The news
On August 31, 2026, at VMware Explore 2026 in Las Vegas, Broadcom (AVGO) introduced VMware Private AI Cloud, an approach for running AI inference, AI agents and traditional workloads on one private platform, aimed at enterprises worried about AI costs and data control.
The offering is built on VMware Cloud Foundation (VCF), Broadcom's private cloud software. Broadcom said it targets three cost drivers: hardware spending, operational complexity and token economics, meaning the cost of the tokens, or chunks of text, that AI models process. Features include token monitoring, sharing one model across multiple tenants, GPU usage tracking and an AI metrics dashboard.
Broadcom also announced VMware AI Factory, which it called the software-defined foundation of Private AI Cloud. The company said VCF's automation can cut the time from deploying a bare-metal server to serving a first AI model from weeks to hours. AI Factory pairs VCF with certified Dell servers and VCF AI ReadyNodes from Cisco, Lenovo, Supermicro and others, plus a collaboration with AMD on its Instinct GPUs. According to Network World, AI Factory is not a separate product and does not cost extra.
Broadcom said VCF customers can run more than 150 open source and commercial models, and that models from Google, NVIDIA, NEC, Alibaba Cloud and Z.ai, including Gemma 4, Nemotron 3 and GLM 5.2, had been validated on VCF. It cited its own Private Cloud Outlook 2026 survey, which found 56% of enterprises running or planning to run production AI inference on private cloud.
The company made AgentMinder generally available the same day. The software acts as a central control plane for AI agents: it treats each agent as an identity tied to a declared mission, approved tools and authorized resources, checks each tool call at runtime and logs agent activity for audit. Broadcom said it runs AgentMinder internally at peaks of nearly 36 million customer-related and seven million workforce-related API calls a day.
Not everything is shipping yet. Broadcom listed an AI Gateway with token and usage rate limits, and secure sandboxes for code that agents generate, among new and forthcoming services.
The numbers
- Models VCF customers can run (Broadcom)
- More than 150
- Newly validated model families named
- 5 (Nemotron 3, Gemma 4, cotomi, Qwen3.8-27B, GLM 5.2)
- Bare metal to first model (Broadcom claim)
- Weeks cut to hours
- Enterprises running or planning production inference on private cloud (Broadcom survey)
- 56%
- Daily API calls AgentMinder handles at Broadcom, peak (company figure)
- Nearly 36 million customer-related plus 7 million workforce-related
Why CEOs should care
For CIOs and CTOs deciding where inference should run, this gives existing VMware shops an in-house option that does not require a new platform. Separate what ships now from what is forthcoming, such as the AI Gateway, and ask which of the 150-plus models Broadcom will support in production rather than merely run. Run a pilot with your own workloads before accepting the weeks-to-hours deployment claim.
CFOs should model the trade-off directly: on-premises inference swaps per-token cloud bills for GPU purchases, power and staff. The case depends on utilization, so ask for total-cost comparisons built on your expected usage, not vendor benchmarks. Also weigh that building an AI stack on VCF deepens dependence on Broadcom, a vendor many VMware customers already view negatively, as the survey data below show.
CISOs should look closely at AgentMinder, because agent identity and permissions are becoming a real control gap. Test how it fits your existing identity and authorization tools, which Broadcom says it supports through the AuthZEN standard, and whether its audit logs meet your compliance needs. Boards should ask management who is accountable for AI agent governance, whichever vendor supplies the tools.
The bigger picture
Broadcom's pitch runs against some market data. Network World, citing an Omdia survey from August 2026, reported that 59% of inference workloads currently run in the cloud and that companies expect that share to reach 63% in three years. The same article cited a Gartner survey from April 2026 in which 76% of IT leaders and CIOs had a negative outlook toward Broadcom's ownership of VMware, up from 64% in 2025 and 33% in 2024.
Rivals are also courting VMware customers. Network World reported that Nutanix added 3,000 customers over the past year and has about 10,000 in total, against more than 200,000 for VMware. Private AI gives Broadcom a new reason for that installed base to stay.
What happened next
On September 2, 2026, Broadcom reported that infrastructure software revenue rose 29% to $8,752 million in its fiscal third quarter ended August 2, 2026. In its 10-Q report, dated September 9, 2026, it attributed the segment's growth primarily to strong demand for VCF. Broadcom also said on September 1 that VMware Explore 2027 will move to May 3-6, 2027, at Resorts World Las Vegas.
What to watch: release dates for the AI Gateway and agent sandboxes, and whether partners follow through. Network World reported that Flexential, a VMware partner, plans offerings built on VMware AI Factory for early 2027.
