Google Cloud Next 2026 Recap: What Actually Happened

TL;DR: Google Cloud Next 2026 ran April 22-24, 2026 at the Mandalay Bay Convention Center in Las Vegas, not April 8-10 as some listings had it. Thomas Kurian keynoted, Sundar Pichai opened. The headline reveal was the Gemini Enterprise Agent Platform, a full rebrand of Vertex AI. Google also shipped 8th-generation TPUs, an Apache Iceberg lakehouse, and folded the $32 billion Wiz acquisition into a combined security platform. One attendee put the crowd near 40,000 in a venue built for fewer.

I track Google Cloud Next every year for one reason. What gets renamed and re-architected on that keynote stage in April tends to ship as a real feature inside the AI SaaS tools I review six to twelve months later. Vertex AI becoming the Gemini Enterprise Agent Platform is exactly that kind of announcement. Most readers here will never touch Google Cloud’s console, but the agent tooling Google previewed is already showing up in vendor pitch decks. That is why Google Cloud Next sits on my AI events hub alongside every other event I track for the same reason.

The dates on this page were wrong for months

Google Cloud Next 2026 ran April 22-24, 2026, at the Mandalay Bay Convention Center, 3950 South Las Vegas Boulevard, Las Vegas, Nevada. Not April 8-10, which is what this page’s ACF data carried until recently. Google Cloud’s official Day 1 recap blog is dated April 23, 2026, with embedded photos timestamped 0422. Pichai’s companion post is dated April 22. The 7-highlights wrap-up is dated April 24. A handful of third-party listings add April 21 for pre-conference partner activity outside the core schedule.

Next year is already locked. The official Next event FAQ confirms Google Cloud Next 2027 runs April 13-15, 2027, back at Mandalay Bay (checked 2026-08-24, googlecloudevents.com).

Detail	Value
Dates	April 22-24, 2026 (April 21 for some partner activity)
Venue	Mandalay Bay Convention Center, Las Vegas, NV
Format	Hybrid: in-person plus free on-demand Next Digital
Keynote	Thomas Kurian, Google Cloud CEO; Sundar Pichai opened
Attendance	Not officially confirmed; 30,000-35,000 pre-event, ~40,000 per one attendee
Next edition	April 13-15, 2027, same venue (checked 2026-08-24)

Gemini Enterprise Agent Platform is the announcement worth reading

Vertex AI is gone. Google fully rebranded it as the Gemini Enterprise Agent Platform and says all Vertex AI services and roadmap route exclusively through the new stack. That is not a cosmetic rename. New pieces include Agent Studio for low-code building, an upgraded Agent Development Kit (ADK) that natively supports Model Context Protocol and works across Gemini, Claude, and other models, a re-engineered Agent Runtime backed by a persistent Memory Bank, and governance layers named Agent Identity, Agent Registry, and Agent Gateway.

Direct model access on the platform now includes Gemini 3.1 Pro, Gemini 3.1 Flash Image (Nano Banana 2), Lyria 3 for audio, and Anthropic’s Claude Opus 4.7 running inside a Google-run platform. That last part is worth naming out loud. Google Cloud shipping Anthropic’s flagship model as a first-class tenant is a real signal about how much Google trusts the demand for model choice at the enterprise layer.

If you build on what used to be Vertex AI, this is not optional reading. The migration conversation is coming.

Kurian argued for one connected stack. Pichai brought the receipts.

Kurian’s core argument from the stage: enterprises need chips designed for models, models grounded in your data, agents and applications built with those models, and the whole stack secured by the infrastructure. Kurian’s framing organized every announcement of the week into one story rather than five separate product lines. It is also a defensible position because Google actually built each of those layers this year.

Pichai brought the internal numbers to back it up in his companion post on blog.google. Google’s first-party models process more than 16 billion tokens per minute through direct customer API use, up from 10 billion the prior quarter. 75% of new code written at Google is AI-generated and engineer-approved, up from 50% the prior fall. Gemini Enterprise’s paid monthly active users grew 40% quarter-over-quarter in Q1 2026 (checked 2026-08-24, blog.google).

The 8th-generation TPU family splits into two purpose-built chips. TPU 8t for training scales to 9,600 TPUs and 2 petabytes of shared memory per superpod at triple the processing power of the prior Ironwood generation. TPU 8i for inference connects 1,152 TPUs per pod at 80% better performance per dollar. Google Cloud is also among the first offering NVIDIA’s Vera Rubin NVL72 systems, alongside its own Arm-based Axion N4A processors, now generally available.

Data platform got a real architectural shift. Cross-Cloud Lakehouse (formerly BigLake) standardizes on Apache Iceberg and lets customers query data sitting in AWS or Azure without copying it. A zero-copy federation layer extends into Palantir, Salesforce, SAP, ServiceNow, and Workday. Knowledge Catalog (formerly Dataplex) becomes a Gemini-powered context graph mapping business meaning across a company’s data estate.

The $32 billion Wiz acquisition, closed March 2026, showed up fully integrated in the keynote and kept its own brand name instead of being folded into generic Google Cloud Security. Three new SOC agents launched inside Google Security Operations: Threat Hunting, Detection Engineering, and Third-Party Context. Google Cloud Fraud Defense arrived as the next evolution of reCAPTCHA, built to tell legitimate AI agents apart from malicious bots.

Workspace Intelligence ties Docs, Sheets, Slides, Gmail, Drive, and Chat together with an “Ask Gemini in Chat” feature that summarizes projects and drafts documents without leaving the window. Gemini Enterprise for Customer Experience shipped a Shopping agent and a Food Ordering agent, already live at Papa John’s.

I did not attend this year, and the attendance gap in the record is worth naming

Google has not published a confirmed final attendance number. Pre-event marketing projected a range, with some listings citing 30,000 (the 2024 baseline) and conference-prep content pegging 35,000+. The most specific number came from an attendee actually on the ground, not from Google’s marketing. AtScale’s independent post-event recap opened by describing “the 40,000 people crammed into a venue built for significantly fewer of them,” and described gridlocked hallways, session rooms filling 20 minutes early, and bathrooms, food service, and escalators “buckling under the load.”

That is one attendee’s estimate, not an official headcount. It also lines up with a repeatedly mentioned complaint from multiple attendees about the venue struggling with this crowd. If you sat this year out and are weighing 2027, the venue reality is worth pricing in alongside the keynote content.

Who this event actually mattered for

Platform engineers on Vertex AI got the most directly actionable news of the week. The transition to Gemini Enterprise Agent Platform is not optional, so anyone with production workloads on the old naming needs to understand Agent Studio, Agent Runtime, and the new governance layer regardless of whether they attended.

Data architects and data engineering leads had a strong reason to be in the room for the Cross-Cloud Lakehouse news. The bet on Apache Iceberg and the new zero-copy federation into Salesforce, SAP, ServiceNow, and Workday reshape a multi-year data strategy.

Security teams evaluating a cloud-native SOC stack should care most about the Wiz integration and the three new Google Security Operations agents. A $32 billion acquisition folded into a keynote demo without a rebrand is a signal worth reading.

Founders building AI agents generally, not specifically on Google Cloud, still got real value from the MCP-everywhere platform stance. As one independent recap put it: “every Google service is now MCP-enabled by default. That is not a feature, that is a platform stance.” The same thread runs through my MCP server leaderboard coverage.

Pure end-user Workspace admins with no infrastructure decisions on their plate got very little that required being on-site. Workspace Intelligence rolls out to tenants automatically.

My verdict splits by role

I get asked this every year, and the honest answer is that it depends on what you build. For the roles the event is actually built for, Google Cloud Next 2026 delivered real, usable substance. Even a critical independent recap called the Gemini Enterprise Agent Platform “the most impressive thing at the show, full stop.”

For everyone else, the verdict is less flattering to the in-person format. Gridlocked hallways, 20-minutes-early room fills, and infrastructure buckling under the load is a crowd-management failure, not just Vegas being Vegas. If the choice is between fighting that crowd for three days or watching the free Next Digital replay, the replay is the better use of your time unless your job specifically requires hallway conversations.

One other honest limitation. Independent coverage described the customer testimonials as “vague, fulsome, and interchangeable, the kind of quotes that make you wonder if anyone at those companies has actually put the product into production.” The technical announcements are independently verifiable and real. The customer-success theater on top deserves the same skepticism I apply to any tool’s own case studies before recommending it in a hands-on review.

For Google Cloud Next 2027 (April 13-15, 2027, back at Mandalay Bay), the eight months of lead time and Google’s already-public dates give you real prep runway. If Kurian carries the “agentic era” framing forward the same way, the Agent Registry and Agent Gateway pieces are what to watch for at the enterprise governance layer.