Broadcom unveiled VMware AI Factory at VMware Explore 2026 in Las Vegas, introducing a software-defined foundation within VMware Private AI Cloud designed to help enterprises deploy, govern, and secure AI workloads from bare-metal infrastructure through to live model inference.
The announcement, made on August 31, 2026, positions the platform as a direct response to enterprise frustration over the slow, costly, and complex path from raw hardware to production-ready AI.
Paul Turner, chief product officer of VMware Cloud Foundation Division at Broadcom, said enterprises want AI running where their data already lives, but the journey from metal to model has historically been too cumbersome.
VMware AI Factory addresses this by automating infrastructure deployment, unifying lifecycle management, and giving organizations flexibility to select their own hardware and vetted AI models while keeping private cloud costs predictable.
The platform’s core value proposition is speed. VMware Cloud Foundation‘s automation capabilities can reportedly cut the time from bare-metal server deployment to serving a first AI model from several weeks down to just hours, by fully automating hardware provisioning, software stack enablement, and end-to-end lifecycle operations.
Security and governance are baked into the private AI services layer that underpins the factory, as highlighted in the Broadcom product announcement.
GPU resources are pooled and shared across teams rather than dedicated per workload, and a unified model gallery gives IT and data science teams centralized visibility into model deployment, retrieval-augmented generation workflows, token throughput, latency, and compute utilization.
Three capabilities stand out from a cybersecurity perspective. Multi-tenant Model Sharing lets organizations share AI models across business units through isolated namespaces, preserving data privacy while avoiding redundant GPU-heavy deployments.
AI Gateway centralizes governance across on-premises and cloud environments through a single interface, adding intelligent prompt routing, token- and usage-rate limiting, and application-level authorization controls.
Most notably, Secure AI Sandboxes and Governance introduce virtualized container spaces that isolate agent-generated code execution, paired with a control layer that defines how autonomous agents are invoked, which tools they can access, and how their outputs are validated before any action is taken a direct answer to growing concerns about unchecked agentic AI behavior in enterprise environments.
| Core Capability | Architectural Component | Operational & Security Function |
| Multi-Tenant Sharing | Isolated Namespaces | Eliminates redundant GPU hardware while safeguarding private data |
| AI Gateway | Unified Management Interface | Enforces prompt routing, token rate-limiting, and app authorization |
| Secure AI Sandboxes | Virtualized Containers | Isolates agent-generated code execution and governs tool access |
| Hardware ReadyNodes | Cisco, Dell, Lenovo, Supermicro, AMD | Enables zero-touch provisioning across vSphere, vSAN, and Kubernetes |
| Governed Model Gallery | 150+ Open & Commercial Models | Delivers sovereign models-as-a-service on private infrastructure |
Broadcom is pairing the software layer with certified VCF AI ReadyNodes from Cisco, Dell Technologies, Lenovo, and Supermicro, and is also collaborating with AMD to combine VMware Cloud Foundation with AMD Instinct MI350 Series GPUs and the open ROCm software ecosystem.
Zero-touch provisioning will orchestrate deployment across vSphere, vSAN, Kubernetes, and the AMD GPU operator stack.
A separate partnership with MetalSoft brings integrated bare-metal automation to VCF, cutting physical server provisioning time from weeks to minutes and letting IT teams manage heterogeneous hardware directly through the VCF console, eliminating the need for vendor-specific tools.
On the model side, VCF customers can now access more than 150 open-source and commercial models, including Nemotron 3, Gemma 4, cotomi, Qwen 3.7-Max, and GLM 5.2, delivered as governed models-as-a-service, giving enterprises a data-sovereign, cost-controlled route to running AI on their own infrastructure without relying entirely on external cloud providers.
Prevent incidents due to slow investigations. Power your Tier 1 with threat intelligence from 15K SOCs: Integrate TI Lookup in your SOC
