Introduction: For over a decade, NVIDIA and Google Cloud have co-engineered a full-stack AI platform that spans every layer—from optimized libraries to enterprise cloud services. At Google Cloud Next in Las Vegas, they unveiled a new wave of innovations designed to bring agentic and physical AI from the lab into production. From next-generation GPU instances to advanced AI frameworks, these developments empower developers, startups, and enterprises to build smarter agents and digital twins. Here are seven key breakthroughs you need to know about this evolving partnership.
1. A Decade of Co-Engineering: The Foundation of AI Innovation
NVIDIA and Google Cloud’s collaboration began over ten years ago, focusing on performance-optimized libraries and enterprise-grade cloud services. This partnership has built a robust AI platform that enables everything from large-scale training to real-time inference. The latest milestones at Google Cloud Next expand this foundation to support the next wave of agentic and physical AI workloads. By combining NVIDIA’s GPU leadership with Google Cloud’s scalable infrastructure, they’ve created a seamless environment for developers to move from research to production—whether managing complex workflows or deploying robots on factory floors.
2. Next-Generation Infrastructure: A5X Instances Powered by Vera Rubin
Google announced A5X bare-metal instances built on NVIDIA Vera Rubin NVL72 rack-scale systems. Through extreme codesign across chips, systems, and software, these instances deliver up to 10x lower inference cost per token and 10x higher token throughput per megawatt than previous generations. A5X leverages NVIDIA ConnectX-9 SuperNICs and next-generation Google Virgo networking, scaling to 80,000 Rubin GPUs in a single cluster and up to 960,000 in a multisite setup. This infrastructure is ideal for running the largest AI workloads with optimized performance and sustainability.
3. Comprehensive Blackwell GPU Portfolio for Any Workload
Google Cloud’s NVIDIA Blackwell portfolio spans from A4 VMs with HGX B200 systems to A4X VMs with GB200 NVL72 and A4X Max with GB300 NVL72, plus fractional G4 VMs using RTX PRO 6000 Blackwell Server Edition GPUs. Customers can right-size acceleration—from one-eighth of a GPU to tens of thousands of Blackwell GPUs interconnected via NVL72 racks. This flexibility allows teams to train, tune, and serve everything from frontier models to agentic and physical AI workloads, balancing cost, performance, and sustainability.
4. Gemini on Google Distributed Cloud with Blackwell GPUs
In preview, Google Gemini runs on Google Distributed Cloud powered by NVIDIA Blackwell and Blackwell Ultra GPUs. This enables enterprises to deploy AI at the edge with low latency and data privacy. The integration brings advanced language models to remote locations, supporting applications like real-time analytics and autonomous systems. By combining Gemini’s multimodal capabilities with NVIDIA’s acceleration, Google Cloud offers a secure, scalable solution for industries ranging from manufacturing to healthcare.
5. Confidential VMs with NVIDIA Blackwell GPUs for Secure AI
Google Cloud introduced confidential virtual machines (VMs) powered by NVIDIA Blackwell GPUs. These VMs encrypt data in use, protecting sensitive workloads during training and inference. They are designed for financial services, healthcare, and government sectors where data privacy is critical. With hardware-based isolation, users can run agentic AI and physical AI models without exposing proprietary data to cloud infrastructure. This milestone strengthens trust in cloud-based AI by ensuring end-to-end security.
6. Agentic AI with Nemotron Open Models and NeMo Framework
Building on the Gemini Enterprise Agent Platform, NVIDIA contributes Nemotron open models and the NeMo framework to enable advanced agentic AI. Developers can build agents that manage complex workflows, reason across data, and take actions autonomously. NeMo provides tools for customizing and deploying large language models, while Nemotron offers efficient model architectures. This combination allows enterprises to create intelligent assistants that interact with enterprise systems, supporting tasks like customer service, supply chain optimization, and decision support.
7. Physical AI and Digital Twins: From Lab to Factory Floor
The partnership extends to physical AI, enabling robots and digital twins on factory floors. Using NVIDIA’s simulation platforms and Google Cloud’s compute power, companies can design, test, and deploy autonomous machines in virtual environments before real-world use. Digital twins—virtual replicas of physical systems—allow continuous optimization of manufacturing processes. This accelerates the adoption of AI in industrial automation, reducing costs and improving efficiency. The integration of agentic and physical AI creates a seamless bridge between digital intelligence and tangible outcomes.
Conclusion: The NVIDIA-Google Cloud partnership continues to push the boundaries of AI infrastructure, from the massive scale of Vera Rubin to secure confidential computing and edge deployment. These seven breakthroughs demonstrate a commitment to making agentic and physical AI accessible and practical across industries. As the next decade unfolds, this collaboration will remain a cornerstone for building intelligent systems that transform how we work, manufacture, and innovate.