Broadcom has revealed its new VMware AI Factory at VMware Explore 2026 in Las Vegas, aiming to simplify how enterprises deploy, manage, and secure AI workloads on private infrastructure. Designed to span everything from raw hardware to fully operational AI models, the platform addresses common challenges like long setup times, complexity, and inflated costs. The launch took place on August 31, 2026.
From Weeks to Hours: Speeding Up AI Deployments
The VMware AI Factory builds on VMware Cloud Foundation (VCF) and brings automation to deployment workflows. Rather than waiting weeks, enterprises can progress from bare-metal servers to serving first AI model in hours. This includes automated hardware provisioning, software stack enablement, and full lifecycle operations.
Security and governance are integrated throughout the stack. Resources like GPUs aren’t locked to individual workloads; instead, teams share pooled compute, reducing overhead. A governed model gallery gives visibility into how models are deployed and managed, including metrics like latency, token throughput, and compute utilization.
Key Security Features
Among the notable security capabilities are:
- Multi-Tenant Model Sharing:Uses isolated namespaces so business units can share models while keeping data separate—preventing redundant GPU usage.
- AI Gateway:Centralizes governance with unified interfaces and features like prompt routing, usage rate-limiting, and application-level authorization. Works across on-premises and cloud environments.
- Secure AI Sandboxes and Governance:Employs virtualized containers to isolate agent-executed code, controls which tools agents can access, and adds validation layers before output is acted upon. A direct approach to avoiding uncontrolled AI agent behavior.
Hardware, Models, and Partners
Broadcom plans to deliver the software paired with certified AI-ready hardware from Cisco, Dell Technologies, Lenovo, and Supermicro. A collaboration with AMD adds support for AMD Instinct MI350 GPUs and the ROCm software stack.
The platform supports zero-touch provisioning across environments such as vSphere, vSAN, Kubernetes, and the AMD GPU operator stack. Automating physical server deployment with MetalSoft is also part of the strategy, reducing provisioning tasks from weeks to minutes.
On the AI model front, users will have access to a gallery of over 150 open-source and commercial models. Some of the models available include Nemotron 3, Gemma 4, Cotomi, Qwen 3.7-Max, and GLM 5.2. Each model is delivered under a governed model-as-a-service structure, allowing enterprises to run AI workflows while maintaining sovereignty over their data.
Enterprises have long struggled with getting AI from hardware to production securely and cost-efficiently, especially when handling sensitive data. The VMware AI Factory offers a unified, governed approach that wraps infrastructure, models, and operations under one roof. The added hardware ecosystem and broad model support help reduce vendor lock-in and infrastructure sprawl. What to watch next: how fully automated deployments perform at scale, how governance actually works in real scenarios, and whether enterprises can reduce their reliance on external cloud providers for secure AI workloads.
