Google Cloud Expands AI Infrastructure Partnership with Thinking Machines Lab for Frontier Model Training

Google Cloud announced an expanded agreement with Thinking Machines Lab during the Cloud Next '26 conference in Las Vegas. This deal provides the AI research firm with massive capacity on the Google Cloud AI Hypercomputer to accelerate the development of its frontier models. The partnership focuses on utilizing the new A4X Max virtual machines, which integrate the latest NVIDIA GB300 NVL72 architecture.

Thinking Machines Lab will be among the first customers to deploy the NVIDIA Blackwell-powered systems at scale. Early testing shows a twofold increase in training and serving speeds compared to previous GPU generations. This performance boost is supported by Google’s high-bandwidth Jupiter network, which is essential for the rapid weight transfers needed in reinforcement learning.

Beyond hardware, the laboratory is leveraging Google’s integrated AI stack to manage its global operations. Services like Google Kubernetes Engine and Cluster Director help orchestrate massive training jobs while ensuring automated hardware fixes. This allows the team to maintain continuous uptime even when hardware failures occur at the node level.

The deal directly fuels the growth of Tinker, the flagship managed developer platform for model fine-tuning. By abstracting away the complexities of resource recovery and scheduling, Thinking Machines aims to democratize access to efficient model adaptation. The industry can expect the company to release increasingly sophisticated open-weight models that are optimized for scientific auditability.

Mark Lohmeyer, VP at Google Cloud, emphasized that the AI Hypercomputer provides an optimized architecture for the agentic AI era. Founding researcher Myle Ott noted that the seamless integration allows the lab to focus on unique aspects of their stack rather than infrastructure management. This collaboration highlights the growing necessity of rack-scale system optimization for modern AI research.

Thinking Machines Lab has grown rapidly since its founding in early 2025 by industry veterans. After securing a historic seed round, the company has positioned itself as a leader in human-AI collaboration. This latest expansion with Google Cloud ensures they have the compute runway needed to maintain their trajectory.