Google Cloud Next ’26 Unveils Gemini Enterprise, 8th‑Gen TPUs, and New AI Models
Observe the new AI‑Native Cloud and its inference router to reduce cost and latency in production AI workloads.
Explore integrating the Inference Router into your AI workflows to optimize cost and latency.
Summary
The 2026 edition of Google Cloud Next, held in Las Vegas, marked a major push into agent‑centric artificial intelligence. The conference introduced Gemini Enterprise, a unified platform that consolidates Google’s former Agentspace into a single product for building, scaling, governing, and optimizing AI agents. Gemini Enterprise offers a natural‑language interface that allows non‑technical users to create and run agents, while the underlying hardware is powered by the new eighth‑generation Tensor Processing Units (TPUs). TPU 8t training units can scale to 9,600 cores in a super‑pod, and TPU 8i inference units deliver an 80 % improvement in cost‑performance with near‑zero latency, providing the backbone for high‑throughput workloads.
Alongside the hardware, Google announced a suite of data‑and‑security enhancements. The Agentic Data Cloud redesigns the data foundation for agent workloads, and Agentic Defense incorporates Wiz technology to autonomously remediate vulnerabilities across Red, Blue, and Green teams. Workspace Intelligence embeds agents across Gmail, Docs, Sheets, Drive, Meet, and Chat, turning everyday productivity tools into agent‑powered assistants. Seven live demos showcased the platform’s capabilities, from multi‑agent systems and memory addition to debugging at scale and no‑code agent sharing, all available as GitHub repositories and Google Codelabs.
In April 2026, Google expanded its AI portfolio with the release of Gemma 4, described as the most capable open model available. Gemma 4 has already been downloaded over 500 million times and can process more than 16 billion tokens per minute via the API, a significant jump from the 10 billion tokens per minute benchmark of the previous quarter. The announcement also highlighted Deep Research Max for autonomous data synthesis and a new coding tutor in Colab that assists developers in writing code faster. Nearly 75 % of Google Cloud customers now use AI products, and the company noted a sharp rise in conversational search behavior, with AI‑overviews coverage increasing from 36 % to 82 % for B2B technology queries and from 18 % to 83 % for education queries.
These developments signal Google’s intent to position itself as a fully owned, end‑to‑end stack for the “Agentic Era,” integrating custom silicon, data infrastructure, security, and productivity tools. By offering high‑throughput hardware, a versatile agent platform, and powerful open models, Google is equipping developers and enterprises to build next‑generation AI applications that are both scalable and secure.
Key changes
- Five‑layer AI‑Native Cloud architecture unifying compute, storage, networking, inference router, and managed services.
- Inference Router (public preview) routes requests by cost, latency, quality, and residency, eliminating hardcoded model logic.
- Dedicated inference with bring‑your‑own‑model support enables custom and fine‑tuned models on dedicated GPU infrastructure without Kubernetes complexity.
- Expanded model catalog includes 25+ new models, including NVIDIA Nemotron 3 Nano Omni, with day‑0 access and TensorRT‑LLM tuning.
- Managed PostgreSQL & MySQL Advanced Edition, Managed Weaviate, and fully managed RAG Knowledge Base provide hyperscaler‑grade reliability.
- Early customers report 67 % cost reduction (Workato), 2× throughput (Character.ai), and 40 % lower latency (Hippocratic AI).