The Network’s Role Is Changing. That Changes Everything
Contributed by Anupama Mahabhashyam, Broadcom.

For most of the Internet era, the network’s primary job was transport: move data reliably between endpoints, at scale, with as little disruption as possible. Telcos built extraordinary operational disciplines around that objective. Five-nines availability, rigorous change management, meticulous vendor certification processes. These were not bureaucratic habits. They were the architecture of trust on which the digital economy depends.
That objective has not gone away. But it has been joined by something more demanding. The emergence of the industrial AI cloud means that compute is no longer centralized. AI inference, edge workloads, and distributed training jobs are spreading across a landscape that now includes traditional data centers, private clouds, and thousands of edge sites simultaneously. The network is no longer just connecting those environments. It is increasingly the base on which they depend.
This is a meaningful shift in kind, not just in scale. When the network becomes the backbone of distributed compute, the expectations placed on it change. Latency, observability, security posture, and configuration compliance are no longer network operations concerns alone. They become AI reliability concerns. And the operational frameworks telcos have built, however robust, were not designed with this in mind.
AI is not an abstract future possibility. It is an operational reality that network teams are already managing, with infrastructure and tooling designed for a different era. That gap between operational capability and operational requirement drives the discussion that follows.
Sovereignty Is Part of the Picture
Regulatory and geopolitical pressures are also reshaping where AI workloads can run and who can operate them. Telcos, already accustomed to operating under national regulatory frameworks, are naturally positioned to provide sovereign cloud infrastructure for enterprises that need to keep data and AI workloads within defined jurisdictions. This is an extension of what telcos already do, not a new capability they need to build from scratch.
Governance and compliance culture are already embedded in how telcos manage application and infrastructure lifecycles, security standards, and regulatory obligations. Physical presence at the edge, the result of sustained investment in radio, fiber, and facility infrastructure, is something most technology companies cannot duplicate.
While hyperscaler outages make headlines and disrupt millions of users in an instant, the telecom industry quietly built something the cloud world is still catching up to: carrier-grade reliability. Designed for five-nines uptime across sprawling, multi-vendor networks, this discipline of fault tolerance, redundancy, and graceful degradation is precisely the infrastructure foundation that agentic AI systems demand.
For enterprises that need AI infrastructure but lack the on-premises capacity to run it, telcos are a natural partner. This extends beyond GPU availability. Inference workloads run across a range of compute options, including CPUs, making the breadth of compute capacity that telcos can offer across their distributed, sovereign infrastructure all the more valuable. Sovereign compute, paired with the right hardware mix, becomes a genuine competitive differentiator for enterprises operating under regulatory constraints.
The DevSecOps Challenge: Fragmented Compliance and Security Data
Bridging the gap between where infrastructure stands today and what the industrial AI cloud demands is a genuine engineering and organizational undertaking, and it is worth being specific about where the difficulty actually lies.
The challenge is not primarily a skills problem or a strategy problem. It is a DevSecOps data problem. Telcos managing large, heterogeneous networks face a compounding challenge: millions of subscribers, across dozens of domains and vendors, each with its own lifecycle timelines, security advisories, and configuration requirements. Keeping that information current is a significant engineering undertaking in its own right, before any AI ambition is added.
The result is a compliance and security data bottleneck. Engineers spend significant time on manual research, cross-referencing vendor documentation, validating lifecycle data, and assessing device vulnerability across vertical siloed stacks, work that requires consistency and attention but not the kind of judgment that justifies the cost of experienced engineers doing it.
Technical debt accumulates in this environment. Devices approach end-of-life without coordinated remediation plans. Configuration drift goes undetected. Security posture degrades incrementally. Without a unified, continuously maintained view of the network’s compliance and security state, the infrastructure layer that AI operations depend on remains fragile. This is precisely the problem that purpose-built DevSecOps tooling for network security and compliance is designed to solve.
The Operational Impact: Shifting from Reactive to Continuous
Much of the early discourse around AI in network operations focused on conversational interfaces – ask a question, get an answer. That capability has value, but it is not the primary opportunity. The more significant shift is toward agentic AI: systems that can monitor continuously, reason across complex datasets, initiate multi-step workflows, and verify outcomes for mission-critical networks, operating within defined guardrails that keep humans in control without requiring intervention at every step.
The distinction matters operationally. A tool that answers a question requires an engineer to ask the question at the right time. An agent that monitors OS lifecycle state across the system, correlates it against known vulnerability data, and surfaces prioritized remediation recommendations on an ongoing basis does not depend on an engineer remembering to check. It changes the operational posture from reactive to continuous.
AI also changes what “normal” looks like on the network. Distributed AI workloads have traffic signatures, east-west communication patterns, and latency sensitivities that differ from traditional application traffic. Network operations teams accustomed to particular baselines will find those baselines shifting as AI workloads grow. Detecting anomalies, diagnosing degradation, and maintaining application experience in this environment requires observability tooling that can adapt to new patterns, not just measure against historical ones.
Importantly, this does not require building new infrastructure from scratch, or hiring teams of specialists who did not exist in the organization before. The value is in making existing infrastructure and existing expertise more effective, by giving both access to better, more unified information.
Infrastructure Must Be Coherent Before Automation Can Act
Any serious conversation about operationalizing agentic AI in a network environment arrives at the same prerequisite: the infrastructure layer must be coherent before automation can act on it reliably. Fragmented infrastructure produces fragmented data. Fragmented data produces unreliable automation. In a carrier-grade environment, unreliable automation is not an acceptable outcome.
VMware Telco Cloud Platform built on VMware Cloud Foundation is a software-defined horizontal platform that helps telcos and businesses build, run, and manage AI-ready workloads along with VNFs and CNFs across complex, multi-vendor environments. By providing a common operational layer across data centers and private clouds it turns infrastructure heterogeneity from an operational liability into something the organization can consistently manage and leverage.
Existing infrastructure can be enrolled into unified management and automation frameworks. Existing observability data can be unified rather than left in separate systems. Existing engineering teams can be supported by agentic workflows that handle the repetitive, data-intensive tasks that consume disproportionate time.
Visibility Is Not a Feature. It Is the Enabling Condition
Infrastructure consolidation addresses one dimension of the AI readiness gap. But unified infrastructure alone isn’t sufficient. The other is insight. Telcos operating at scale lack the ability to unify disparate data sources into a coherent operational picture, particularly as infrastructure extends into cloud edges that sit outside traditional management boundaries.
Network Observability by Broadcom provides end-to-end visibility across physical underlay, virtual overlay, cloud environments, and application experience simultaneously. They extend visibility to the parts of the estate that are hardest to see: unmanaged cloud edges, external network segments, and the application experience layer that determines whether workloads are getting what they need.
This is where agentic AI becomes practically deployable. An agent operating on unified, high-quality observability data can assess OS compliance across the full estate, continuously correlate configuration state with performance anomalies, and flag security exposure based on current vulnerability data. The quality of what an agent can do is a direct function of the quality of the data it works with.
Looking Ahead
The AI infrastructure transformation is just beginning. As the size of AI models and the complexity of workloads increase, data center networks will face ever-growing demands.
This transformation requires three converging capabilities: infrastructure that can adapt to agentic workloads, observability that provides continuous insight across hybrid environments, and automation that acts on that intelligence reliably. When these elements align, the network evolves from reactive troubleshooting to predictive, self-correcting operations.
Broadcom’s position across the stack enables telcos to transform their operations without abandoning their investments. The network’s new role demands new operational capabilities, but it builds on foundations telcos already have. The question is no longer whether telcos can adapt to the AI era. It’s how quickly they can operationalize the advantage they’re positioned to deliver.



