The dust has begun to settle following Broadcom’s assimilation of VMware, and VMware Cloud Foundation 9 has had more than a year to reach production in customer environments, setting the stage for the next phase of the company’s private AI push: an AI factory.
VMware Cloud Foundation (VCF) brought an aggressive rationalization of VMware’s 168 products, which had previously been scattered across multiple business units at the private cloud infrastructure vendor. Broadcom slashed the number of separately sold VMware products to four main license tiers, but discontinued perpetual licenses and raised prices, prompting lawsuits from some large customers and creating ripe opportunities for competitors. Broadcom also drastically restructured VMware’s partner programs.
Still, 70% of VMware’s 10,000 largest customers migrated to VCF after its initial launch in 2024, and 3,000 customers have now migrated to VCF 9, which shipped last year, according to Broadcom officials on a press briefing call last week.
VCF 9 delivered product integrations, first previewed by Broadcom CEO Hock Tan in 2024, under a single interface that included VMware’s vSphere hypervisor, VMware NSX and VSAN for network and storage virtualization, plus add-on services for network security and disaster recovery, including vDefend, the Avi load balancer and VMware Live Recovery.
One customer who converted to VCF 9 told TechTarget last week that the newly consolidated software forms a reliable base layer for his on-premises data center infrastructure.
“It’s the bedrock that has provided the data center stability for us to operate with a lean staff and achieve availability and scalability goals,” said Jeffrey Ericson, CISO and senior vice president of infrastructure and vendor management for MedRisk, a healthcare company based in West Conshohocken, Penn.
MedRisk was able to dodge some of the disruption felt by other customers following the Broadcom acquisition. Ericson said the company had renewed its contract with the vendor just before the acquisition was completed in 2023 and maintained its relationship with its preferred channel partner.
MedRisk reaped benefits from streamlining products under VCF 9 and from Broadcom professional services’ help in consolidating six on-premises data centers into two, with plans to further reduce that to one on-premises data center and a public cloud installation for redundancy, Ericson said.
“We went from 175 physical servers, none of which were alike, to eight physical VMware hosts running a billion-dollar business, and VMware was a very important part of orchestrating that,” he said.
VCF 9 underpins Private AI Cloud
During the annual VMware Explore conference this week, Broadcom rolled out a bevy of product updates, including updates to VCF and add-on software services that will be sold as VMware Private AI Cloud. Broadcom also previewed planned Tanzu services for agentic orchestration and data management.
All these updates are meant to further the private AI strategy Broadcom unveiled last year by persuading VCF users to expand their usage of the platform, said Stephen Elliot, an analyst at IDC.
“It’s been a journey for customers, but they have seen the execution, engineering and evolution the acquisition has brought, [while] AI has put private cloud back in the spotlight,” Elliot said. “They’ve been on cruise control for a while, and now all of a sudden it’s like, ‘Hey, what’s your private cloud strategy? What’s the role of it with AI?’
“Large companies have access to Broadcom’s technical experts and engineering teams and a much more streamlined engagement process with the company, so it’s pretty easy to dictate where they’d like to see a product line go,” Elliot said.
Private AI services Broadcom began delivering with the general availability of VCF 9.1 in May included an AI assistant and infrastructure management features optimized for AI hardware. Among those infrastructure optimizations was integration with Nvidia vGPU drivers, which allow both VMs and containers to share GPUs without requiring a separate software license. VCF 9.1 added support for Model Context Protocol (MCP), a large language model gallery and LLM performance monitoring.
As of this week, Private AI Services will include multi-tenant model sharing, enabling administrators to deploy an LLM once and delegate shared access across multiple business units or other groups, while managing data privacy and isolation within each tenant’s namespace.
Further software updates for Private AI Services, previewed this week, include a new AI gateway that will provide prompt routing, token and usage rate limiting, and application-level authorization for LLMs, as well as a Secure Agent Framework that provides sandbox isolation and a harness for AI agents.
Private AI Cloud rolls up into AI Factory
Further VMware private AI add-ons will be available under the new, broader VMware AI Factory. A planned AgentMinder tool will include observability and agentic security features, such as agent identity management and enforcement of least-privilege access controls.
Broadcom plans updates to its vDefend and Avi load-balancer products for agentic AI security as part of VMware AI Factory. vDefend will automatically detect and identify agents, associated MCP servers, LLMs, data and tools on VCF networks, and enforce security policies to curb shadow AI. It will use an agentic AI pipeline to generate threat signatures that inform planned distributed virtual patching features. Avi recently increased its maximum throughput by 88% to 12 Tbps in anticipation of higher agentic AI network traffic.
MedRisk is primarily evaluating AI management tools from vendors other than its on-premises infrastructure providers, exploring applications with Palantir and data services in Snowflake on Azure. But the vDefend and Avi updates will be important to supporting AI, Ericson said.
“Where VMware has a leg up is that just about everything that’s going to pass through our on-prem infrastructure would pass through VMware, and using vDefend and the traffic and intelligence that it can capture, and then be able to report on that or get visibility, seems ideal,” Ericson said. “With the caveat that if we have things that are external to our on-prem environment, say, in the cloud, we may or may not see that through VMware.”
Enterprise AI’s shifting competitive landscape
Broadcom’s emphasis on private cloud in on-premises data centers has opened up at least some opportunities for vendors focused on orchestrating AI workloads across hybrid cloud environments, including IBM/Red Hat and Nutanix, but that won’t necessarily hold it back from a prominent role among enterprise AI vendors, said Matthew Flug, an analyst at IDC.
“Hock [Tan] had a vision, and I don’t think it’s a bad one. I don’t think it leaves them in a bad spot,” Flug said. “They’re really the only ones doing private cloud this way… We can argue over the course of the next five years about how many workloads will really stay in the private cloud. But there’s definitely going to be a significant amount.”
Still, Broadcom’s new Private AI Cloud features are largely in line with what other large vendors have rolled out this year, from Google Cloud to Cisco, Dell and HPE, Flug said, and many of the newest features are not yet generally available.
“Every company I’ve spoken to in the past two months, I think I can safely say, has released some kind of AI gateway or model router,” he said. “Broadcom’s being very deliberate and building out a really robust [product], but it is going to be a little late.”
For Broadcom’s customer base that looks to it for reliable products, however, private AI in production is just getting started, said Steven Dickens, founder and CEO at HyperFrame Research.
“You can do all kinds of arm-waving up the stack, but enterprises are looking for a safe pair of hands,” Dickens said. “Broadcom is speaking to operational reality for the mass enterprise market.”
AI Factory ties in AMD GPUs…Nvidia next?
Private AI Cloud software updates, combined with a catalog of more than 150 pre-tested open source, open weight and cloud-based AI models, form the software basis for the new VMware AI Factory, a “metal to model” stack that will include pre-tested hardware from AI processor, server and storage partners as well as partner MetalSoft’s bare metal provisioning tools.
Hardware partners at launch for VMware AI Factory are AMD for GPUs and certified server node makers Cisco, Dell Technologies and Lenovo. Broadcom officials didn’t rule out adding Nvidia and other hardware vendors as partners in the future. Nvidia’s Nemotron models are among those supported by AI Factory’s model as a service feature.
“[These are] the first set of partners that we worked with over the last several weeks to do factory validated server configurations and validated tested architecture that we are comfortable to show to the market,” said Prashanth Shenoy, vice president of product marketing in the VCF Division at Broadcom. “Obviously, we are going to expand the set of ecosystem partnerships beyond AMD, as well as beyond the OEM and ODM server partners that we listed out… Nvidia is definitely top of mind for us.”
Meanwhile, AMD hardware is likely easier and less expensive to get amid the global memory shortage and rising Nvidia prices, said Rob Strechay, lead analyst at VB Intelligence.
Enterprises building on-premises AI are increasingly considering a mix of AI chips and processors, including Google’s TPUs, Amazon’s Inferentia, Microsoft’s Maia 200 and Arm AGI CPUs, he added.
“AMD has been doing a very big push in the space because they don’t have all the software assets that Nvidia has, and to be honest, the AMD product is not the same. But in many ways it’s good enough,” Strechay said. “I think people look at it and go, ‘You can get AMD chips, and it’s not going to bankrupt you.'”
Beth Pariseau, senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism. Have a tip? Email her or connect on LinkedIn.