Google's Control‑Plane Pitch

Google heads into Cloud Next this week selling an “AI control plane” — a layer for running, securing and tuning agents — more than a single model brand. (siliconangle.com) Google Cloud Next 2026 runs April 22-24 at Mandalay Bay in Las Vegas, and Google’s official event page says the conference will focus on generative AI, infrastructure and security. Google opened registration in December and has used the event to tee up hands-on labs and keynote announcements. (googlecloudevents.com) (cloud.google.com) In cloud computing, a control plane is the management layer: it decides where workloads run, what data they can reach and which security rules apply. Google already has pieces of that stack in Vertex AI Agent Engine, which its documentation says lets developers deploy, manage and scale AI agents in production. (docs.cloud.google.com) Google also has an Agent Development Kit, or ADK, that walks developers from building an agent in Python to deploying it on Vertex AI Agent Engine Runtime. The quickstart shows Google is pushing not just model access, but a full path from code to production service inside Google Cloud. (docs.cloud.google.com 1) (docs.cloud.google.com 2) That packaging comes after Google spent the last year tying AI software more tightly to its own hardware. At Cloud Next 2025, Google introduced Ironwood, its seventh-generation Tensor Processing Unit, and said it was the first TPU designed specifically for inference, the stage where a trained model answers prompts. (blog.google) Google’s current TPU documentation says TPU7x, the Ironwood generation, is the latest TPU available on Google Cloud and is built for large-scale training and inference. A single platform pitch that spans agent tools, security controls and TPU tuning gives Google a way to argue that customers should buy the whole stack together. (docs.cloud.google.com) That can simplify deployment for large companies that want one vendor to handle orchestration, access controls and hardware optimization. It can also make startups more dependent on Google-specific services if they build agents with ADK, run them on Agent Engine and tune them for Google’s TPU fleet. (siliconangle.com) (docs.cloud.google.com) Google is not alone in pushing that message. GitLab said on April 14 that its Duo Agent Platform would run on Vertex AI models and count usage against existing Google Cloud commitments, showing how partners are already plugging into Google’s managed agent stack rather than treating models as stand-alone products. (tmcnet.com) The test in Las Vegas is whether Google announces enough concrete tools, prices and customer deployments to make “control plane” sound like a product instead of a theme. If it does, the company’s pitch will be that buying AI from Google means buying the operating layer around it too. (siliconangle.com)