Enterprises are navigating rising AI costs while avoiding vendor lock-in as an increasingly complex infrastructure layer presents additional challenges.
Vendors are starting to recognize the struggle enterprises face with managing AI infrastructure. Last week Broadcom introduced VMware Private AI Cloud and its software-defined foundation VMware AI Factory. The foundation automates infrastructure deployment capabilities such as hardware provisioning, end-to-end lifecycle management and software stack enablement.
Integrating hardware and software operations lets enterprises reduce operational complexity while quickly scaling AI workloads, the company said. VMware AI Factory and Private AI Cloud are Broadcom’s “first full-throated response” to helping enterprise customers along their AI transformation journeys, according to Matt Kimball, principal data center analyst at Moor Insights & Strategy.
With environments growing larger and becoming more challenging to govern, even the best system administrators will get lost, Kimball said in an email to CIO Dive.
“It’s not about capabilities,” Kimball said. “It’s about the complexity of all these new demands tied to GPUs, memory tiering and performance tuning. Especially as we start to see AI emerging as a new workload with such unique requirements.”
Tools such as the latest platform releases from Broadcom are maturing in a way that will help enterprises address infrastructure complexity, a key friction point for AI adoption.
“What VMware has delivered is a good opening salvo, but I expect to see VCF — the VMware AI Factory and Private AI Cloud — rapidly evolve into that autonomous AI control plane,” Kimball said.
Simplifying the infrastructure layer
A concept largely led by Nvidia, AI factories focus on providing advantages of centralized, integrated stacks and giving enterprises all the performance benefits of AI, Brad Shimmin, VP and practice lead of data intelligence, analytics and infrastructure at The Futurum Group, told CIO Dive. At the same time, they enable optionality without introducing complexity and technical debt, he said.
“It has and will continue to be a multicloud, multivendor, multimodal environment,” Shimmin said. “Every single layer of that AI stack companies are building right now is taking advantage of this ever-moving conveyor belt of innovation that’s coming out of both the frontier model makers and the clouds that are hosting and building on top of these models.”
VMware AI Factory is a “great move for Broadcom,” which is known for its ability to simplify infrastructure, he added.
To streamline deployment of the VMware AI Factory, Broadcom is partnering with MetalSoft to reduce time spent on bare-metal provisioning from weeks to minutes, the company said. IT executives will be able to provision physical servers from multiple vendors directly through the VMware Cloud Foundation management console, “unifying the software and hardware lifecycle into a single operational model,” the company said.
VMware AI Factory also gives enterprises a path to run leading AI models on-premises. Broadcom announced during an Aug. 31 event that frontier AI models from providers including Google, Nvidia and Alibaba Cloud are validated to run on VMware Cloud Foundation, which can be accessed in Private AI Cloud. VCF customers have access to more than 150 open source and commercial models.
VMware AI Factory enables enterprises to choose both preferred hardware and vetted models, enabling better control over AI tokenomics more broadly, Paul Turner, chief product officer of the VMware Cloud Foundation division at Broadcom, said in an announcement.
Broadcom’s move to abstract underlying hardware complexity through automated infrastructure capabilities also addresses a lot of skills gap issues, but doesn’t fully solve them, Kimball said.
“It’s a great first step,” Kimball said. “But a first step nonetheless.”
