Broadcom announced VMware Cloud Foundation 9.1, a new version of its private cloud platform that delivers a clear message to businesses: run AI workloads in production within controlled infrastructures, with enhanced security, more hardware options, and a promise of cost reduction compared to models relying solely on public cloud.
This update comes at a pivotal time for IT departments. AI is no longer limited to internal testing or small innovation projects. Many organizations are beginning to deploy inference, AI agents, and model-based applications within real business processes, forcing a reevaluation of costs, privacy, data sovereignty, performance, and operational capacity.
According to a preview of Broadcom’s Private Cloud Outlook 2026 report, 56% of surveyed organizations are running or planning to run inference in production on private cloud. The use of public cloud for the same purpose stands at 41%, reflecting a 15% year-over-year decline. The company also notes that 62% of IT leaders are highly or extremely concerned about infrastructure costs related to generative AI.
Private AI, Kubernetes, and support for mixed hardware
VMware Cloud Foundation 9.1 presents itself as a native private cloud platform for AI and Kubernetes, capable of running virtual machines, containers, inference loads, and agent-based applications on a single infrastructure layer. The goal is to prevent organizations from needing to maintain separate stacks for traditional applications, cloud-native services, and new AI deployments.
One of the key features of this version is support for mixed compute infrastructure, with AMD, Intel, and NVIDIA platforms. Broadcom aims to strengthen the concept of an open ecosystem where companies can choose CPUs, GPUs, and networking hardware without being limited to a single architecture. Notably, the integration with AMD and NVIDIA accelerators, Intel Xeon processors, NVIDIA ConnectX-7, BlueField-3, and standards-based connectivity like EVPN and VXLAN combined with Arista’s Universal Cloud Network are highlighted.
The update also includes efficiency improvements. Broadcom claims up to a 40% reduction in server costs through memory tiering techniques in clusters with mixed AI and non-AI workloads, up to 39% lower total cost of ownership (TCO) for storage thanks to enhanced compression and deduplication for AI data pipelines, and up to 46% decrease in Kubernetes operating costs. It’s important to note these figures are based on Broadcom’s internal estimates or testing; real-world impacts may vary depending on the environment.
Zero trust security and data sovereignty
Security is a major focus in the announcement. VCF 9.1 includes capabilities designed to protect models, training data, and inference loads from the infrastructure layer up to applications. Broadcom emphasizes that organizations deploying AI in production need tighter controls over privacy, intellectual property, regulatory compliance, and incident recovery.
Features include ransomware recovery for on-premises environments, integrated validation with CrowdStrike Falcon Endpoint Security, zero-trust lateral segmentation, distributed IDS/IPS protection for Kubernetes workloads, and live patching without downtime in specific scenarios. Some of these features are available through advanced services for VCF sold separately, an important detail for organizations evaluating overall adoption costs.
Sovereign recovery is also a key aspect. For regulated sectors or companies with sensitive data, avoiding cross-border movement of models, internal information, or training data can be essential. Broadcom positions VCF 9.1 as an alternative for organizations seeking to leverage AI without ceding control over data location or governance.
Reducing fragmentation for traditional workloads and new applications
VCF 9.1 also aims to simplify the deployment of modern applications. The platform includes scalability and performance improvements in Kubernetes, with a 2.6x increase in cluster scale, 70% faster deployments, and 75% shorter update windows compared to beta versions, according to Broadcom.
AI observability is another highlighted feature. The platform offers metrics such as time to first token, token generation performance, and GPU utilization across different accelerators. For organizations investing in expensive hardware, these metrics can help assess whether infrastructure is being utilized effectively or if there’s underutilized capacity.
Broadcom also introduces application stack blueprints for live environments, designed to capture multi-VM setups as reusable templates. This can reduce configuration errors and accelerate the creation of development, testing, and production environments—particularly valuable for platform teams handling numerous internal requests.
This release aligns with a clear trend: enterprise AI will be driven not just by model power, but by the ability to run those models securely, governably, and cost-effectively. VCF 9.1 reflects Broadcom’s response to this landscape, emphasizing private cloud as the natural environment for production AI workloads where data control, security, and cost predictability matter as much as elasticity.
Frequently Asked Questions
What is VMware Cloud Foundation 9.1?
It’s Broadcom’s latest version of the private cloud platform designed to run production workloads, Kubernetes, virtual machines, and AI within controlled enterprise infrastructures.
Why does Broadcom associate VCF 9.1 with private AI?
Because many companies want to deploy inference and AI agents without moving sensitive data to the public cloud, maintaining control over security, costs, compliance, and data sovereignty.
What hardware does VCF 9.1 support for AI workloads?
Broadcom highlights support for AMD, Intel, and NVIDIA platforms, along with integration with high-performance networks and standards like EVPN and VXLAN for scalable AI deployments.