At Google Cloud Next 2026, NVIDIA and Google Cloud announced a deepened long-term partnership, fully integrating new GPU infrastructure, confidential computing, and Google's restructured enterprise-grade intelligent agent technology stack to jointly advance the construction of physical AI factories. Google is also shifting its enterprise AI strategy focus from general chat tools to autonomous intelligent agents, and unifying Vertex AI and related services under the Gemini Enterprise brand.
I. Physical AI Factories Powered by Rubin Computing Architecture
At the infrastructure level, Google Cloud announced that it will launch A5X bare metal instances based on NVIDIA's Vera Rubin NVL72 system in the second half of 2026. NVIDIA stated that this architecture can scale to 80,000 Rubin GPUs in a single cluster and up to 960,000 GPUs in multi-cluster environments across regions, enabling Google Cloud to support ultra-large-scale AI training and inference environments, rather than traditional enterprise-level AI pilot projects.
The core significance of this scale lies not in the numbers themselves, but in the shared understanding of industry needs: AI applications are shifting from isolated model hosting to highly coupled, integrated infrastructure encompassing agents, simulation, robotics, and digital twins. As NVIDIA previously disclosed when deepening its industrial software collaboration, this technology stack is increasingly geared towards engineering and manufacturing processes, rather than being limited to chatbot scenarios in data centers.
II. Physical AI Factories Also Need Secure Deployment
Beyond computing power scale, the two companies are also expanding deployment models. NVIDIA announced that Google will pilot Gemini models on Google Distributed Cloud based on NVIDIA Blackwell/Blackwell Ultra GPUs to meet the needs of customers requiring cutting-edge models to operate close to sensitive data. Google also launched a confidential G4 virtual machine powered by NVIDIA RTX PRO 6000 Blackwell server GPUs, now available for global preview, providing hardware-level security for regulated industries or multi-tenant AI workloads.
III. Agents, Models, and Industrial Workflows
Software-level upgrades are equally crucial. At this conference, Google launched the Gemini Enterprise Agent platform, serving as a unified environment for building, governing, and scaling agent deployments. NVIDIA, on the other hand, deeply integrated its Nemotron 3 Super model and NeMo-based training tools into this technology stack. This provides industrial users with a complete path from model development to production deployment, supporting applications in robot simulation, digital twins, vision systems, and workflow automation.
Currently, this collaboration focuses more on platform synergy, and no products directly targeting end users have been launched. However, it clearly demonstrates the strategic positioning of both parties: Google Cloud serves as the operating system layer for enterprise agents, while NVIDIA provides the underlying computing power and model tools for next-generation physical AI factories.
Tags：