Google Cloud Next '26 Opening Keynote
Summary
Google Cloud CEO Thomas Kurian opened Next '26 by declaring the end of the AI pilot phase, noting that nearly 75% of Google Cloud customers now leverage AI products in production. He introduced the "Agentic Enterprise Blueprint" — a unified five-layer stack designed to help businesses deploy AI at enterprise scale.
Sundar Pichai shared Google's internal AI transformation data: 75% of new code at Google is AI-generated (up from 50% the prior fall), a complex code migration was completed 6x faster using a three-role agent system (planners, orchestrators, coders), marketing teams achieved 70% faster campaign turnaround and 20% higher conversion using Gemini, and security operations agents now triage tens of thousands of threat reports monthly, reducing threat mitigation time by over 90%.
The keynote's centerpiece was the Gemini Enterprise Agent Platform, described as "mission control for the agentic enterprise." Key capabilities include a low-code Agent Studio for building agents via natural language, an Agent Registry for centralized governance, a Skills and Tools Registry for reusable modular packages, native Model Context Protocol (MCP) integration exposing all GCP services as MCP servers, agent-to-agent orchestration with both generative and deterministic patterns, zero-trust verification with cryptographic agent identities, and Agent Observability for full execution tracing. The platform also announced support for Anthropic's Claude Opus 4.7 alongside existing Claude Opus, Sonnet, and Haiku models. A notable partnership was revealed: Apple is using Google Cloud as a preferred provider to develop next-generation Apple Foundation Models powered by Gemini technology for future Apple Intelligence features.
On the infrastructure front, Google announced eighth-generation TPUs split into two specialized variants — TPU8T for training (2.8x improvement over Ironwood, 121 exaflops FP4 per pod, scaling to 9,600 chips) and TPU8I for inference (9.8x performance over Ironwood, 1,152 chips per pod, near-zero latency). Google also announced it will be among the first to offer Nvidia's Vera Rubin NVL72, a new Virgo network linking 134,000 chips with 47 petabits/second of bandwidth, and managed Lustre storage with up to 10 terabytes/second throughput.
The Agentic Data Cloud introduced four innovations: the Knowledge Catalog (a universal context engine that auto-enriches unstructured data using Gemini), the Data Agent Kit (integrating AI skills into VS Code, Claude Code, and Gemini CLI), the Lightning Engine for Apache Spark (2x price performance over the previous market leader), and the Cross-Cloud Lakehouse (borderless analytics over data in AWS, Azure, or Google Cloud using Apache Iceberg, with zero egress fees).
Agentic Defense announced the formal integration of Wiz into Google Cloud, introducing an AI Application Protection Platform with red/blue/green security agents for autonomous vulnerability detection, triage, and remediation. The mean time to exploit a vulnerability has dropped to minus-7 days (exploits before patches), and handoff between threat actors now takes 22 seconds, underscoring the urgency of machine-speed security.
Customer stories included: Virgin Voyages reducing production timelines by 60% and achieving a 28% month-over-month sales increase; Signal Aduna reaching 80% employee AI adoption with 11,000 agents built and 400% surge in weekly users; KPMG hitting 90% adoption with 100+ agents in one month; Walmart deploying Gemini Enterprise to store leaders via Pixel Fold devices; and Citadel Securities running workloads 2-4x faster at 30% lower cost using TPU Ironwood. Google Workspace announced migrations from Microsoft 365 are now up to 5x faster, and the new Workspace Intelligence layer eliminates context fragmentation across Docs, Sheets, Slides, Gmail, and Chat.
Key Takeaways
- •The pilot phase is over — 75% of Google Cloud customers are now running AI in production, and the primary challenge has shifted to scaling AI across the entire enterprise, not experimenting with it.
- •Google internally generates 75% of new code with AI (up from 50% six months prior) and completed a major code migration 6x faster using a multi-agent system with planner, orchestrator, and coder roles — a repeatable blueprint for enterprise migrations.
- •The Agentic Enterprise requires a unified stack, not fragmented point solutions: chips designed for models, models grounded in data, agents built on models, all secured by infrastructure.
- •Every GCP service is now exposed as an MCP server, enabling any external agent or tool to orchestrate Google Cloud infrastructure directly via the Model Context Protocol.
- •Security must operate at machine speed: the mean time to exploit a vulnerability is now minus-7 days (exploits before patches exist), and threat actor handoff takes 22 seconds — requiring autonomous AI security agents, not human-speed SOC teams.
- •Cross-Cloud Lakehouse breaks the assumption that your analytics engine and data must live in the same cloud — organizations can query AWS S3 or Azure data from BigQuery with zero egress fees using Apache Iceberg open standards.
- •Agent identity and governance are now non-negotiable at scale: every agent should have a cryptographic ID, auditable authorization policies, and centralized management through an agent gateway.
- •The Knowledge Catalog auto-enriches unstructured data (PDFs, images, video) the moment it lands in storage, making it agent-ready without manual data engineering pipelines.
- •TPU specialization has diverged: TPU8T is optimized for training (2.8x over Ironwood, scales to 9,600 chips), while TPU8I is optimized for inference and reinforcement learning (9.8x over Ironwood, near-zero latency at scale).
Transcript
I've got dungeon master's I've got 12sided die. I've got kitting there for me. Yes, I do. I do. I've got posters on the wall. My favorite rock crew kiss in the god. In the cross I feel safe. No one else is in the ground where I belong. No one hears me. No one hears me. No one hears me. No one hears me. No one hears me. Please welcome to the stage CEO of Google Cloud, Thomas Curran. Hello, >> welcome everyone to Google Cloud Next. Just one year ago, we stood on this same stage and promised a new ...
