The message Broadcom brought to Mobile World Congress 2026 in Barcelona sounds like a warning to industry insiders: telco private clouds need to become more efficient, manageable, and “sovereign”, because hardware costs — especially memory — are no longer just a budget detail but a strategic risk. With this in mind, the company announced VMware Telco Cloud Platform 9, an evolution of its private cloud platform for operator data centers that, according to the announcement, is built on VMware Cloud Foundation 9 and adds specific capabilities for telecommunications.
The ambitious promise, packaged in figures, is as follows: estimated total cost of ownership (TCO) savings of 40% over five years compared to siloed architectures, energy consumption reduction of 25–30% thanks to denser virtual machine deployment and better server performance, 38% less TCO in memory and servers using Advanced NVMe Memory Tiering, and 38% less TCO in storage through global deduplication in vSAN ESA. Broadcom frames these numbers as estimates from an ACG Research report and internal testing/estimations, which are subject to change.
Despite the marketing, the reality is clear: in practice, telcos are attempting to juggle between 4G/5G (and what comes next), the rise of Artificial Intelligence workloads, and increasing regulatory pressure which, in Europe, has turned data sovereignty into a contractual concern.
Unifying 5G and AI workloads on a single “horizontal layer”
Broadcom’s proposal emphasizes one idea: operators no longer want a fragmented infrastructure divided into domains (core, edge, analytics, corporate IT) with different technologies and processes. They want a horizontal foundation over which to run network functions (4G/5G Core) and also AI workloads that consume increasing amounts of data and compute capacity.
In the words of Paul Turner (VMware Cloud Foundation Division), the goal is to reduce CAPEX and OPEX at a time when “hardware costs are skyrocketing,” and the demand for memory driven by AI will continue pushing server prices upward.
“Monetizing AI” from a telco: private AI-as-a-Service and shared GPU resources
One of the most significant shifts in the announcement is the attempt to bring the discourse of monetization into a traditionally conservative area. Broadcom proposes that Telco Cloud Platform 9 incorporate capabilities for operators to offer:
- Private AI-as-a-Service: native tools like Model Store, Model Runtime, and vector databases to deploy turnkey AI environments with data isolation and regulatory compliance.
- Virtualized GPU: partitioning a physical GPU among multiple virtual machines to increase utilization and reduce costs, instead of dedicating hardware per customer or service.
- GPU-as-a-Service (multi-tenant): on-demand access to virtualized GPU power, with logical data separation.
- Advanced real-time GPU/vGPU monitoring to avoid manual management of acceleration inventory.
- Automated Lifecycle Management (LCM): deploying private AI environments “in minutes instead of weeks.”
- Agent Builder Service: a low-code framework for building AI agents by orchestrating models, information retrieval, and tools.
This segment reflects a market intuition: if telcos want to capture value beyond connectivity, they must turn part of their distributed infrastructure into a sellable product, especially for highly regulated sectors that cannot (or prefer not to) send data outside certain borders.
Efficiency: “extended” memory with NVMe and global deduplication in vSAN ESA
The most “engineering-focused” aspect relies on two key innovations:
- Advanced NVMe Memory Tiering: leveraging high-speed NVMe storage to “extend” memory and increase load density without matching DRAM growth. In environments where DRAM becomes an economic bottleneck, this technique aims to improve cost per VM and per service.
- vSAN ESA Global Deduplication: global deduplication to reduce storage footprint by eliminating duplicate blocks at the cluster level, optimizing the utilization of existing capacity.
Broadcom also supports high-core-density CPUs and includes Intelligent Resource Scheduling to optimize performance per watt, particularly relevant for 5G and AI workloads.
Operations: GitOps, unified observability, and live patching in ESX
In telecommunications, the challenge isn’t just deploying but operating without disruption. That’s why the announcement includes automation and operational tools:
- GitOps automation: enabling ETSI or unified GitOps blueprints with ArgoCD to synchronize network functions and infrastructure from a “single source of truth.”
- Centralized dashboard for fleet management, cost control, and licensing, complemented by Network Fabric Observability to correlate issues across physical and virtual layers from core to edge.
- Continuous Operations (Live Patching): ESX live patching to apply critical updates without maintenance windows or VM interruptions, suitable for high-availability scenarios.
- AI-assisted operations, with a “human-in-the-loop” approach: recommending actions without automating critical decisions without supervision.
Sovereignty as a requirement: keys, jurisdiction, and audit evidence
The announcement positions Telco Cloud Platform 9 as a “sovereign-ready” infrastructure, aligned with the growth of digital sovereignty frameworks and laws. In Europe, the EU Cloud Sovereignty Framework formalizes goals and assurance levels (SEAL), referencing initiatives like Gaia-X and cybersecurity standards. Broadcom lists “planned” capabilities to support sovereign deployments:
- In-jurisdiction operations: managing data and control planes within defined borders.
- Cryptographic authority: exclusive local control over encryption keys.
- Audit trail: logging and automation for verifiable compliance testing.
- Automated compliance: using hardened kits and Kubernetes policy management via OPA.
- Centralized SecOps: confidential computing enclaves (AMD/Intel) and microsegmentation with vDefend to reduce lateral movement.
Summary table of what’s announced for Telco Cloud Platform 9
| Block | What it includes | Broadcom’s promised benefits |
|---|---|---|
| Cost & Energy | VM density, intelligent scheduling | 40% lower TCO over 5 years; 25–30% less energy consumption |
| Memory | Advanced NVMe Memory Tiering | 38% less TCO per memory/server |
| Storage | vSAN ESA Global Deduplication | 38% less TCO in storage |
| AI as a Service | Model Store/Runtime, vector databases, LCM | Faster deployment and isolation |
| GPU | Virtualization, GPU-as-a-Service, monitoring | Better utilization, less dedicated hardware |
| Operations | GitOps (ArgoCD), observability, live patching | Fewer maintenance windows, increased consistency |
| Sovereignty | Jurisdiction management, keys, audit evidence, OPA | Enhanced compliance and local control |
Frequently Asked Questions
What is VMware Telco Cloud Platform 9, and what is it for?
It’s a private cloud platform designed for telco data centers to run 4G/5G network functions and AI workloads on a unified foundation.
What does “40% less TCO over five years” mean?
It’s an estimated savings in total ownership costs compared to siloed architectures, based on Broadcom’s cited analysis and automated platform approach.
Why is NVMe Memory Tiering so frequently mentioned?
Because it aims to increase density using NVMe to extend memory capacity, reducing reliance on expensive and critical DRAM in AI environments.
What does the “sovereign” approach add for a telco?
Capabilities to keep data and operations within specific jurisdictions, control encryption keys locally, and generate audit evidence for regulatory compliance.
via: Broadcom