Nvidia, Google Cloud partner

Nvidia and Google Cloud said this week they are expanding their alliance to sell more of Nvidia’s newest artificial intelligence systems through Google’s cloud. (blogs.nvidia.com) The companies said Google Cloud will offer A5X bare-metal instances built on Nvidia Vera Rubin NVL72 systems, and those deployments are designed to scale to nearly 1 million Rubin graphics processing units across data centers. (blogs.nvidia.com) (nvidia.com) A graphics processing unit, or GPU, is the chip that handles the heavy math behind training and running AI models, and cloud providers rent those chips to companies that do not build their own data centers. Google Cloud already offers A4 virtual machines with Nvidia B200 chips and A4X virtual machines with Nvidia GB200 NVL72 systems. (cloud.google.com) (blogs.nvidia.com) Google and Nvidia framed the new push around “agentic” AI systems that can carry out multistep tasks and “physical” AI used in robotics, manufacturing, drug discovery and energy. They also said Gemini models will run on Google Distributed Cloud with Nvidia Blackwell and Blackwell Ultra hardware. (blogs.nvidia.com) The announcement came during Google Cloud Next in April 2026, where Google also introduced its Gemini Enterprise Agent Platform and new Tensor Processing Unit chips. That puts Nvidia’s hardware alongside Google’s in-house AI processors instead of replacing them. (crn.com) (blogs.nvidia.com) Nvidia said the Rubin-based A5X systems are aimed at lowering inference cost per token and raising token throughput per megawatt versus the prior generation. Inference is the stage where a trained model answers prompts, and it has become the main cost center as chatbots and coding tools handle more user traffic. (blogs.nvidia.com) (nvidia.com) Google also said it is widening access to Blackwell capacity, including confidential virtual machines with Blackwell GPUs and a broad lineup that runs from full rack-scale systems to smaller fractional instances. That gives cloud customers more ways to buy premium Nvidia hardware without reserving an entire cluster. (blogs.nvidia.com) (storagenewsletter.com) The partnership extends a pattern from the past year: Google has been one of the first cloud providers to roll out Nvidia’s newest AI systems, including preview access to GB200-based A4X machines announced in 2025. Nvidia is now using that same channel to push Rubin into commercial cloud deployments. (cloud.google.com) (blogs.nvidia.com) For customers building AI products, the practical change is simple: more of Nvidia’s top-end chips are moving into Google’s cloud, and Google is promising larger clusters and lower serving costs as those systems arrive. (blogs.nvidia.com)