Broadcom Launches AI Factory Based on VMware Cloud Foundation

Broadcom launched a VMware AI Factory at VMworld Explore, providing a software-defined framework for deploying AI models on shared GPU infrastructure, and validated over 150 AI models including NVIDIA Nemotron 3, Google Gemma 4, Alibaba Qwen 3.7-Max, NEC cotomi, and Z.ai GLM 5.2 to run on VMware infrastructure. The company also introduced AgentMinder and extended vDefend and Avi Load Balancer to govern and secure AI agents, while reporting that 3,000 IT organizations now run VCF 9.

TL;DR — Key Takeaways - Broadcom’s VMware AI Factory aims to give enterprises a software-defined foundation for deploying and operating AI models across shared GPU infrastructure. - The broader VMware stack is adding agent-focused governance and security through Tanzu, AgentMinder, vDefend and Avi Load Balancer, extending controls to AI agents, MCP servers and sensitive data. - Broadcom is positioning VCF as a way to contain rising AI infrastructure and token costs while giving enterprises greater control over private and self-hosted AI workloads. Broadcom at its VMworld Explore https://www.vmware.com/explore/us conference launched multiple artificial intelligence AI initiatives that extend the reach and scope of its core IT infrastructure portfolio, including a VMware AI Factory that provides a software-defined framework for deploying AI models. Additionally, Broadcom announced today that NVIDIA Nemotron 3, Google Gemma 4, Alibaba Qwen 3.7-Max, NEC cotomi and Z.ai GLM 5.2 models have all been validated to run on VMware infrastructure. In total, more than 150 AI models have now been validated by Broadcom. The VMware AI Factory is based on an existing VMware Private AI Cloud, a new offering that enables IT teams to deploy AI inference models in a multi-tenant on-premises IT environment or on a cloud service using VMware Platform based on the Cloud Foundry platform-as-a-service PaaS environment running on the VMware Cloud Foundation VCF 9 platform that provides compute, networking and storage services. Broadcom today also revealed it has allied with MetalSoft, which also reduces the time needed to provision bare-metal infrastructure to run VCF from weeks to minutes. The VMware AI Factory makes it possible to automate the provisioning of isolated pools of shared server resources that include graphical processor units GPUs along with governance, security and observability capabilities that can be used to track token costs Additional capabilities that have been added to the VMware Tanzu Platform to specifically run AI agents in a PaaS environment are hardened sandboxes, a catalog, a harness that enables developers to build and deploy AI applications, an AI gateway to monitor and rate limit AI agents, and an ability to process both structured and unstructured data. Broadcom is also adding to its portfolio AgentMinder, a tool based on the OpenTelemetry framework for collecting telemetry data that verifies agent identity and authorizes each action against its declared mission, intent, context, and risk level. At the same time, VMware vDefend has now been extended to automatically identify Model Context Protocol MCP servers, large language models LLMs and data stores running on VCF. VMware Avi Load Balancer, meanwhile, has been extended to restrict agents from accessing unauthorized MCP tools and inspect transaction content to block malicious execution. It will also establish normal agentic traffic baselines for AI agents in addition to preventing unauthorized exfiltration of credentials, personally identifiable information PII , and sensitive financial data. Ultimately, Broadcom is extending VCF to make it simpler to securely deploy any type of AI model cost-effectively as the cost of IT infrastructure continues to rise, says Prashanth Shenoy, chief marketing officer for the VCF division at Broadcom. The NVE memory tiering capability that Broadcom provides enables IT teams to deploy AI models in a way that reduces the total cost of AI, he adds. “We are helping them manage and overcome AI tokenomics,” says Shenoy. Broadcom reports that 3,000 IT organizations are now running VCF 9. The percentage of those organizations that are running AI workloads on VCF is unknown, but over time, Broadcom is clearly betting that many of those workloads will be running in some type of self-hosted IT environment running VCF.