VMware Explore 2026 recap

VMware VCF VCF9.1 release Explore VCF9.1.1

VMware Explore 2026 ran August 31 untill September 3 at The Venetian Convention and Expo Center in Las Vegas. A lot of "Explore coverage" floating around right now is really recapping VCF 9.1, which went GA back in May 2026, and vDefend 9.1 features that shipped in release notes and design blueprints months before the event. If we skip those the following announcements came out.

VMware Private AI Cloud: The Umbrella Broadcom Introduced

On day one Broadcom introduced VMware Private AI Cloud as a new umbrella positioning, tying VCF, VMware AI Factory, Tanzu, and its security stack together as a single governed platform for running inference workloads, autonomous agents, containers, and traditional VMs side by side. This is less a product than a new framing for existing and new pieces, but it's the organizing theme for everything below.

VCF 9.1.1: Shipped During Explore, With Real Agentic AI Baked In

Coinciding with the event, Broadcom shipped VCF 9.1.1. The first maintenance release for VCF 9.1 going GA September 3. Two things are worth calling out beyond the usual maintenance-release bug fixes.

AI Assistant for VCF

This part is wat kind of excites me, bringing AI to the VCF admin. It ships in 9.1.1 as an actual configurable capability in VCF Operations. It's a conversational AI chat interface you enable via Extensions > Extension Management (activating the "VMware Intelligent Assist" and "Advanced Configuration" extensions) and it natively supports either VCF Private AI Services (PAIS) or a private Google Gemini instance as the LLM backend, with your choice of model configured directly in the UI.

Three concrete capabilities ship with it:

  • Conversational diagnostics gives you the AI chat-style option to talk to your VCF environment (e.g. "summarize all vCenter Server issues"), with responses that include contextual links back to the actual resources in VCF Operations.
  • Intelligent health visibility correlates proactive health alerts, configuration state, and logs across vSphere and VCF management services to speed up root cause analysis.
  • Management Packs Builder with AI assistance gives admins tooling to build management packs (integrations that pull third-party/adjacent infrastructure data into VCF Operations) without needing to write API integration code themselves.

There's also a new AI Observability view (token usage, average tokens/sec, request latency, skill usage, audit logs) so you can actually monitor what the assistant is doing and costing you.

Practical notes if you're planning to test this: the chat session currently isn't persisted (navigate away and your history is gone) and because the service runs on VCFMS worker nodes, those nodes need outbound connectivity to either the cloud LLM endpoint you configure (no option in air-gapped or restricted environments) or your own LLM running in VCF on PAIS.

Full-stack Kubernetes and agentic-adjacent observability

Alongside the AI assistant, 9.1.1 adds real-time VKS (vSphere Kubernetes Service) observability. This means 2-second metric streaming instead of the previous 5-minute polling, using OpenTelemetry plus multi-cluster VKS monitoring with zero-code OpenTelemetry collection and native Grafana dashboard import. For anyone running agentic workloads or short-lived containerized inference jobs on VKS, this closes a real blind spot: short-lived pods and transient bottlenecks were easy to miss on the old polling interval.

Other 9.1.1 additions worth a mention if you're doing design or upgrade planning: dynamic on-demand identity (real-time AD/OpenLDAP group evaluation instead of pre-provisioned users), expanded certificate lifecycle management (now covering NSX Edges, vSphere Supervisors, License Servers), a new compact form factor cutting CPU/memory footprint up to 40%, and IPv6 support in the license server. There's also a defined patch sequence exception specific to 9.1.1 ,Fleet LCM has to go to 9.1.1 before other components, and VCFMS has to update before Identity Broker/Salt Master.

VMware AI Factory

Announced August 31, VMware AI Factory is a new software-defined layer aimed at compressing the path from bare metal to a running model. It combines VCF with certified AI ReadyNodes from Cisco, Dell, Lenovo, Supermicro and others, plus new automation for AI-ready infrastructure deployment and Day 2 operations. According to Broadcom it is taking deployment time from "weeks to hours."

During this week a new partnership with MetalSoft was announced. It brings heterogeneous bare-metal automation into VCF , provisioning or repaving physical servers from multiple vendors directly through the VCF management console. With MetalSoft you get a cutting bare-metal provisioning from weeks to minutes and removing the need for vendor-specific firmware/hardware tooling.

AgentMinder

Broadcom unveiled AgentMinder, a new governance and runtime-control platform for autonomous AI agents, generally available the same day. It gives agents enterprise identities bound to a declared mission, approved tools, and authorized resources, enforces policy at each tool call via a gateway, and logs sessions through OpenTelemetry for audit and anomaly detection. This is a genuinely new product, not an extension of something previously shipped and worth tracking if agentic workloads are anywhere on your roadmap.

New Agentic AI Security Capabilities for vDefend and Avi

Separate from the vDefend 9.1 feature set that shipped earlier in the year, Broadcom announced new agentic-AI-specific security capabilities at Explore: vDefend can now discover MCP servers, LLMs, datastores, and tools running in the environment, and flag unauthorized "shadow AI" deployments. This is additive to (not a restatement of) the vDefend 9.1 release that was already GA.

AI-Ready Data Foundations for Tanzu Platform

This announced pairs governed AI agents with governed data: a pipeline that takes multimodal data, adds semantic context, applies governance and lineage tracking and once ready publishes it as curated data products for agent consumption. It comes with an isolated credential store so agents can't directly access raw credentials, aimed at reducing credential theft and prompt-injection risk.

Leading AI Models on VCF

Broadcom announced model-as-a-service style availability of validated third-party models directly on VCF. Models named were Google Gemma, Alibaba Qwen, NVIDIA Nemotron, NEC cotomi, and Z.ai GLM , and their goal is the support for 150+ models overall as part of the Private AI Cloud push.

NSX Networking: What Was Previewed in Sessions

This one is important to frame correctly. NSX networking itself did not get a new release at Explore. VCF 9.1 already went GA back in May and this is the version the sessions were actually built around. What Explore added was depth: breakout sessions and workshops that walked through the VCF 9.1 networking model in more technical detail than the original release notes did.

Worth knowing if you're doing design work on top of the current GA release:

  • Virtual network appliance (know as VNA) for centralized NAT and load balancing. No NSX edge required and will be required if you want these services while using EVPN.
  • Transit gateway span and VLAN-backed subnets. Gives finer control over workload placement.
  • Enhanced Direct Path. Near-bare-metal network throughput without giving up vMotion or DRS. This is the part that matters most: SR-IOV normally forces that trade-off and Enhanced Direct Path is built to avoid it.
  • AI agent isolation at the network layer.
  • A first look at an agentic operations layer for NSX itself. Not AI for the workloads running on top of the network but AI helping run the network day to day.

Treat this section as "already shipped in VCF 9.1, explained properly at Explore" rather than a new announcement. If you want the feature-level detail confirmed against your own environment go to the VCF 9.1 release notes directly instead of the session recordings.

TrueSource: Expanded Open-Source Supply Chain Coverage

Broadcom also announced expanded coverage in TrueSource, its curated catalog of secure, verifiably-built open-source libraries and images, adding coverage for Java, Python, and Node.js ecosystems. Relevant if you're doing any software supply-chain hardening work alongside VCF/Tanzu deployments.

What's next?

Two things worth looking into.

  • AI Assistant for VCF in 9.1.1 seems worth lab-testing before deciding on a Gemini vs. PAIS-backed rollout, especially given the outbound-connectivity requirement and the lack of chat persistence in this first release.
  • The new vDefend/Avi shadow-AI discovery capability. Very intereseting to discover if you already have AI workloads running unmanaged in a VCF environment.

The other announcements are a bit further away from myself, being a VCF Architect and Admin. From a Private Cloud development view these will contribute into a more complete platform competing with the public cloud.

The last "big" announcement at Explore was the confirmation for another Explore next year, but with new dates and a new venue! VMware Explore will move to Resorts World Las Vegas with new spring dates of May 3–6 2027.