VMware Cloud Foundation 9.1 went GA in May 2026, the whole platform is moving to a unified API-first model, where all fragmented automation tools across SDDC Manager, vCenter, NSX, and vSAN is now consolidated behind a single set of API contracts and that changes how you think about automation, lifecycle management, and day-2 operations.

vCenter: Patching

This is the one people have been waiting for. vCenter Quick Patch changes the patching model fundamentally — instead of updating every RPM regardless of what changed, it only touches the specific binaries that are actually different in the patch payload. The result is downtime measured in seconds or zero, not hours. VM deployments, Kubernetes operations, and API workflows keep running. You can apply a critical security fix without scheduling a maintenance window. That’s a genuine operational improvement.

A few other vCenter highlights worth knowing:

• VM hardware version moves from vmx-10 to vmx-17 — long overdue
• vCenter operations are up to 20% faster overall
• Resize vCenter via a single API call (PATCH /deployment/size) — scale compute and disk without digging through the UI
• RDU now supports online depot — connected vCenters no longer need to mount an ISO for 9.1.x patches
• File Integrity Monitoring (FIM) runs every 4 hours by default, aligned to NIST SP 800-115
• OAuth 2.0 API tokens, a new Query API for fast inventory access, and a unified vCenter Group Federated API (VGFA) that lets you manage multiple vCenter instances through a single endpoint

VCF Management Services

The biggest structural change in 9.1 is VCF Management Services — a new common runtime that consolidates the License Server, Software Depot, and Salt RaaS into one place. If you’ve ever dealt with the sprawl of managing these independently, this is a welcome cleanup.

Two other platform-level additions worth calling out: on-prem Ransomware Recovery to isolated clean rooms is now built in, and Continuous Compliance Enforcement means the platform can continuously remediate drift across the full VCF stack rather than just flagging it. For regulated environments, that’s significant.

And if you’re planning an upgrade, Broadcom released an open-source upgrade planning tool on GitHub specifically for 9.1. Given how many different upgrade paths exist — standalone vSphere, mixed vSAN/NSX environments, full-stack VCF — it’s worth running before you commit to anything.

Compute: ESX Gets Smarter

Zero Touch Provisioning (ZTP) is the replacement for Auto Deploy

On the security side, User-Level Monitor (ULM) is now the default for all VMs. It completely reimplements the VM monitor outside of privileged kernel mode, which meaningfully reduces the hypervisor attack surface.

For AI and HPC workloads, NVIDIA ConnectX-7, BlueField-3, and AMD MI350 GPU support lands with Enhanced DirectPath — and critically, you still get vMotion, Storage vMotion, and Live Patch with these devices. That’s a big operational improvement over traditional passthrough.

Storage: vSAN Gets Serious About Efficiency

Two big additions to vSAN ESA in 9.1.

Global Deduplication now operates across the entire cluster, not just per-disk-group. It runs continuously in the background, works in encrypted environments, and combined with enhanced compression, Broadcom is claiming up to 39% lower storage TCO. For regulated industries where encryption is non-negotiable, the background encrypted dedup support is particularly relevant.

Enhanced NVMe Memory Tiering is the one to watch for AI and data-intensive workloads. The idea is simple: hot data lives in DRAM, colder pages are intelligently offloaded to NVMe. In practice, 20–25% of memory accesses are offloaded without impacting application responsiveness. Native RAID1 mirroring protects the NVMe tier, and config changes no longer require a host reboot. Broadcom is claiming up to 40% lower TCO through the higher consolidation ratios this enables.

NFS v4.1 VAAI-NAS Unmap : finally arrives (space reclamation on thin-provisioned NFS v4.1 datastores),

Native S3 Object Storage is still tech preview but is landing in an upcoming 9.1.x release — removing the need for a separate object storage solution like Mimio S3 for DSM workloads that need it.

Kubernetes: VKS Gets the Scale Treatment

VKS sees some of the biggest numbers in this release. Support for up to 500 Kubernetes clusters per Supervisor, 70% faster cluster provisioning, and multi-network support so cluster nodes can have multiple vNICs isolating application, storage, and management traffic. That last one matters for anyone running production Kubernetes where network segmentation is a requirement, not an afterthought.

Linked Clone Fast Deploy is also worth mentioning — Ubuntu workloads spin up in seconds, which is directly useful for CI/CD pipelines and dev environments where fast instance provisioning is a daily concern.

DRS and vMotion: Under the Hood Improvements

vMotion cuts source CPU utilization by around 67% and destination by 22%

The Streaming vMotion Orchestrator removes the old static 8-concurrent-migrations limit and replaces it with dynamic concurrency tuned to your hardware

Non-disruptive maintenance mode is also new — DRS now tells you why a VM can’t be evacuated rather than forcing a disruptive evacuation that degrades performance.

A Few More Things Worth Knowing

The vSphere Client gets an Event Tracking page that gives you a proper timeline view when HA kicks in — host failures, failover decisions, restarts, all correlated in one place rather than pieced together from logs. VPC Transit Gateway, Traceflow, and Live Traffic Analysis are now all configurable from the vSphere Client rather than requiring a separate NSX context switch.