Private AI cloud, agentic infrastructure dominate VMware Explore

Broadcom announced the VMware AI Factory, a packaged AI infrastructure stack for on-premises production AI, at VMware Explore in Las Vegas, aiming to reduce deployment time from weeks to hours. The company also unveiled an agent governance platform and an open-source security portfolio, with the AI Factory integrated into VMware Private AI Cloud and supporting accelerators from Nvidia and AMD.

Broadcom https://www.networkworld.com/article/4180754/will-broadcoms-vmware-strategy-keep-paying-big-dividends.html announced a new packaged AI infrastructure stack it’s calling the VMware AI Factory today at VMware Explore https://www.vmware.com/explore/us in Las Vegas. The company also announced a new agent governance platform and an open-source security portfolio. The bet is that companies will want to run their production AI on premises https://www.networkworld.com/article/4182967/ai-inference-moving-to-private-clouds-broadcom-says.html using VMware Private AI Cloud. “The VMware AI Factory is not a separate product nor does it cost extra,” says Prashanth Shenoy https://www.linkedin.com/in/prashanthshenoy/ , CMO and vice president of marketing for the VMware Cloud Foundation division at Broadcom. Instead, it’s something that customers build using the infrastructure automation capabilities in VMware Cloud Foundation, he tells Network World , “which can reduce the time from bare metal server deployment to serving the first AI model from weeks to a matter of hours.” It works with AI accelerators from Nvidia https://www.networkworld.com/article/3562856/nvidia-latest-news-and-insights.html and AMD and is certified to work with servers from Cisco, Dell, Lenovo, Supermicro and others. It can also run more than 150 open-source and commercial models. Five models have already been tested and validated, including models from Nvidia, Google, NEC, Alibaba, and Z.ai. There’s built-in observability into token throughput, latency, compute, and memory utilization, Broadcom says. And it already supports secure sharing of AI models between tenants or lines of business. In the future, VMware AI Factory will also have secure virtualized container space to isolate agent-generated code, with a control layer defining how agents are invoked, what tools they access, and how their outputs are validated. The challenge is getting all the pieces together—the GPUs, the networking, the storage, the containers, the AI software stack, and the AI models themselves—Shenoy said in a media briefing last week “Putting all this together is a very painstaking process,” he said. “The VCF AI Factory provides them with a turnkey, integrated solution that helps our customers accelerate the time to their first AI model deployment.” This isn’t VMware’s first foray into private AI. The company announced VMware Private AI Services https://www.networkworld.com/article/4053013/broadcom-touts-ai-native-vmware-but-gains-arent-revolutionary.html at last year’s conference, and it became part of the VCF bundle https://www.networkworld.com/article/4008384/broadcom-resets-private-cloud-strategy-with-vmware-cloud-foundation-9-0.html in early 2026. And that offering is still around. “Private AI Services is most definitely not deprecated,” Shenoy tells Network World . It is integrated with the VCF platform, he says. The VMware AI Factory is the infrastructure foundation, he says. But it’s just one part of the broader VMware Private AI Cloud, which combines the AI Factory with other VMware services, including VMware vDefend and Avi Load Balancer for agentic security, and the Tanzu platform’s AI-ready data foundations https://www.networkworld.com/article/4053013/broadcom-touts-ai-native-vmware-but-gains-arent-revolutionary.html , hardened agent sandboxes, agentic harnesses, and auditable agent governance. The world of cybersecurity has changed dramatically since last year’s conference, Umesh Mahajan https://www.linkedin.com/in/umeshmahajan/ , general manager for application networking and security at Broadcom, told reporters last week. “AI is coming at us at galactic speed,” he says. “Every mediocre actor has become an elite attacker thanks to AI.” To help customers defend against these threats, VMware has continued to expand the use of AI in its security portfolio, improving dashboards and AI assistants. But the biggest announcements were all about securing AI agents being deployed by enterprises. “We’ve had 50 years to figure out how to control and manage employees,” says Clayton Donley https://www.linkedin.com/in/claytondonley/ , vice president and general manager of the Identity Management Security division at Broadcom. “We’ve had 25 years, perhaps, managing extranets and giving customers access to do things in our systems.” The question today, he said in the media briefing last week, is how to put the same kinds of controls around AI agents. “And how do we control it when it’s actually happening,” Donley adds, “not just at the governance level after the fact?” To solve this problem, Broadcom announced the release of AgentMinder https://www.broadcom.com/agentminder , available today, that’s designed to act as a traffic controller for AI agents. It verifies agent identities and authorizes each action against the agent’s declared mission, intent, context, and risk. Broadcom is already using the tool internally, the company says, and it can support peak loads of nearly 43 million API calls per day. “It has enabled company wide deployment of our AI agentic pipeline,” says Broadcom CIO Alan Davidson https://www.linkedin.com/in/alan-davidson-0a90286/ in a press release. “We’ve achieved massive global scale paired with zero downtime, even during maintenance and upgrades.” AgentMinder is one of multiple AI gateway tools that VMware now offers or plans to offer. In the “coming soon” category, there will be an AI gateway in the VCF Private AI Services, which will include application authorization as well as token and usage rate limiting and intelligent prompt routing. There’s an AI gateway in the Tanzu Platform, which monitors, rate-limits and logs agent actions and manages agent credentials. This is slated for release later this year. “They can help the customer not only understand what the agents are doing but also be able to monitor token usage and actual consumption of the AI,” says Justin Giardina https://www.linkedin.com/in/justingiardina/ , CTO at 11:11 Systems, a managed services provider and VMware partner. “That’s a solution to a real problem today.” These AI gateways aren’t the only AI security announcements this week. The Avi Load Balancer service will soon have Agentic Threat Defense, which includes restrictions on MCP tools, prompt inspection, and exfiltration prevention. It also has what Broadcom calls “zero-day attack detection”—if an AI agent starts acting weird, it will flag and isolate it in real time. Then, in VMware vDefend, zero-trust lateral security will soon extend to agentic workloads, with automatic discovery of MCP servers, LLMs and other unauthorized shadow AI tools. In addition, the VMware Tanzu Platform will include new agentic AI capabilities later this year, including out-of-the-box agentic harnesses that include pre-approved skills, workflows, memory services, and human-in-the-loop controls as well as hardened agent sandboxes, and a curated marketplace of AI models and tools and data products. “If you think about it, an agent is just another microservice,” says Purnima Padmanabhan https://www.linkedin.com/in/ppadmanabhan/ , general manager of the Tanzu Division at Broadcom. “We have taken that microservice architecture and applied it and scaled it out to build agents very easily.” Another significant announcement is about securing the open-source ecosystem. “We have talked a lot about how we are scanning our existing commercial software with frontier AI, and how we are aggressively patching all of that,” Padmanabhan says. “But Broadcom also is the primary committer to open-source projects like Spring.” These open-source efforts are now combined under TrueSource. In addition to the existing support for Spring Enterprise, Broadcom announced two new services: TrueSource Trusted Artifacts, which provides secure clean-room builds for Java, Python and Node.js; and TrueSource Data Services, which provides artifacts and support for PostgreSQL, RabbitMQ, MySQL, and Valkey data engines. According to Broadcom, its engineers continually scan Spring and its dependency tree using frontier models, finding vulnerabilities before attackers do, then verifying each patch by hand. In the past five months, engineers have used more than 12 billion tokens to do this. Similarly, Broadcom engineers find and fix problems in the Java ecosystem, Python, and Node.js. “Everything is scanned, fixed, contributed back,” Padmanabhan said at the pre-briefing. “We provide clean room bills, and wherever we need to patch, we will work with the maintainer. If not, we’ll be the maintainer of the last resort.” Spring Enterprise, TrueSource Trusted Artifacts, and TrueSource Data Services are available today. They’re not bundled in VCF but can be purchased separately, with tiered site licenses. As enterprise customers operationalize AI, if they do so on the VMware stack, that could lock them further into the platform. That could be an advantage, as competitors—and clouds—continue to siphon off VMware’s customers. Read more: Will Broadcom’s VMware strategy keep paying big dividends? https://www.networkworld.com/article/4180754/will-broadcoms-vmware-strategy-keep-paying-big-dividends.html Last week, for example, VMware competitor Nutanix announced https://ir.nutanix.com/static-files/a3c778d0-41c3-4d3a-a3bd-74eff0c253ec that it gained 3,000 new customers over the past year. According to 6sense https://6sense.com/tech/converged-infrastructure/nutanix-market-share , Nutanix currently has about 10,000 customers, while VMware has over 200,000 https://6sense.com/tech/virtualization/vmware-market-share . Today, most organizations continue to rely heavily on VMware because it’s been a de-facto standard for more than a decade, migration is risky, and many third-party applications and virtual appliances are certified or optimized for VMware, research firm Gartner says https://www.gartner.com/en/documents/7690625 . But many customers are dissatisfied with the pricing and licensing changes https://www.networkworld.com/article/3813523/thinking-of-moving-off-vmware-gartner-tallies-cost-of-large-scale-vmware-migration.html that occurred after Broadcom acquired VMware https://www.networkworld.com/article/3484025/vmware-by-broadcom-product-service-and-support-news.html in 2023 and want to migrate, according to Gartner, and, by 2029, 55% of enterprises will migrate 100% of their VMware workloads. According to a survey the research firm released in April, 76% of IT leaders and CIOs have a negative outlook toward the ownership of VMware by Broadcom—up from 64% in 2025 and 33% in 2024, respectively. As a result, 67% are looking for VMware alternatives while 19% are accelerating cloud adoption plans. And 35% have already either fully migrated or are in the process of migrating their entire portfolios away from VMware. Colocation provider Flexential is a VMware user and partner, but plans to look for alternatives had been on its roadmap even before the pricing and licensing changes. “But the alternatives are not fantastic,” says Craig Cook https://www.linkedin.com/in/craigcookitsolutions/ , Flexential’s senior vice president of cloud and data protection. “VMware has a very dominant market share, and for a reason. They have a great set of technologies that are mature, robust, and stable. We are currently all-in.” In fact, Flexential is now building its own set of offerings on top of the VMware AI Factory, set to be released early next year. “Basically, you can do everything there,” Cook tells Network World . “You can bring your own model. You can spin up your agents. You can orchestrate it. And they’ve got observability, which is critical.” Flexential will bundle that with hosting and services for a complete private AI solution. “Their stuff sits next to most of the data in the enterprise today, and that latency, that proximity, is important,” he says. Running AI on premises instead of in the cloud can offer cost, latency, and compliance benefits, but it can also be a management challenge, and organizations are still struggling to find the right balance. According to an Omdia survey https://blogs.cisco.com/news/cloud-or-on-premises-new-report-shows-why-ai-workload-placement-matters of 1,201 IT leaders released in August, 96% currently use a mix of cloud, on-premise, and edge infrastructure for AI. As share of workload, 59% of inferencing workloads currently run in the cloud, and 41% on-prem. In three years, companies expect to run 63% of AI inferencing workloads in the cloud and 37% on-prem. Meanwhile, other workloads are moving in the opposite direction—according to the survey, 60% of companies have repatriated some workloads back from the cloud to on-prem. For companies running AI workloads on prem, integrating AI security with a virtualization platform makes sense, says Ryan Sheehan https://www.linkedin.com/in/ryan-sheehan-8b17763/ , senior vice president of advanced solutions at SHI International, an IT consultancy. “That’s where I think VMware is in a very good position,” he tells Network World . “They’re close to the infrastructure.” That puts VMware in the right place to enforce controls. “They’re at the hypervisor,” he says. “They are the hypervisor.”