Broadcom used VMware Explore in Las Vegas to announce a set of validated models for VMware Cloud Foundation, so enterprises can offer model-as-a-service on their own hardware. Five are named: Nemotron 3 from Nvidia, Gemma 4 from Google DeepMind, cotomi from NEC, Qwen 3.7-Max from Alibaba Cloud and GLM 5.2 from Z.ai.
Five and 150 are not the same claim
The release states both numbers in adjacent paragraphs, and they mean different things. Five models have been tested and validated by Broadcom on specific configurations. More than 150 is the count of open-source models the vLLM runtime can technically serve — a capability statement with no vendor guarantee attached. Treating the two as equivalent inflates Broadcom's commitment thirtyfold.
The interesting names on the list
Two of the five validated models are Chinese. Qwen 3.7-Max is described in the release as a proprietary multimodal model with a one-million-token context window, offered for sovereign, on-premises access; GLM 5.2 is Z.ai's open-source general language model, positioned for local coding and reasoning agents. For an enterprise that cannot send data to a public cloud, this is a supported path to Chinese open-weight models on hardware it owns — with a US vendor's stack around it. NEC's Japanese-optimised cotomi sits in the same list for the same reason: sovereignty, not benchmark leadership.
Two numbers that are not measurements
The release cites MLPerf Inference v5.1 as a measurement programme. That is a reference to a methodology, not a published MLPerf result — no performance figures appear anywhere in the announcement. And the frequently quoted 56% of enterprises running production inference in private cloud comes from Broadcom's own Private Cloud Outlook 2026 survey, not an independent analyst house.
What the packaging says
Validating a fixed short list on named hardware is how infrastructure vendors have always sold databases and hypervisors: pick the supported combinations, own the support contract. Applying it to model serving means the enterprise choice narrows from "any model on Hugging Face" to "the five we will answer the phone about".