Nvidia + Google Cloud push enterprise AI

Nvidia and Google Cloud said on April 22 they are expanding their partnership so companies can build larger AI systems on Google’s cloud and in their own data centers. (blogs.nvidia.com) At the center of the announcement is Google Cloud’s A5X instance, a server offering built on Nvidia’s Vera Rubin platform that Nvidia said can scale to nearly 1 million Rubin graphics processing units, or GPUs. Google and Nvidia have also been previewing support for Rubin NVL72 systems in Google Cloud infrastructure this year. (blogs.nvidia.com) (cloud.google.com) A GPU is a chip built to handle many calculations at once, which makes it useful for training and running artificial intelligence models. Google Cloud already sells GPU-backed AI Hypercomputer systems for dense clusters, and its documentation says those systems are designed for performance-optimized AI and machine learning workloads. (docs.cloud.google.com) (cloud.google.com) The new pitch is not only bigger clusters but more controlled ones. Nvidia said customers will be able to run Gemini on Google Distributed Cloud with Blackwell and Blackwell Ultra GPUs, which keeps Google-managed AI services closer to on-premises data for customers with residency or compliance requirements. (blogs.nvidia.com) (cloud.google.com) Google’s own distributed-cloud product has been moving in that direction for months. In August 2025, Google said Gemini on Google Distributed Cloud was available for customers that wanted to use the model on-premises rather than only in a public cloud region. (cloud.google.com) Nvidia and Google also tied the hardware push to software for so-called agentic AI, which refers to systems that can plan and carry out multistep tasks. Nvidia said its Nemotron models and NeMo software are being integrated with Google’s Gemini Enterprise Agent Platform. (blogs.nvidia.com) (docs.cloud.google.com) Security is a second part of the enterprise sales pitch. Google Cloud documentation says customers can run confidential virtual machines with attached GPUs, and Nvidia said the partnership adds confidential computing support for Blackwell-based workloads. (docs.cloud.google.com) (blogs.nvidia.com) That matters for companies that want to use proprietary documents, code, or customer records in AI systems without sending raw data into a standard shared environment. Google and Nvidia are effectively selling a mix of public-cloud scale, on-premises deployment, and hardware-level isolation in one stack. (cloud.google.com) (blogs.nvidia.com) (docs.cloud.google.com) Nvidia’s separate enterprise infrastructure push shows the same theme at a smaller scale. In a technical blog published April 22, Nvidia said its RTX PRO 4500 Blackwell Server Edition with vGPU 20 software is aimed at virtualized enterprise workloads and delivered nearly 1.9 times the graphics acceleration of previous Nvidia architectures in the company’s testing. (developer.nvidia.com) The through line is that Nvidia and Google are no longer talking only about training giant models in remote clusters. They are packaging AI as production infrastructure that can span cloud regions, private facilities, and regulated environments using the same Nvidia hardware and Google software layers. (blogs.nvidia.com) (cloud.google.com)