Talent shortages, tool sprawl, and cloud complexity vex NetOps teams as AI workloads arrive on the network, EMA research finds.

Enterprise network operations teams are struggling to keep pace with the demands placed on them, and the challenge is growing as enterprises prepare their networks and observability tools for AI workloads. According to a recent Enterprise Management Associates (EMA) benchmarking study, roughly 31% of IT professionals surveyed said their organization’s network operations strategy is completely successful, a significant decrease from 42% just two years ago. The findings, published in EMA’s Network Management Megatrends 2026 report, are based on a survey of 352 IT professionals across North America and Europe. The report confirms that network teams today face multiple pressures: a persistent talent shortage, tool sprawl, hybrid and multi-cloud complexity, and the arrival of AI workloads on networks that were not originally designed to manage them.

“Network operators clearly know they need to do better, but they aren’t getting the support they need,” said Shamus McGillicuddy, EMA’s vice president of research for network infrastructure and operations. “They need budget to fill empty seats on their teams. They need better tools. They need more automation. They need more influence over modern architectures, like hybrid and multi-cloud networks. CIOs need to step up and give network operators the support they deserve, especially if those CIOs want to succeed with AI transformation. Networks will make or break those projects.”

The state of the NOC

Tool sprawl remains a chronic condition for network operations teams. The typical IT organization uses four to ten monitoring and troubleshooting tools to manage its network, a number that EMA says has barely moved in more than a decade. Yet EMA found no significant correlation between the size of a toolset and operational success. The data shows how much room for improvement exists, regardless of how many tools a team has:

• 58% of network problems are detected proactively before users experience their impact
• Only 37% of alerts generated by network monitoring tools are indicative of a real problem
• Manual administrative errors cause 28% of network problems
• 29% of the average network professional’s day is spent troubleshooting

“IT pros believe that 53% of the network problems that they are dealing with on a day-to-day basis could be prevented with better tools, so that gives you some color around why only 31% of the people we surveyed felt like they are completely successful with network operations strategy,” McGillicuddy explained. “Tool replacement is widespread. Seventy-three percent of the people we surveyed said they are likely to replace, at least somewhat likely to replace, a network observability or network monitoring tool within the next two years.”

The Network Operations Center (NOC) has long been the nerve center of enterprise IT, responsible for ensuring connectivity, performance, and security. But as networks grow more complex with hybrid work, cloud migration, and edge computing, traditional NOC workflows are breaking down. Engineers are buried in alerts, many of which are false positives, leading to alert fatigue and missed critical incidents. The manual effort required to triage and diagnose issues drains time that could be spent on strategic initiatives. According to the EMA report, the average network professional spends nearly 30% of their day on troubleshooting, a figure that underscores the inefficiency of current tools. The industry is now at a crossroads: either invest in smarter observability and automation, or risk falling further behind as AI workloads multiply.

Megatrend 1: The talent crisis is getting worse

The share of organizations that find it somewhat or very difficult to hire network technology experts has risen sharply from 26% in 2022 to 41% in 2024 to 52% today. According to EMA, the shortage is most acute at the senior and mid-career levels, where cloud, security, and automation skills are most needed. The talent gap is not just a hiring problem; it is a retention and skills development challenge. As experienced engineers retire or move to more lucrative roles in cloud and DevOps, fewer replacements are entering the field with the requisite depth of knowledge. One monitoring architect with a Fortune 500 entertainment company summed it up: “We’re being asked to do more with less. What used to be done by a 25-person team, management now wants us to do with a ten-person team.” This sentiment echoes across industries, from finance to healthcare to retail.

The talent gap is also driving urgency around automation. Short-staffed teams need tools that can handle routine tasks automatically, so that the engineers they do have can operate at a higher level. However, the skills gap itself can be the biggest barrier to achieving that automation. Network teams cited several top barriers to automation in the EMA survey:

• Skills gaps within the team: 46%
• Tool limitations or lack of integration: 36.4%
• Insufficient data quality or visibility: 31.8%
• Risk aversion or governance constraints: 31.8%
• Budget constraints: 29.8%
• Organizational resistance to change: 27.3%
• Lack of trust in automation: 25%

These barriers create a vicious cycle: without automation, teams are overwhelmed by manual tasks, and without skilled staff, they cannot implement the automation that would free them up. To break this cycle, organizations need to invest in training, cross-skilling, and platforms that abstract complexity. The EMA report suggests that successful teams prioritize automation despite the challenges, and they view AI as a force multiplier rather than a replacement for human expertise.

Megatrend 2: The push to automate day-two operations

Network automation has historically focused on provisioning and configuration, considered day-zero and day-one work. But the new priority is day-two operations: the ongoing detection, triage, diagnosis, and remediation of network problems in production. According to the EMA report, 79% of respondents rate automating these tasks as a high or very high priority. Organizations are looking for AI-driven, agentic automation—tools capable of reasoning about network conditions and taking autonomous or semi-autonomous action. The report found that 55% of respondents say AI features are a requirement when evaluating new tools, and AI-driven insights and automation is the top reason they would replace an incumbent. The day-two tasks organizations most want to automate include:

• Security response and containment: 54.3%
• Capacity and performance optimization: 49.7%
• Incident remediation and self-healing: 44.3%
• Configuration optimization: 40.3%
• Event correlation/alert noise reduction: 37.5%
• Change validation and rollback: 26.4%

EMA found that an emerging enabler is Model Context Protocol (MCP) support, which gives AI agents a standard interface to interact with multiple network management tools. Successful NetOps organizations were more likely to prioritize MCP support for agentic AI access to tools. “The MCP access points become like an abstraction layer across your tool sprawl,” McGillicuddy noted. This abstraction is critical because tool integration has long been a pain point; without standard interfaces, automation scripts become brittle and hard to maintain. By adopting MCP, teams can create a unified automation layer that works across disparate monitoring, logging, and ticketing systems. For example, an AI agent could automatically correlate a performance anomaly detected by one tool with network configuration data from another, then execute a remediation action—all without human intervention. The push toward day-two automation represents a fundamental shift in how network operations will function in the age of AI.

Megatrend 3: Hybrid and multi-cloud networks remain ungoverned

Nearly seven in ten (69%) surveyed organizations operate hybrid cloud environments, and 66% are multi-cloud. Yet only 36% say they are completely effective at managing their cloud networks. This gap reflects both technical complexity and cultural friction between network teams and cloud engineering groups. EMA found the core challenges are familiar: proprietary networking constructs that vary across providers, inconsistent telemetry, skills gaps on the network team, and limited end-to-end visibility across cloud and on-premises environments. “I still talk to network observability vendors that haven’t got feature parity across the big three cloud providers yet,” McGillicuddy said. “They might be good at collecting and analyzing data from AWS, but they’re still kind of far behind on things with Google Cloud Platform, and they haven’t even thought about some of the secondary ones yet.”

Organizations that have managed to integrate IP address management and extend network observability tools across hybrid environments report better overall outcomes, but both remain works in progress for most. The lack of unified governance means that network policies are often inconsistently applied, leading to security gaps and performance issues. For instance, a misconfiguration in a cloud virtual network could expose critical data, but without visibility into that cloud environment, the network team might not discover the issue until after a breach. EMA’s data suggests that successful teams prioritize integration over consolidation—they focus on security insights, workflow integration, and data sharing across their toolset rather than trying to reduce its size. They also build unified visibility and security controls that span both on-premises and cloud infrastructure. As more workloads move to the cloud, the ability to manage hybrid networks effectively will become a competitive differentiator.

Megatrend 4: AI networks need managing, and few tools are ready

Nearly half of respondents (47.7%) said AI training or inference workloads are already deployed on their networks. Most of the rest expect to deploy within the next two years. But only 35% say their current network observability tools are completely ready to manage those workloads. The performance concerns are specific to AI infrastructure: isolating problems across networks, applications, and GPU clusters simultaneously; managing inference tail latency; and gaining visibility into GPU utilization as a network signal. The tool enhancements teams most want to close the gap include:

• AI-powered troubleshooting and remediation: 51.3%
• Proactive alerting for AI-related performance risks: 49.3%
• AI workload awareness via real-time packet analysis: 46.9%
• Real-time streaming telemetry to replace polling intervals: 40.2%
• Correlation of GPU, application, and network performance metrics: 34.3%

AI workloads impose unique demands on the network. Training runs generate massive east-west traffic between GPU nodes, requiring ultra-low latency and high bandwidth. Inference workloads, on the other hand, are sensitive to jitter and tail latency; even a small delay can degrade user experience in real-time applications. Traditional network monitoring tools, designed for periodic polling and coarse-grained metrics, cannot provide the granular, real-time visibility needed. Tools that rely on polling intervals may miss microbursts or transient congestion that cause AI jobs to slow down. Streaming telemetry, which pushes data continuously rather than waiting for poll requests, is essential for catching these events. Moreover, correlating GPU utilization with network performance is a new challenge—most monitoring tools today treat GPU metrics as a separate domain, but in AI workloads, the two are deeply intertwined. A network slowdown can cause GPUs to idle, wasting expensive compute resources. EMA’s research underscores the urgency: organizations that delay upgrading their observability tools risk derailing their AI transformation projects.

What successful teams are doing differently

EMA’s research also identified the practices separating successful organizations from those falling short. Successful teams hold network observability data to a strict accuracy standard. They have moved beyond scripts and runbooks to AI-driven and agentic management tools, and they prioritize integration over consolidation. They focus on security insights, workflow integration, and data sharing across their toolset rather than trying to reduce its size. And the successful organizations are building unified visibility and security controls that span both on-premises and cloud infrastructure. “AI networking, or networks for AI, is going to require some retooling. I recommend you talk to your vendors about whether they’re thinking about this. Most of them aren’t, probably because they’re not hearing from you,” McGillicuddy advised. The message is clear: network teams must be proactive in demanding better tools and support from their IT leadership and vendors. Without that advocacy, the gap between operational reality and business expectations will only widen.

Source: Network World News