Broadcom Validates Five AI Model Families for VMware’s On-Premises Model Service

Enterprises using VMware Cloud Foundation now have a validated route to run five additional AI model families on premises and offer them internally as a model-as-a-service. Broadcom validated NVIDIA’s Nemotron 3, Google DeepMind’s Gemma 4, NEC’s cotomi, Alibaba Cloud’s Qwen 3.7-Max and Z.ai’s GLM 5.2.

The change is aimed at organizations that want to serve models from their own infrastructure rather than simply give individual teams direct access to a model provider. Broadcom says the validation is intended to let customers deploy the named models on premises and deliver them as a service to internal users.

A wider catalog sits behind the five validations

The five families are named validations within a broader runtime catalog. Broadcom says VCF uses vLLM as its default model runtime and can support more than 150 open-source models. The company also says VCF supports mixed compute across AMD, Intel and NVIDIA hardware, giving customers a choice of CPU and GPU infrastructure.

The services that turn a model into a shared resource

VMware AI Factory supplies the surrounding deployment package. It combines VCF with customer-selected AI software, accelerator architectures and certified AI ReadyNodes from vendors including Cisco, Dell, HPE, Lenovo and Supermicro. VMware says it automates hardware provisioning, software-stack enablement, infrastructure deployment and lifecycle management.

Three services address separate deployment tasks

• Multi-tenant model sharing uses isolated namespaces so tenants or business units can access shared models without separate deployments. VMware says this is intended to reduce redundant deployments and GPU waste.
• AI Gateway provides one interface for models hosted on premises and in the cloud, with prompt routing, token and usage limits, and application authorization.
• Secure virtualized container spaces isolate agent-generated code. A control layer governs agent invocation, tool access and output validation before an action is taken.

Validation narrows the infrastructure question

Broadcom is not presenting the five validations as a single prescribed model stack. Its approach combines a default runtime, support for a larger model catalog and mixed hardware options with services that govern how teams share, route and run AI workloads. The immediate result is a tested VCF path for the five newly named families; enterprises still choose which models and infrastructure fit their applications.