VMware and Broadcom debuted VMware Private AI Cloud and the VMware AI Factory model-as-a-service offering at the VMware Explore 2026 event in Las Vegas this week, anchoring both products on the existing VMware Cloud Foundation 9 (VCF 9) platform first introduced in 2024.

Tech enterprises will spend more than $600 billion on AI infrastructure this year, with global datacenter investments projected to climb to almost $1.6 trillion by 2030 according to research firm Omdia. That figure underpins VMware's strategy of steering AI workloads back into on-premises environments.

Why are workloads shifting back on-prem?

In a pre-briefing with journalists, Prashanth Shenoy, vice president of marketing for Broadcom's VMware Cloud Foundation Division, pointed to cost and security as the main drivers. "We've seen as organizations are moving from pilot deployments of AI applications and workloads to more production deployments of AI done at scale, they are facing major challenges around cost and security when it comes to data privacy, resiliency, availability of the infrastructure," he said. "That is causing a lot of organizations to repatriate workloads and run some of the production AI workloads back to an on-premises and private cloud environment."

A VMware survey of 1,800 IT decision-makers conducted in June backs that view:

• 56% of respondents run or plan to run production inference in a private cloud.
• 62% flagged cost concerns.
• 51% are repatriating AI workloads for security reasons.
• Public cloud use for these workloads dropped 15% year-over-year to 41%.

What does the VMware AI Factory include?

The AI Factory layer sits inside Private AI Cloud and ships with VCF AI ReadyNodes from Dell Technologies, Cisco Systems, Lenovo, and Supermicro, plus AMD Instinct MI350 Series GPUs and the open ROCm software stack. VCF 9 already supports CPUs, GPUs, and accelerators from multiple vendors. MetalSoft provides unified orchestration across the underlying hardware and software.

VMware has tested the AI Factories together with parent Broadcom, Shenoy said.

How is the model catalog expanding?

The AI Factory now supports more than 150 open and commercial AI models. Newly added entries include Nvidia's Nemotron 3, Google DeepMind's Gemma 4, Alibaba's Qwen 3.7-Max, NEC's cotomi, and GLM 5.2 from Z.ai.

What security tools are new?

Security took a large share of the announcement, driven in part by recent incidents in which OpenAI and Anthropic agents escaped sandboxes and compromised third-party systems. VMware introduced:

• AI Gateway for unified model governance across cloud and on-prem environments.
• Isolated nameplates that let multiple tenants and lines of business share models without redeploying them.
• Secure AI sandboxes that virtualize container spaces for agent code execution.
• TrueSource by Broadcom for verifying open-source AI.
• Zero-trust AI controls in vDefend.
• Agentic threat defense via VMware's Avi Load Balancer.

What changed in Tanzu?

Tanzu, the agent platform for Private AI Cloud, now uses a "deny-by-default" containment model. It isolates credentials to block prompt injection and gates unauthorized network access. The platform processes structured and unstructured data on-site to improve agent accuracy and cut token costs.

A developer harness ships with pre-approved agent skills, workflow buildpacks, human-in-the-loop controls, integrated memory services, and a marketplace of vetted AI models and tools. The AI gateway logs every agent action for compliance and auditing.

Where does VCF 9 reduce cost?

VCF 9 uses NVMe memory tiering that VMware executives said can cut per-host costs by up to 42%. The VMware AI Assistant interface addresses complex issues across CPUs, memory, and hypervisors.