Nvidia and Google Cloud stack up

Nvidia and Google Cloud said on April 22 they are packaging chips, models, software and security into a single enterprise artificial intelligence stack for what they call “AI factories.” (blogs.nvidia.com) The announcement came at Google Cloud Next 2026 in Las Vegas, where Google also introduced Gemini Enterprise Agent Platform as its new system for building, deploying and governing artificial intelligence agents. (cloud.google.com 1) (cloud.google.com 2) In plain terms, an AI factory is a data center tuned for training models and running automated software workers at scale, not just storing files or hosting websites. Nvidia said the setup will use Vera Rubin-powered A5X instances on Google Cloud that can scale to nearly 1 million Rubin graphics processing units across sites. (blogs.nvidia.com) Google’s side of the stack centers on Gemini Enterprise Agent Platform, which folds model selection, agent building, orchestration, DevOps and security into one product. Nvidia’s side adds Nemotron models and NeMo software for building and tuning agents on top of that platform. (cloud.google.com) (blogs.nvidia.com) The companies are also pitching the package to customers that cannot send sensitive data to a public cloud region. Nvidia said Gemini models will run on Google Distributed Cloud with confidential computing on Nvidia Blackwell systems, which are designed to keep data encrypted while it is being processed. (blogs.nvidia.com) Google has spent the last year moving from chatbots that answer prompts to agents that can carry out multi-step work across software systems. In March, Google Cloud said companies were shifting from chatbots to agents that automate more complex workflows, and this week it recast Vertex AI into Gemini Enterprise Agent Platform. (cloud.google.com 1) (cloud.google.com 2) Nvidia has been pushing the same idea from the infrastructure side: sell not just graphics processors, but the full recipe around them, including networking, model tooling and deployment software. The Google deal extends that strategy into a cloud platform that already sells its own Tensor Processing Units alongside Nvidia hardware. (blogs.nvidia.com) (blog.google) Google is framing part of the answer as openness. Its Gemini Enterprise agent marketplace page says customers can use Google-made, third-party and in-house agents in one place, and says Google is backing the Agent2Agent protocol so agents can interoperate across models and platforms. (cloud.google.com) That leaves enterprises with a familiar tradeoff: a tighter stack can reduce integration work, but it can also make one vendor pairing feel like the default architecture. The next test is whether customers treat this Nvidia-Google bundle as a fast on-ramp or as the blueprint they build around. (blogs.nvidia.com) (cloud.google.com)