VMware VCF 9.1 release: The biggest changes enterprise infrastructure teams need to know about

The DRAM problem gets a real

NVMe memory tiering was introduced in VCF 9.0 and it worked. The performance numbers held up, the TCO story was real, and customers were interested. But it required a reboot to enable, and it depended on hardware RAID controllers, which made adoption slower than it should have been. Too much friction for a feature that was supposed to reduce friction.

VCF 9.1 fixes that. It offers reboot-free enablement, software-based mirroring that eliminates the RAID controller dependency, and smarter cold page detection. The result is up to 40 percent lower server TCO.

Why this matters

DRAM is the single most expensive line item on a server order. In a supply chain environment where hardware costs are rising and lead times are stretching, anything that reduces your per-server cost while maintaining application performance is a lever worth pulling. This is structural, not incremental. I strongly believe this feature will become as ubiquitous in hypervisors as deduplication is in enterprise storage systems.

Storage gets real. Finally.

Speaking of deduplication. vSAN global deduplication and enhanced compression went GA in 9.1, and for the first time, they work alongside data-at-rest encryption.

This was the missing piece. No enterprise security team ever loved turning off encryption to get dedupe savings. Now they don’t have to choose. Up to 39 percent storage TCO reduction, with encryption intact. That’s a conversation that used to end at “not yet.” Now it starts with “when.”

Kubernetes is growing up on VCF

The Kubernetes improvements are significant on paper with up to 46 percent lower operational costs, faster deployment, and larger cluster scale. Those numbers matter.

But the real story is accessibility. VCF is making container management more approachable for the same ops teams that run VMs today. If your biggest Kubernetes barrier isn’t technology but talent (and for most enterprises, it is) this is the kind of platform shift that changes the math on adoption. AI workloads require this shift and as such containers usage and orchestration are becoming more ubiquitous. It is updates like these that truly make the difference.

Security extends where it needs to

Distributed IDS/IPS now extends to Kubernetes workloads for the first time with 9Tbps inspection throughput and live patching in 80% of use cases, including TPM-enabled hosts.

This is what zero trust looks like at the infrastructure layer: not a slide deck, not a checkbox, but a platform capability. AI workloads running in containers need the same security posture as VMs. Now they can get it.

Key difference between VCF 9.0 and VCF 9.1

VCF 9: The foundation (Released June 2025)

• Unified operations — Single management interface, quick Start deployment, one-click patching.
• VMs and containers and AI — Native VKS; run Kubernetes alongside VMs as first-class workloads.
• NVMe memory tiering — NVMe as secondary memory tier; up to 38% lower server TCO by reducing DRAM dependency.
• Enhanced data path — Up to 3x faster network switching; kernel optimizations and optional DPU offload.
• AI-ready infrastructure — GPU vMotion, NVIDIA Blackwell support, private AI services included at no extra cost.
• Zero-trust security — Confidential computing, vDefend micro-segmentation, IDS/IPS, ransomware recovery.
• Fleet management — Centralized lifecycle, fleet-level upgrades, license portability across on-prem/cloud/edge.

What’s new in VCF 9.1 (Released May 2026)

• NVMe tiering enhanced — Reboot-free, no RAID controllers, smarter cold page detection; up to 40% lower server TCO.
• vSAN global dedupe GA — Generally available with data-at-rest encryption; up to 39% storage TCO reduction.
• Elastic provisioning — Parallel imaging, automated host discovery, pull-based edge orchestration.
• Kubernetes at scale — 6x cluster scale, 75% faster deploy, 75% shorter upgrades (vs preview); up to 46% lower K8s ops costs.
• 4x faster cluster upgrades — Fleet scales to 5,000 hosts (2x increase); asynchronous and live patching.
• Zero-trust for Kubernetes — IDS/IPS for K8s AI workloads (9 Tbps); live patching in 80% of cases incl. TPM hosts.
• Unified AI platform — Inference and agentic AI on single layer; AMD and NVIDIA GPU support; multi-tenant isolation.
• Networking — Avi LB and vDefend replace HW appliances; EVPN-VXLAN fabric interop; automated inter-VPC firewalls.

Key efficiency gains — 9.0 versus 9.1