Broadcom’s VMware unit just renamed and expanded its private AI tools. The new package carries a name long associated with Nvidia. Yet this version runs first on rival AMD silicon.
The announcement landed Monday at VMware Explore in Las Vegas. It arrives as enterprises hunt for ways to control exploding inference costs without handing data to public cloud providers. VMware AI Factory promises exactly that: automated infrastructure from bare metal to model serving, all inside VMware Cloud Foundation.
Call it co-opetition at its finest. Nvidia popularized the “AI factory” phrase years ago to describe end-to-end systems built around its GPUs, networking and software. VMware, a longtime Nvidia partner that helped virtualize those expensive cards, now applies the same label to a stack centered on AMD Instinct GPUs and the open ROCm ecosystem. The move highlights a market shifting from raw GPU scarcity to operational efficiency and predictable pricing.
Prashanth Shenoy, vice president of product marketing at Broadcom’s VCF division, described the offering in The Register as “an evolution” of the earlier VMware Private AI Foundation with Nvidia. “VMware AI Factory represents a full-stack, automated operational system designed to treat AI token generation as a continuous production pipeline,” Shenoy said. He noted the platform supports multiple accelerator architectures and called Nvidia the company’s longest-standing GPU vendor partnership.
But for now the spotlight sits on AMD. Broadcom and AMD are collaborating to deliver a VMware AI Factory that pairs VCF with AMD Instinct MI350 Series GPUs and the open AMD ROCm software ecosystem, according to Broadcom’s official release. Zero-touch provisioning orchestrates the entire stack from vSphere and vSAN through Kubernetes and the AMD GPU operator. An AMD DVX driver attaches GPUs to large VMs consumed by a VMware vSphere Kubernetes Service cluster.
Paul Turner, chief product officer of Broadcom’s VMware Cloud Foundation Division, put the value proposition plainly in coverage by Investing.com: “VMware AI Factory changes that. We give customers a software-defined foundation that automates infrastructure deployment, unifies lifecycle management, and lets them choose their preferred hardware and vetted models.”
The economics matter. Public cloud inference bills can spiral. On-premises setups often suffer from underutilized GPUs locked away in departmental silos. VMware AI Factory lets organizations deploy a model once and share it securely across tenants or business units through isolated namespaces. Token monitoring, rate limiting and an AI Gateway enforce governance. Secure AI sandboxes isolate agent-generated code and limit tool access. And an observability dashboard tracks utilization, latency and cost.
Servers from Cisco, Dell Technologies, Lenovo and Supermicro carry official VCF AI ReadyNode certification. A new integration with MetalSoft slashes bare-metal provisioning from weeks to minutes. The result: time from raw hardware to first model serving shrinks from weeks to hours, multiple sources report.
Model support looks broad. VCF customers can run more than 150 open source and commercial models. Five already validated include Nvidia’s Nemotron 3, Google’s Gemma 4, NEC’s cotomi, Alibaba’s Qwen 3.7-Max and Z.ai’s GLM 5.2, according to Broadcom’s press release and coverage in Network World.
Shenoy told SDxCentral the initial hardware partners focus on AMD but Nvidia remains “top of mind.” The platform does not replace Nvidia’s own AI factory reference architectures. It simply gives customers another validated, automated path that avoids per-token cloud pricing.
Analysts see the announcement as part of a larger industry move toward private AI clouds. A Broadcom survey cited in its materials found 56% of enterprises already run or plan to run production AI inference on private infrastructure. Data sovereignty, cost control and security concerns drive the shift.
AMD itself has pushed hard into rack-scale systems. Its Helios platform, built on Instinct MI400-series GPUs and 6th Gen EPYC Venice CPUs, targets the same inference-heavy workloads now dominating global AI compute. Shipments ramped in recent months with commitments from OpenAI, Anthropic, Meta and Microsoft measured in gigawatts. The VMware partnership validates AMD’s ROCm software in enterprise private clouds and offers a software-defined control plane on top of that hardware.
Yet challenges remain. Nvidia’s CUDA still dominates developer mindshare. ROCm has narrowed the gap but enterprise adoption outside hyperscalers has lagged. VMware’s long history virtualizing Nvidia GPUs gives the new AMD-focused factory credibility, yet customers will watch real-world performance numbers closely.
The broader VMware Explore agenda reinforced the private AI theme. Announcements around agent governance, Tanzu data foundations and strengthened open-source security for Python and Java libraries painted a picture of infrastructure built to handle agentic workloads securely at scale. Those agents, which generate code, call tools and access resources, require exactly the sandboxing and policy controls VMware highlighted.
Enterprises face a simple choice. They can keep buying discrete GPU clusters for every team and watch costs climb. Or they can treat inference as a shared production pipeline with centralized governance, usage-based accounting and hardware choice. VMware AI Factory bets the latter wins.
Whether the borrowed branding confuses buyers or simply borrows mindshare from Nvidia’s marketing remains to be seen. What matters more is the underlying promise: faster deployment, lower and more predictable costs, and the ability to keep sensitive data and models behind the firewall. For CIOs tired of surprise cloud bills and shadow AI projects, that message lands at the right time.
Broadcom has not ruled out a dedicated Nvidia-centric AI Factory variant. For the moment, though, the company is using its virtualization strengths to give AMD a stronger foothold in the enterprise private cloud market. The competition between the two GPU vendors just gained a powerful new layer of software abstraction.
from WebProNews https://ift.tt/APGL3lZ
No comments:
Post a Comment