Google Cloud leans on infra

Google Cloud is pitching artificial intelligence as an infrastructure sale, not just a model sale, at Cloud Next 2026. (cloud.google.com) In an April 1 guide to Cloud Next sessions, Google put its infrastructure spotlight on “AI Hypercomputer,” Google Kubernetes Engine, Compute Engine and Google Distributed Cloud, with talks led by Mark Lohmeyer, Muninder Sambi and Drew Bradstock. The session list also includes “What’s new for AI on GKE: Training, serving, and agents.” (cloud.google.com) Google’s custom Tensor Processing Units are central to that pitch. Google says its sixth-generation Trillium chips can scale to 256 Tensor Processing Units in one pod, and that Trillium is more than 67% more energy-efficient than Tensor Processing Unit v5e. (cloud.google.com) Google is also tying those chips to managed software that companies already buy. Its Tensor Processing Unit page says customers can run large artificial intelligence workloads through Google Kubernetes Engine or Vertex AI, while Dynamic Workload Scheduler is designed to reserve all needed accelerators at once. (cloud.google.com) That framing comes as cloud vendors compete on operating the full stack for businesses that want artificial intelligence without rebuilding their platforms. Google’s Cloud Next agenda marks 10 years of Google Kubernetes Engine and pairs that anniversary with new sessions on training, serving and agents. (cloud.google.com) The same pattern is showing up in Google’s developer event plans. Google says I/O 2026 runs online on May 19 and 20, with a Google keynote at 10 a.m. Pacific time and a developer keynote at 1:30 p.m. Pacific time on May 19. (io.google, io.google) In its February preview for I/O 2026, Google said the event will cover “agentic coding” and Gemini model updates across Android, Chrome, Cloud and other products. An Android Developers post published April 14 also pointed developers to sessions including “What’s new in Google AI” and “Build core skills to thrive as an AI-era developer.” (developers.googleblog.com, android-developers.googleblog.com) Google is also keeping open models in the mix. On April 9, Google Cloud said Gemma 4 would run on Tensor Processing Units across Google Kubernetes Engine, Compute Engine and Vertex AI, and that parts of the model family would also be offered as fully managed and serverless in Model Garden. (cloud.google.com) The through line is that Google is selling the control plane around artificial intelligence workloads: the chips, the cluster manager, the scheduler and the managed platform. At Cloud Next now and I/O on May 19 and 20, the company is putting those operating layers at the center of its artificial intelligence message. (cloud.google.com, developers.googleblog.com)