A4X Max VM with NVIDIA GB300 GPU and multiple Google Cloud services accelerate model exploration and training
las vegas, April 22, 2026 /PRNewswire/ — Cloud Next’26 — Google Cloud today announced a new agreement to provide expanded AI infrastructure capabilities and capacity to Thinking Machines Lab. This new agreement expands Thinking Machines’ footprint on Google Cloud, accelerating research, platform development, and training of frontier models using AI hypercomputers.
Thinking Machines leverages A4X Max and NVIDIA Blackwell architecture through Google Cloud, serving as one of the first Google Cloud customers to take advantage of NVIDIA GB300 NVL72. In early testing, Thinking Machines has seen 2x faster training and services using A4X Max VMs compared to previous generation GPUs. This performance is powered by Google Cloud’s Jupiter network, which enables the near-instantaneous weight transfer required for TML reinforcement learning workloads.
In addition to NVIDIA’s highly optimized compute, Thinking Machines also benefits from proximity to Google Cloud’s integrated and open AI stack. Today, the company uses services such as Google Kubernetes Engine, Spanner, Cluster Director, Cloud Storage, and Anywhere Cache for large-scale orchestration to help build its frontier model and fine-tuning product, Tinker.
By combining Cloud Storage, Spanner for transactional metadata, and a custom node-level caching solution, Thinking Machines enables its infrastructure to support continuous training while processing production workloads on a global scale.
“By leveraging A4X Max and our AI hypercomputer integration stack, Google Cloud can now run at record speeds with the reliability we demand,” said Miles Ott, founding researcher at TML.. “By seamlessly integrating high-performance computing, fast storage, GKE orchestration, and automated remediation with Cluster Director, we can now focus on the unique aspects of our stack like Tinker and reinforcement learning.”
Mark Lohmeyer, vice president and general manager of AI and compute infrastructure at Google Cloud, said: “Through this new agreement and our deep partnership with NVIDIA, we will help further accelerate Thinking Machines using Google Cloud’s AI hypercomputers, which bring together purpose-built hardware, open software, and flexible consumption models in an optimized architecture.”
“As models grow in size and reinforcement learning workflows become more complex, system-level optimization becomes critical,” said Ian Buck, vice president and general manager of hyperscale and HPC at NVIDIA. These advances running on top of the stack make the platform stronger, faster and smarter, enabling TML to extend and build on what researchers around the world are creating at NVIDIA.”
Thinking Machines began working with Google Cloud in 2025.
About Google Cloud
Google Cloud offers a powerful and optimized AI stack, including AI infrastructure, leading models such as Gemini, data management capabilities, multicloud security solutions, developer tools and platforms, and agents and applications to help organizations transform their businesses for the agent era. Customers in more than 200 countries and territories rely on Google Cloud as their trusted technology partner.
Source Google Cloud
