NVIDIA and Google Cloud extend their long-standing collaboration by delivering a full‑stack AI platform designed to take agentic and physical AI from experimental stages into large‑scale production across industries. The partnership centers on next‑generation infrastructure such as A5X bare‑metal instances powered by NVIDIA Vera Rubin NVL72 systems, which provide “up to 10x lower inference cost per token and 10x higher token throughput per megawatt than the prior generation,” enabling clusters that can scale to nearly a million Rubin GPUs for the most demanding AI workloads. Alongside this, a broad portfolio of NVIDIA Blackwell‑based offerings on Google Cloud allows organizations to right‑size acceleration, from fractional GPUs to massive NVL72 rack‑scale configurations, supporting mixture‑of‑experts reasoning, multimodal inference and complex simulations for robotics and physical AI.
Security and data sovereignty are addressed through Gemini models running on NVIDIA Blackwell and Blackwell Ultra GPUs in Google Distributed Cloud, where confidential computing keeps prompts and fine‑tuning data encrypted even from infrastructure operators. Confidential G4 VMs with NVIDIA RTX PRO 6000 Blackwell GPUs extend these protections to multi‑tenant public cloud environments, making high‑performance AI more accessible to regulated industries. On the software side, the platform supports frontier models like Gemini and Gemma as well as NVIDIA Nemotron open models, giving developers tools to build agents that can reason, plan and act. NVIDIA Nemotron 3 Super is integrated into the Gemini Enterprise Agent Platform, while Managed Training Clusters and a new reinforcement learning API based on NVIDIA NeMo RL streamline large‑scale training and customization of open models.
Industrial and physical AI use cases are enabled through NVIDIA Omniverse libraries and NVIDIA Isaac Sim on Google Cloud Marketplace, which allow the creation of physically accurate digital twins and robotics simulation pipelines before deployment. Companies such as CrowdStrike, Snap, Schrödinger and multiple startups leverage this co‑engineered stack to accelerate threat detection, optimize data pipelines, shrink drug discovery cycles and build AI‑native applications, highlighting its impact from emerging ventures to global enterprises. With more than 90,000 developers joining the joint community and NVIDIA receiving Google Cloud Partner of the Year awards, this cloud‑scale platform underpins a transition from prototypes to production systems that secure fleets, automate factories and enable new AI applications.
Reference
Buck, I. (2026, April 22). NVIDIA and Google Cloud collaborate to advance agentic and physical AI. NVIDIA Blog. https://blogs.nvidia.com/blog/google-cloud-agentic-physical-ai-factories/