Google Renames Vertex AI as Gemini Enterprise Agent Platform and Launches Agentic Data Cloud at Cloud Next 2026
At Cloud Next 2026, Google overhauled its AI cloud stack: Vertex AI becomes the Gemini Enterprise Agent Platform, a new Agentic Data Cloud offers cross-cloud data access via Apache Iceberg, and the Virgo Network delivers 47 Pb/s to connect 134,000-chip training clusters.
Overview
At its annual Cloud Next conference in Las Vegas on April 22, 2026, Google used its largest cloud event of the year to declare the official arrival of what it calls the “agentic enterprise.” The keynote — themed around moving AI from experimentation into production across entire organizations — delivered three major platform shifts: the replacement of Vertex AI with the Gemini Enterprise Agent Platform, the introduction of an Agentic Data Cloud anchored by a cross-cloud lakehouse, and the unveiling of the Virgo Network, a new megascale data center fabric capable of connecting 134,000 chips at 47 petabits per second.
The announcements came as Google Cloud reported $17.7 billion in Q4 revenue, up 48 percent year-over-year, and disclosed a $240 billion revenue backlog — more than double the figure from the prior year, according to Fortune.
Gemini Enterprise Agent Platform: Vertex AI’s Successor
The centerpiece announcement was the formal rebranding and expansion of Vertex AI into the Gemini Enterprise Agent Platform, available in general availability starting April 22. According to Google’s official launch post, the move is a response to the scale of complexity enterprises face when deploying agents that interact across multiple systems simultaneously, often without adequate security or governance controls.
All existing Vertex AI services will be delivered through the new platform going forward. The addition is not cosmetic: the platform introduces four new capability layers — Build, Scale, Govern, and Optimize — each with distinct enterprise-focused tooling.
On the build side, Agent Studio provides a low-code visual interface for assembling agents, while the Agent Development Kit (ADK) offers a code-first environment with graph-based sub-agent architecture. An Agent Sandbox provides a hardened execution environment for model-generated code, and an Agent Garden ships pre-built templates for use cases such as financial analysis, code modernization, and invoice processing.
For scaling, the platform introduces Agent Runtime with sub-second cold starts, support for multi-day autonomous workflows, a Memory Bank for persistent long-term context, and Agent Sessions that allow mapping of interactions to internal databases or CRMs via custom session IDs, according to the Gemini Enterprise Agent Platform announcement.
Governance capabilities — a category largely absent from earlier AI platforms — include Agent Identity, which assigns unique cryptographic IDs to every agent and produces auditable trails; Agent Registry for centralizing approved tools and skills; Agent Gateway for unified security policy enforcement; and Agent Anomaly Detection for real-time behavioral monitoring.
The platform provides access to more than 200 models through an expanded Model Garden, including Google’s own Gemini 3.1 Pro and Gemini 3.1 Flash Image, open-source Gemma 4, and third-party models from Anthropic, according to The Next Web’s coverage of the event. Google cited more than 6 trillion tokens processed monthly on Gemini models through the Agent Development Kit as a measure of current platform adoption.
Agentic Data Cloud and Cross-Cloud Lakehouse
Alongside the platform rebrand, Google announced the Agentic Data Cloud, a redesign of its data infrastructure to support the throughput and contextual demands of production AI agents. As Andi Gutmans, VP of Data Cloud, said according to SiliconAngle, existing data infrastructures were designed as “static repositories” unsuitable for the real-time, multi-system requirements of agentic AI.
The Agentic Data Cloud rests on three pillars. First, the Knowledge Catalog — an evolution of Google’s Dataplex Universal Catalog — scans organizational documents including accounts, PDFs, presentations, and images to extract entities and relationships, building a navigable schema that helps agents understand company-specific definitions and business logic rather than generic web knowledge.
Second, a suite of Gemini-powered data agents targets specific workflows: a Data Engineering Agent, a Data Science Agent integrated with Claude Code and VS Code, and a Database Observability Agent.
Third, and most strategically significant, is the Cross-Cloud Lakehouse, standardized on Apache Iceberg, which allows AI agents to access data stored in AWS S3 buckets or Azure data lakes as if that data were stored locally in Google Cloud — without migrating it or paying egress fees. The system uses Google’s Cross-Cloud Interconnect alongside Apache Iceberg REST catalog support to enable zero-copy access. According to the SiliconAngle report on data infrastructure, the lakehouse integrates with enterprise platforms including Databricks, Palantir, Salesforce Data360, SAP, ServiceNow, Snowflake, and Workday. Azure support is expected later in 2026; AWS support was available at launch.
The Agentic Data Cloud also includes a Lightning Engine for Apache Spark, which Google claims runs 4.5 times faster than open-source Spark equivalents and delivers two times better price-performance than leading competitors, according to the day-one recap from the Google Cloud blog.
Virgo Network: 47 Petabits Per Second for Training at Scale
Underpinning both the agent platform and the data infrastructure is a new physical network layer called Virgo. As reported by SiliconAngle, Virgo is a megascale data center fabric designed to connect up to 134,000 chips — including the new TPU 8t training processors previously reported by The Machine Herald — at 47 petabits per second of bidirectional bandwidth.
That represents more than four times the bandwidth per accelerator compared to the previous generation. Google achieved the improvement by “flattening” the network topology to reduce the number of layers traffic traverses, and by separating independent switching planes for fault isolation. The system includes deep observability tools and automated routing to handle failures at scale.
Virgo also supports what Google describes as “Fluid Compute,” with CPU instances capable of delivering up to 95 million packets per second — which Google claims is up to 40 percent faster than competing hyperscalers — alongside an Agent Gateway controller that governs protocols including Anthropic’s Model Context Protocol and Google’s Agent2Agent (A2A) standard. A2A reached version 1.2 at the event with the addition of cryptographic signatures for domain verification and, according to The Next Web, had reached production deployment at 150 organizations ahead of the conference.
Workspace Studio and Workspace Intelligence
For business users rather than developers, Google introduced two complementary products. Workspace Intelligence is a new semantic layer that maps emails, chats, files, collaborators, and active projects across the Google Workspace suite into a unified context graph, enabling Gemini-powered agents to synthesize information across apps without manual handoffs.
Workspace Studio is a no-code agent builder that lets business users describe automations in plain language and deploy them across Gmail, Docs, Sheets, Drive, Meet, and Chat. It integrates with third-party platforms including Asana, Jira, Mailchimp, and Salesforce, with rollout to Google Workspace business, enterprise, and education customers. Google separately announced that Meet’s AI note-taking feature has grown 8.5 times year-over-year to more than 110 million monthly users, according to 9to5Google.
What We Don’t Know
Several significant questions remain open. Google did not disclose detailed pricing for the Gemini Enterprise Agent Platform, Agent Runtime, Memory Bank, or the Cross-Cloud Lakehouse beyond standard Google Cloud pricing structures. The timeline for Azure support in the Cross-Cloud Lakehouse — described as “later in 2026” — has no specific quarter or date attached.
Google’s governance features, including Agent Identity and Agent Anomaly Detection, are newly introduced and unproven at production enterprise scale. How they perform against established compliance and audit frameworks (SOC 2, ISO 27001, FedRAMP) has not been detailed.
The platform’s competitive positioning also raises questions. Google CEO Thomas Kurian stated that competitors are “handing you the pieces, not the platform,” framing Google’s full-stack ownership — from TPU silicon through productivity suite — as a differentiator. Whether enterprises with existing investments in AWS Bedrock, Azure AI Foundry, or Anthropic’s managed agents will find Google’s integrated approach compelling enough to justify migration remains the central commercial test for the announcements made this week.
Analysis
The renaming of Vertex AI, combined with the Agentic Data Cloud and Virgo Network, reflects a coherent strategic argument: that the next competitive frontier in cloud is not raw model capability but enterprise infrastructure — governance, data access, memory, and observability — that makes AI agents safe and manageable in production. Google’s $240 billion backlog and 48 percent revenue growth suggest the market is already moving in this direction. The question is whether the Gemini Enterprise Agent Platform, with its expanded governance layer and cross-cloud data access, arrives before or after enterprises have already committed to competing stacks.