How VMware Cloud Foundation will help you privatize AI workloads faster

Most enterprises have deployed AI in some form, and increasingly they want to run it on their own infrastructure. But most aren’t ready… yet.

A10 Networks’ State of the AI Infrastructure Report found that 53 percent of organizations are only “somewhat confident” in their infrastructure’s ability to support future AI needs.

The result is a widening gap between AI ambition and AI infrastructure, and for most organizations, “deployed” still means a Copilot subscription or a ChatGPT enterprise license rather than a private LLM running against proprietary data.

Closing that gap usually requires a budget cycle, an infrastructure decision, and an executive sponsor. And over the next 18 months, nearly every VMware customer will have all three.

The VCF 9.x elephant in the room

VMware vSphere 8 reaches the end of general support on Oct. 11, 2027. After the EOL date, technical guidance continues for two more years, but new patches, security updates, and hardware certifications stop. So, what’s next? VMware Cloud Foundation (VCF) 9.x has been generally available since June 17, 2025, and is the upgrade path forward.

When you’re ready to make the shift, the technical migration is easy to scope. But the licensing change is a more complex discussion. vSphere 8 was the last version that could be perpetually licensed, which means moving to VCF 9.x also means moving to a VMware by Broadcom subscription. There are also hardware considerations. VCF 9.x has higher technical requirements than vSphere 8 and most current environments will need a compute refresh to run it well.

But let’s pause here. Two features in VCF 9.x are worth understanding before any budget conversation, because they change the value of the upgrade itself.

NVMe memory tiering

The first is NVMe memory tiering, which becomes critical in a market where memory supply is constrained and prices are elevated. VCF 9.x Our team of certified professionals and technical experts can help you determine how and where to leverage memory tiering.

VMware Private AI Foundation with NVIDIA

The second is VMware Private AI Foundation with NVIDIA, which is included with VCF 9.x. It gives customers a secure environment to deploy and run LLMs on their own infrastructure, with NVIDIA NeMo as the underlying framework and support. For regulated industries, NVIDIA’s AI Enterprise containers are now available in configurations that meet FedRAMP High and DoD IL5 requirements.

The private AI on-ramp most enterprises haven’t noticed

Standing up infrastructure to run private AI workloads is where most enterprise AI programs stall. The cost is real, the talent is scarce, and the procurement cycle is long. So organizations default to AI tools that don’t require infrastructure decisions, like chat interfaces and vendor APIs.

VCF 9.x changes that math because the environment for private AI comes with an upgrade that most VMware customers are already going to make. By leveraging Private AI Services (PAIS), teams can easily stand up a secure, isolated environment with an LLM, point it at internal data, run a retrieval-augmented use case, and see what works without a separate infrastructure buy or vendor relationship. From there, the path to a production AI deployment is shorter and better understood, since the platform decisions have already been made.

That doesn’t make every workload a candidate for private AI, nor does it replace the work of identifying high-value use cases or building the data pipelines to support them. But it removes one of the largest barriers to starting.

How to think about the next 18 months of VMware strategic planning for AI

For organizations running VMware today, the planning work falls into four buckets:

1. Inventory the existing hardware and identify what will and won’t meet VCF 9.x requirements. Discover hardware requirements based on capacity and desired expansion.
2. Model the licensing impact under the new per-processor minimums and the move from perpetual to subscription.
3. Align the hardware refresh against compute, memory, and AI-readiness needs.
4. Identify two or three candidate workloads that could meaningfully test private AI in the first 90 days after migration.

SHI is building a full VCF 9.x reference architecture in our AI & Cyber Labs, backed by Dell hardware investment that lets you run hands-on testing against the same stack you’re considering buying — including the NVIDIA GPU configurations needed to exercise private AI.

We are already modeling the new Broadcom licensing impact across hundreds of VMware customer environments this quarter, which means the per-core analysis isn’t theoretical. And for organizations that want to test private AI before committing to a full VCF migration, we offer an alternative entry point through NVIDIA Spark devices that can run a functioning LLM environment on a single device.

As a Dell Titanium Black partner, 2024 VMware Geo Partner of the Year (North America), 2024 VMware Technical Enablement/Support Partner of the Year (Americas), and 2026 NVIDIA Partner Network Rising Star Solution Provider Partner of the Year, we are positioned to handle the financial picture and the AI on-ramp as one project, not two.

Next steps for your VCF 9.x migration

Ready for your VCF 9.x journey? Book SHI’s VMware Strategic Assessment to start the conversation today.

We know a lot is changing in VMware.

Find out why organizations like yours are choosing SHI as their VMware partner.

FAQ

What is VMware Cloud Foundation (VCF) 9.x?

VCF 9.x is VMware’s “full stack” for running a private cloud (compute, storage, networking, and management packaged together). If you’re on vSphere today, VCF 9.x is a common next step as vSphere 8 approaches end of general support in October 2027.

Why does a VCF 9.x migration create an “AI on-ramp”?

Because the upgrade often comes with the hardware refresh and platform changes you’d need anyway to run private AI. Instead of creating a brand-new “AI infrastructure project,” you can use VCF 9.x to set up a safe test environment and prove a couple of AI use cases quickly.

What VCF 9.x capabilities matter most for AI readiness?

If you’re thinking about AI readiness, the two standouts are NVMe memory tiering (to reduce how much expensive DRAM you need) and VMware Private AI Foundation with NVIDIA (a packaged environment for testing LLM workloads privately).

What should we do first if we’re planning a VCF 9.x upgrade?

1. Inventory what you have today and determine what won’t meet VCF 9.x requirements.
2. Model the licensing change early (subscription shift and per-processor minimums).
3. Scope the refresh (compute, memory, storage — and GPUs if private AI is in scope). Align the hardware refresh against compute, memory, and AI-readiness needs.
4. Choose two or three pilot use cases to test (the Experiment phase) so you can decide what to scale (the Adopt phase). candidate workloads that could meaningfully test private AI in the first 90 days after migration.

How does SHI help with this (Imagine. Experiment. Adopt.)?

SHI helps teams Imagine the right AI outcomes through use‑case discovery and readiness workshops, Experiment by validating prototypes in the AI & Cyber Labs (often in 2-6 weeks using real data and integrations), and Adopt by moving the proven approach into production with security, governance, and compliance built in.