Tracking enterprise infrastructure's evolution in the AI era

In-Depth

VCF AI Assistants, Gated Lab Licenses and a New Vegas Venue: My Highlights of VMware Explore 2026

• By Tom Fenton
• 09/10/2026

I kicked off VMware Explore at the VMUG WuPaaS Community Meetup at the Yahoo! Sportsbook at the Venetian the night before the conference officially began. It was good to wind down and relax after the flight to Vegas.

BTW, for the uninitiated, WuPaaS stands for Warm-Up Party as a Service. It is an unofficial opening community pre-party held the night before the official start of VMworld/Explore. It is a laid-back gathering designed to let participants catch up before the conference begins.

It was very well attended, and a shout-out goes to Evolia, Vali Cyber, and The Redesign Group for sponsoring the event and, more importantly, for showing their support for the VMUG community.

Explore 2026
Explore officially kicked off Monday morning, August 31. It was weird not to have the event kick off the morning of the first day with a keynote, as Broadcom retimed its keynotes to the afternoon of the first day.

But the afternoon keynote allowed me to attend a few sessions and interview a few vendors beforehand. I will discuss some of the sessions that I attended after I recap the keynote.

Keynote: Shaping the Future of Private AI Cloud and Agentic Innovation
In a prior post, I discussed the announcements made on the morning of Explore's first day. However, VMware expanded on those announcements and provided additional details during its keynote.

The keynote began with news that Broadcom is partnering with MetalSoft to provide bare-metal automation for the VMware AI Factory. Ram Velaga, president of Broadcom's Infrastructure Software Group, said this will enable them to deliver integrated heterogeneous bare-metal automation for VMware Cloud Foundation (VCF) and the newly launched VMware AI Factory. He then went on to say that IT supply chain constraints are not a passing phase and that hardware will remain a scarce, costly resource for the foreseeable future. He also said that the primary culprit driving enterprise server budgets through the roof is memory.

This set the stage for Paul Turner, chief product officer for VCF, to take the stage.

Paul said to combat rising memory costs, VCF 9.1 integrates NVMe-based memory tiering, which he said can deliver up to a 42 percent reduction in hardware expenses. He then showed a slide in which Broadcom demonstrated a 50/50 DRAM-to-NVMe split managed directly within ESXi. As I detailed in a recent article, using memory tiering, the hypervisor abstracts the storage interface so that pooled NVMe storage appears directly to the operating system as system memory. Turner went on to note that VCF 9.1 optimizations can yield up to 39 percent in overall server cost savings. For enterprises building local AI sandboxes and other memory-intensive applications, memory abstraction like this can yield significant savings.

Hardware Support Expansions: Blackwell, Grace, and ConnectX-7
While earlier announcements mentioned generic GPU support, the keynote presentations provided further information about GPUs that enterprise architects have been waiting to hear when Broadcom confirmed support for NVIDIA B200 Blackwell GPUs, tech preview support for NVIDIA Grace CPUs within VCF 9.1, and optimized offloading integration for NVIDIA ConnectX-7 SmartNICs.

Turner challenged the conventional wisdom that every enterprise AI deployment requires massive, 400-billion-parameter frontier models running on dedicated supercomputing clusters. He emphasized the ability to run smaller language models (SLMs) and open-weight architectures, such as Alibaba Qwen, Google Gemma, and NVIDIA Nemotron 3.5, directly on standard CPU infrastructure or modest GPU pools, where appropriate.

TrueSource by Broadcom: Securing the Open-Source Supply Chain
Perhaps the largest product portfolio unveiling completely absent from our initial article on the announcement was TrueSource by Broadcom.

TrueSource was introduced by Purnima Padmanabhan, vice president and general manager of the Tanzu Division. TrueSource addresses a glaring vulnerability in enterprise AI pipelines: compromised open-source dependencies. Modern applications and autonomous agents are built heavily on open-source libraries and frameworks, such as Spring, Java, Python, Node.js, LangChain, and CrewAI. Padmanabhan pointed out that as frontier models have been weaponized to discover CVEs at machine speed, monthly advisories in the ecosystems have exploded.

Padmanabhan introduced TrueSource as Broadcom's answer to the software supply chain mess, providing clean-room, SLSA Build Level 3-verified builds across three focus areas:

• Spring Enterprise: Proactive security scanning across the Spring framework and its dependency tree, providing human-verified, CVE-only patches that can be applied to pinned production releases without forcing a disruptive framework version upgrade.
• TrueSource Trusted Artifacts: Curated, signed libraries across Java, Python, and Node.js ecosystems, bundled with hardened container images from the Bitnami Secure Images catalog.
• TrueSource Data Services: Production-validated distributions, operators, and Helm charts for essential open-source data backends, including PostgreSQL, MySQL, RabbitMQ, and Valkey.

Purnima cited industry studies showing that only 26 percent of fully automated, AI-generated patches successfully fix vulnerabilities without introducing application-breaking bugs. To counter this, Broadcom is pairing frontier-model scanning with human-verified engineering. The fixes they create will be provided to community maintainers, with enterprise subscribers receiving early access to remediation before the vulnerability is publicly disclosed.

Hands-On VCF Operations: Live Patching and Config Drift
The VCF AI Assistant announcement described it as conversational troubleshooting. To show how it worked, they did a live demonstration that walked through how administrators would interact with it.

In the walkthrough, Sabina Anja, chief technologist at Broadcom, showed how the AI Assistant triaged a VMware Security Advisory alert across hundreds of ESXi hosts. The important thing is that the system did not just surface the alert; it analyzed catalog images, verified which hosts were eligible for live patching without maintenance reboots, executed pre-checks, and flagged configuration drifts in Distributed Resource Scheduler (DRS).

She showed another troubleshooting scenario in which the AI Assistant discovered that an administrator had manually disabled DRS during maintenance weeks earlier and had failed to re-enable it, resulting in uneven CPU creep across the cluster. The platform remediated the imbalance and triggered an integrated Salt compliance rule, automatically alerting the operations team whenever DRS sits disabled for more than an hour.

VMware Customers Discuss Their Experiences
Broadcom had Standard Chartered Bank discuss its migration from a legacy data center to a standardized, fully automated private cloud rack architecture. It has more than 200 clusters with a 99 percent virtualization footprint. By doing this, Standard Chartered Bank achieved a 40 percent to 45 percent reduction in data center power utilization while deploying global microsegmentation using Distributed Firewall (DFW) and NSX.

Then, Marian Lakov from United Airlines discussed adopting VMware vSphere Kubernetes Service (VKS) to modernize critical flight applications.

By implementing consistent Kubernetes APIs across its on-premises VCF infrastructure and public cloud environments, United enabled its developers to deploy using automated Terraform pipelines without having to file tickets for the infrastructure they needed. He talked about their active-active multi-region resiliency using Avi Global Server Load Balancing (GSLB) to route production traffic around unexpected site outages.

At the end of the event, VMware announced that VMware Explore will return to Vegas in the spring of 2027, May 3-6, at a new venue: Resorts World Las Vegas.

The Big Picture
Walking away from the keynote, Broadcom's strategy came into focus for me. The high-level messaging from that morning was all about general AI optimism, but the discussion on the stage focused on the details: slashing DRAM costs, vetting open-source binaries, and patching running ESXi hosts without downtime. That's what IT administrators and engineers really care about.

After the keynote ended, we shuffled out and headed to The Hub for the Welcome Reception.

Welcome Reception
The reception was well attended, and guests seemed to have a good time chatting with each other and the vendors, enjoying a beverage or two, and munching on the food provided.

After the Welcome Reception, I headed over to the VMUG Member Reception at Wakuda. I chatted with other VMUG members and then went to another Intel-sponsored meetup at the Flight Club, where I met a couple of members from the Portland VMUG. The photo below was unintentionally fuzzy, but it seemed to capture the feeling of the night.

The rest of the event was spent attending sessions and chatting with fellow attendees. Below is a look at some of the breakout sessions that stood out to me and will give you an idea of the breadth and depth of the content presented at Explore this year.

Selected Sessions Overview

Memory Economics in the AI Era: NVMe Memory Tiering on Lenovo Platforms
Catherine Maina -- senior product manager for ThinkAgile VX, Lenovo
Arvind Jagannath -- senior product line manager, vSphere Platform, Broadcom.

As I have previously written about memory tiering, and it was mentioned during the keynote, I was interested in what Lenovo, a close hardware partner of VMware, had to say on the topic.

In the session, the presenters focused on how NVMe-based memory tiering can address something we are all facing: escalating DRAM prices and supply chain delays driven by the AI boom.

They showed the testing they did on memory tiering using a Lenovo ThinkAgile VX system equipped with Intel Xeon 6 processors and high-endurance Micron PCIe Gen5 SSDs. They showed how it can leverage a 1:1 DRAM-to-NVMe ratio within VMware Cloud Foundation (VCF) to reduce infrastructure acquisition costs by up to 45 percent while significantly boosting memory density and VM capacity.

They showed how VMware ESXi transparently manages hot pages in DRAM while migrating cold pages to NVMe, maintaining near-native DRAM performance for workloads with active memory footprints under 50 percent, while supporting native cluster configuration, DRS balancing, and deep observability across vCenter and VCF Operations.

Who's Who: Top Virtualization Social Media Sites for Admins
David Frankfort -- virtualization guru, vExpert, VMUG leader, as well as storage, server, and security expert.

I had high expectations for this session, and I was not disappointed as it focused on the power of the VMware community. This was nice, as it was a practical, human-centric counterpoint to the conference's hardware and software technical tracks.

We all know that the official Broadcom documentation provides a baseline for deploying, maintaining, and troubleshooting our systems. Still, David reinforced the power of VMware's vast peer-driven community as the premier source of real-world troubleshooting and guidance on products and projects.

The session offered a look at how technical information consumption has evolved since the early VMworld era, highlighting which community platforms, social channels, and administrator networks have stood the test of time to remain indispensable for day-to-day operations.

Unlocking Performance With Next-Gen Topology-Aware CPU Scheduler in VMware Cloud Foundation
Qasim Ali -- distinguished engineer at Broadcom
Rupesh Shimkhada -- senior manager, technology product management at Broadcom
Haoqiang Zheng -- distinguished engineer at Broadcom.

This was a great session as it really got into the nitty-gritty about the extensive CPU and memory scheduling advancements introduced in VMware Cloud Foundation (VCF) 9.1.

The speakers presented eye-catching benchmark results that they achieved on enterprise workloads. Their benchmarks showed up to 30 percent higher transaction rates on Microsoft SQL Server, 26 percent improvements on Oracle DB, 25 percent faster Java operations, and up to 50 percent lower total cost of ownership (TCO) for SAP HANA deployments.

To achieve these numbers, they explained that modern multicore, multidie architectures like AMD EPYC and Intel Xeon require sophisticated topology awareness. Because remote socket and cross-die memory access introduce significant latency, VCF 9.1 implements enhanced NUMA scheduling algorithms that optimize data locality, manage cache contention, and reduce rebalancing migrations to achieve higher, more predictable throughput and more stable latency.

The presentation got really interesting when they dove deep into architectural mechanics and showed, using the VMmark 4 benchmarking tool, that database latency peaks could be reduced by roughly half compared to earlier versions. They illustrated how "monster VMs" and I/O-intensive workloads benefit from sub-NUMA clustering (SNC) and automatic NUMA-client placement across physical CPU sockets.

Looking ahead, they previewed features such as relaxed co-scheduling algorithms to minimize latency jitter for large, multithreaded VMs, smarter memory-tiering optimizations, and enhanced tail-latency controls.

The session concluded with an engaging Q&A addressing BIOS-level NUMA tuning recommendations, default auto core-per-socket allocation in ESXi, and best practices for running latency-sensitive virtual machines. This was definitely one of the most technical presentations I've attended, and if you're interested in performance tuning, I highly recommend watching it.

VMware Cloud Foundation Certification 2026: Tips, Tricks, and Guidance From the Global Community
Dave Bradley -- senior cloud engineer at Vizient
Kristof Asaert -- infrastructure consultant at Unit-D
Daniel Keer -- project lead and senior consultant at Digitally Accurate Inc.

During this session, Dave, Daniel, and Kristof tackled a topic that hits close to home for anyone who wants to run a home lab: passing VMware certification tests. In the past, subscribing to VMUG Advantage gave you evaluation licenses for the entire VMware stack, but Broadcom reshaped that model into an achievement-based gate. Today, securing non-production home lab keys, which allow you to run VCF along with vDefend and Avi Load Balancer, requires passing VMware Certified Professional (VCP) for VCF Administrator or VCF Architect exams. While turning home-lab licensing into a gated reward ruffled plenty of feathers across the community, it forced administrators to dive deeper into the complete multicloud infrastructure suite rather than just spinning up standalone ESXi hosts and vCenter instances as many of us have done in the past.

Preparing for the VCF Administrator and Architect tracks requires not only a deep understanding of VMware's products but also adapting to Broadcom's product realignment and terminology changes. The presenters underscored that while earlier preparation materials, such as study guides and YouTube architecture walkthroughs, remain technically viable, test takers now must memorize component mappings, such as Aria Operations becoming VCF Operations and Cloud Director morphing into Sovereign Operations.

Passing these tests comes down to leveraging every available test-taking mechanism and community asset. A few of the hints they shared include using Pearson VUE interface features, such as right-clicking answer distractors to strike through eliminated choices, and aggressively flagging complex scenario questions for a second pass rather than losing momentum early.

VCF requires a lot of hardware, and if you lack a live environment, free resources like VMware Hands-on Labs (HOL) remain the best alternative for building your skills in SDDC Manager workflows, NSX edge cluster provisioning, and day-two lifecycle tasks.

On the cost side, the presenters advised candidates to watch for promotional Pearson VUE retake windows and VMUG exam discount codes before registering through the Broadcom support portal. On the plus side, once you become certified, maintaining your free lab keys requires only an open-book, 30-question refresher exam every 12 months.

I was able to catch up with Dave and Daniel and discuss their session at an after-hours event. This is why I enjoy attending events like these, as you get to be up close and personal with the presenters.

Optimizing AI From Model Requirements to GPU Consumption on VMware Cloud Foundation 9.1
Frank Denneman -- distinguished engineer, Broadcom
Shawn Kelly -- senior principal architect, AI, at Broadcom.

This is one of the sessions that I really wanted to attend, but due to its length, 1.5 hours, and my other commitments, I couldn't fit it into my schedule. However, it was recorded, and I fully intended to watch it as soon as possible. This is why I love the fact that VMware recorded the Explore sessions.

An Event Worth Attending
Reflecting on VMware Explore 2026, Broadcom shone a spotlight away from the abstract generative AI hype we have been hearing so much about to the engineering challenges that we are facing, such as tackling soaring DRAM costs, tightening open-source security, and streamlining day-two operations.

Between the keynotes and deep-dive breakouts, Broadcom made it clear that VCF 9.1 is and will remain its centerpiece. Rather than talking in lofty generalities, the technical sessions tackled the exact operational bottlenecks engineers grind through every day: curbing runaway DRAM costs with NVMe tiering, smarter NUMA topology scheduling, and automated drift troubleshooting in the VCF AI Assistant.

This, when coupled with a human-centric session on how to pass your certifications to take advantage of VMUG Advantage lab licensing, the who's who in VMware social media, and real-world customer presentations from Standard Chartered and United Airlines, made Explore well worth attending.