage with other cloud providers as needed. Nevertheless, it showcases Google’s strategy to establish early relationships with emerging AI labs.
Murati transitioned from her role as OpenAI’s chief technologist to establish Thinking Machines in February 2025. The company raised $2 billion in seed funding shortly after its inception, achieving a valuation of $12 billion. While operating under a veil of secrecy, Thinking Machines unveiled its first product, Tinker, in October. This innovative tool automates the creation of custom AI models and relies on reinforcement learning, a training method linked to significant advancements at leading AI labs such as DeepMind.
The details of this latest deal highlight the capabilities Google Cloud can provide to support Thinking Machines’ projects, specifically regarding reinforcement learning workloads. This training approach has been instrumental in recent breakthroughs within the AI field, underscoring the computational demands associated with such work. The scale of this agreement further indicates how crucial high-performance computing will be moving forward.
Thinking Machines stands out as one of the pioneering customers for Google Cloud’s GB300 systems, which promise a twofold improvement in training and serving speeds compared to earlier GPU generations. Myle Ott, a founding researcher at the startup, expressed enthusiasm about the partnership, stating that Google Cloud has enabled them to achieve record operational speeds with the necessary reliability.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
