First published: 06-May-2026 13:49:42

Dubai, UAE – Preliminary results from Broadcom’s “Private Cloud Predictions 2026” report, published by Broadcom, a leading global technology company that designs, develops, and supplies semiconductor and infrastructure software solutions, show that private cloud computing continues to solidify its position as the preferred option for running AI applications in actual production environments. More than half of the organizations surveyed (56%) reported that they are using or planning to use private cloud computing to run their AI production inferences. In contrast, the percentage of public cloud computing used for AI production inferences was 41%, a 15% year-over-year decrease, reflecting a growing shift towards more controlled and secure private environments. Furthermore, 62% of IT leaders expressed significant concern about the infrastructure costs associated with generative AI, while 36% reported that AI applications impose new requirements related to data protection, privacy, security controls, and risk management.

In this context, Broadcom today announced the launch of VMware Cloud Foundation (VCF) 9.1 , the latest version of its integrated platform for building and managing private cloud environments. This platform provides a more cost-effective and secure infrastructure for running AI-powered applications and productivity workloads. VCF 9.1 offers a private cloud environment designed to support AI and Kubernetes applications, with built-in security capabilities and support for a hybrid computing architecture that includes AMD and Intel processors, and NVIDIA AI GPUs . This enables organizations to deploy and distribute AI inference and proxy AI applications more efficiently and at a lower cost, while also enhancing security and giving them greater flexibility in selecting the best processors and GPUs to suit their operational requirements and technical objectives.

The VCF platform offers a more efficient alternative to public cloud computing for running productivity applications and workloads. Through intelligent software, it enhances the efficiency of existing server infrastructure and gives organizations greater control over architecture, data sovereignty, regulatory compliance, and governance—all essential elements for deploying and distributing AI applications in real-world production environments. VCF 9.1 enables organizations to run productivity applications and workloads, including AI inference and AI proxy applications, while achieving the following:

• Reduce server costs by up to 40% through intelligent management of memory layers in clusters running a mix of AI-related and non-AI-related applications and workloads.
• Reduce the total cost of ownership for storage by up to 39% by improving compression techniques and removing duplicate data within AI data paths.1
• Reduce the operating costs of software container technologies by up to 46% when running AI-related applications and workloads on a large scale.1
• Accelerating the updating of collections by a factor of 4, while doubling the capacity of the infrastructure, thereby accelerating the scaling up of AI applications.1

Krish Prasad, senior vice president and general manager of Broadcom's VCF division, said: "As more organizations adopt AI technologies to enhance their competitive advantage, they face three key challenges: data privacy and intellectual property concerns, rapidly rising infrastructure costs, and their readiness for an agent-driven AI world. VCF 9.1 is a single, unified platform that addresses these three challenges and provides advanced proprietary AI infrastructure. We support zero-trust AI applications, reduce costs through intelligent infrastructure optimization and hardware choice, and give organizations the flexibility to run agent-driven workflows and AI-enhanced inference processes on the same platform."

A more efficient and large-scale infrastructure for AI applications and workloads

The VCF 9.1 platform enhances the operational intensity of AI applications and workloads, whether virtual machine ( VM ) or containerized, within existing infrastructure, significantly reducing operational complexity. Through intelligent resource management and automated processes, the platform enables organizations to run a greater number of productive applications and workloads on existing servers, scale efficiently across distributed environments, and eliminate the need for costly infrastructure expansions at a time of hardware shortages and rising costs. Key capabilities include :

• Intelligent resource optimization enhances infrastructure utilization efficiency through advanced memory tiering and next-generation storage compression technologies within AI data paths to increase application density and workloads without impacting performance or requiring costly hardware upgrades.
• Automated processes for managing large-scale infrastructure double the administrative capacity to 5,000 hosts and accelerate cluster updates by 4x across distributed and isolated environments, eliminating the burden of manual updates and supporting rapid expansion of AI infrastructure.
• A multi-user infrastructure for isolating AI applications enables organizations and service providers to run multiple projects and clients on a shared infrastructure within strict security boundaries, maximizing the utilization of costly CPU and GPU resources, while supporting the data sovereignty of sensitive models.
• Integration with an open ecosystem provides the freedom to choose between AMD and Nvidia AI GPU accelerators, along with support for leading AMD and Intel CPU platforms, and standards-based compatibility with EVPN and VXLAN technologies via the Arista Universal Cloud Network , reflecting VCF's commitment to providing the high-performance connectivity and computing flexibility required for productive AI applications.
• High-speed networks for AI applications and workloads through VCF support for NVIDIA’s ConnectX-7 and BlueField-3 network cards, with Enhanced DirectPath I/O enabling faster access, allowing multi-user AI model training and high-speed data transfer, which is essential for demanding generative AI applications and workloads.
• Balancing virtual applications and loads and providing security protection through VMware Avi Load Balancer and VMware vDefend 2 eliminates the need for dedicated physical hardware for AI heuristics and agent-based AI applications, reducing capital expenditures while providing enterprise levels of flexibility and automated lifecycle management.

Accelerated application delivery: A modern platform for running AI, software containers, and virtual servers

VCF 9.1 provides a unified platform to accelerate the deployment and distribution of AI applications by running AI applications and inference loads, agent-driven AI applications, container-based services, and traditional virtual servers within a single infrastructure layer. This reduces operational fragmentation and the costs associated with managing separate environments, while providing developers with the speed and platform governance needed to run AI applications in production environments. Key capabilities include:

• Expanding the scope of Kubernetes technologies and improving their performance for artificial intelligence applications , which contributes to increasing the capacity of collections by 2.6 times, accelerating deployment and distribution processes by 70%, and reducing update periods by 75% compared to previous beta versions, with the possibility of smooth expansion without interruption to production services.
• Unified management of hybrid computing enables efficient handling of agent-based AI workflows that rely more heavily on CPUs, along with AI-enhanced inference processes ( GPUs ), reflecting the actual nature of these applications that require greater processing and decision-making capabilities.
• Advanced AI monitoring and governance capabilities provide accurate metrics such as first token issuance time, token processing rate, and GPU utilization levels across multiple accelerator types, helping organizations maximize the return on infrastructure investments through accurate monitoring of hardware usage, along with the application of centralized policies and data sovereignty controls to ensure compliance with legislation and secure access to models.
• Live, direct operating schemes for application packages enable the conversion of multi-server virtual applications into reusable templates for rapid environment deployment, reducing manual setup errors, preventing configuration differences between development, testing, and production environments, and accelerating infrastructure delivery.

A zero-trust security architecture to ensure the sovereignty and governance of AI data.

The VCF 9.1 platform integrates security protection within the same infrastructure layer to safeguard AI applications and workloads, proprietary models, and training data, from the virtualization layer to the application level. By providing zero-trust security partitioning, sovereign recovery capabilities, and continuous updates without the need for separate additional tools, the platform delivers a level of protection required to run AI applications in production environments that surpasses what public cloud computing environments offer. These capabilities include:

• Local recovery from ransomware attacks , through isolated recovery environments and built-in verification tools, along with new support for CrowdStrike Falcon® Endpoint Security solutions, protects AI models and training data, which represent high-value intellectual assets, from cross-border transfer risks while avoiding high data packet charges during emergency recovery operations.
• Continuous application of compliance requirements for legislation , through centralized monitoring and automated processing of the targeted operational status of applications, workloads and components of the VCF platform, enables organizations to demonstrate their readiness for audits of AI applications operating in production environments, without manual burdens or the need for separate compliance tools.
• Live and uninterrupted security updates , supporting up to 80% of use cases without evacuating hosts or the need for maintenance windows, ensuring the continuity of productive AI inference services and AI proxy applications that require constant availability under service level agreements.
• Zero-trust lateral protection , for the first time, expands the distributed protection of Intrusion Detection and Prevention Systems (IDS/IPS) to include AI-specific Kubernetes applications and workloads, with threat inspection capabilities of up to 9 terabits per second for AI distributed inference, and a 5x increase in application recognition within private cloud computing and internet applications.
• Automated, self-service security protection provides centralized classification, ready-made security profiles, delegated firewall settings, and web application protection at points of entry, enabling organizations and service providers to secure AI applications without operational complexity or multiple separate security tools.

“Analyzing years of news archives via public cloud computing is extremely costly, and the unpredictable pricing makes planning AI projects more complicated,” said V.V. Jacob, Senior General Manager of Systems at Malayala Manorama Co Ltd. “By deploying and distributing VCF Private AI Services on our existing VMware Cloud Foundation platform infrastructure, we will be able to run AI-powered tools for summarizing content, generating headlines, and providing editorial support directly through our private cloud. We believe this provides us with the necessary levels of privacy and security to protect editorial sources, along with greater cost clarity that a local private cloud infrastructure offers.”

Alexander Hopfgartner, Head of Technology at Notruf Niederösterreich , said: "We have been able to achieve greater operational efficiency and increase the overall availability of services by unifying virtual machines ( VMs ) and software containers on the VMware Cloud Foundation platform. As the integrated Kubernetes operating environment in the VCF platform, VMware vSphere Kubernetes gives our operations team the ability to easily deploy, distribute, scale, and manage our critical applications."

Kumaran Seva, vice president of enterprise AI and computing at AMD , said: “As enterprise AI projects move from the experimental to the field production stage, the need for an infrastructure that delivers performance, efficiency, and flexibility across a large-scale ecosystem is becoming increasingly urgent. AMD’s enterprise AI solutions, integrated with the VMware Cloud Foundation 9.1 platform, enable AI workloads to run efficiently and at low cost, helping customers deploy and distribute AI inference and proxy AI applications while ensuring the good performance, advanced security, and data sovereignty required in production environments.”

Jeff Raymond, Vice President and General Manager of Software and Services (EOS) at Arista Networks, said: “Arista Networks and Broadcom share a firm commitment to open, standardized networks that give companies complete freedom to choose the right architecture for their production AI infrastructure. The interoperability of EVPN and VXLAN technologies between Arista Universal Cloud Network and the VMware Cloud Foundation 9.1 platform provides the levels of openness and performance that production AI requires. With direct, standards-based connectivity between ESX servers and fabric , organizations can rely on a scalable AI infrastructure network architecture while simultaneously reducing capital and operating costs.”

Chris Stewart, Vice President of Global Cloud and Technology Partnerships at CrowdStrike , said: “Today, AI applications and workloads are prime targets for cyberattacks, and performing recovery operations without proper verification is a risk that organizations cannot afford. By integrating CrowdStrike with the VMware Cloud Foundation platform, organizations will be able to stop intrusions more quickly, ensure the complete integrity of operational environments before recovery, and prevent re-infection, which is essential for protecting high-value models and data, maintaining data sovereignty, and meeting regulatory compliance requirements.”

“The VMware Cloud Foundation 9.1 platform has been further enhanced to be more efficiently compatible with Intel® Xeon® 6 processors, enabling full utilization of a high-density, AI-ready platform,” said Caitlin Anderson, Intel’s executive vice president of sales for the Americas. “Native integration with Intel® QuickAssist Technology also accelerates Encrypted vMotion operations, while freeing up valuable additional computing resources. Through this collaboration, we continue our commitment to delivering continuous innovation and higher total cost of ownership, helping customers accelerate their efforts to modernize AI applications and containerized environments.”

John Fanelli, Vice President of Enterprise Software at NVIDIA , said : " Businesses need an infrastructure that delivers exceptional AI performance while maintaining data sovereignty and control. Our collaboration with Broadcom will combine the Blackwell architecture, including RTX Pro servers with BlueField-3 network cards and the Blackwell HGX platform, along with DirectPath I/O technology for faster access, with the VCF platform. This will enable organizations to deploy and distribute private AI solutions with the same level of performance as public cloud computing, but with complete control over their models and data. This collaboration delivers superior computing power and enterprise-level governance, meeting the requirements of production-oriented AI."

Additional resources

Read all the blogs related to the VMware Cloud Foundation 9.1 platform to learn about the latest innovations and new features .

Learn more about VMware Cloud Foundation

Follow VMware Cloud Foundation channels on LinkedIn, X , and YouTube.

About Broadcom Inc.

Broadcom, Inc. (NASDAQ: AVGO ) is a global technology leader that designs, develops, and supplies semiconductor and software infrastructure solutions that meet the complex and critical needs of global enterprises .

Broadcom combines long-term investment in research and development with excellence in execution to deliver best-in-class technology solutions at scale. The company, a registered corporation in the State of Delaware, is headquartered in Palo Alto, California. For more information, please visit www.broadcom.com

-I finish-

#Company Data

Disclaimer regarding the content of press releases
The content of this press release is provided by a third-party provider. We do not assume any responsibility for, nor do we have any control over, such content. This content is provided on an "as is" and "as available" basis and is not edited in any way. Neither we, nor our affiliates, will be liable for the accuracy, endorsement, or completeness of any opinions, views, information, or materials contained in this content.
This press release is provided for informational purposes only; the content does not constitute legal, investment, or tax advice, nor does it offer any opinion on the suitability, value, or profitability of any particular portfolio or investment strategy. Neither we nor our affiliates will be liable for any errors or inaccuracies in the content, or for any actions you take based on that content. You expressly agree and acknowledge full responsibility for your use of the information contained in this press release.
To the extent permitted by applicable law, Refinitiv, its parent company, subsidiaries, affiliates, relevant shareholders, directors, officers, employees, agents, advertisers, content providers, and licensors (collectively, the 'Refinitiv Parties') shall not be liable (jointly or severally) to you for any direct, indirect, consequential, special, incidental, punitive, or exemplary damages; This includes, but is not limited to, losses of profits, savings, or revenues, whether due to negligence, tort, contract, or other theories of liability, even if the parties to Refinitiv were advised of the possibility of any such damages or losses occurring or actually anticipated them.