Unexplained network performance issues are alarming IT professionals, particularly during critical work periods such as video conferences and project finalizations. A growing concern is the impact of AI features in SaaS tools on network bandwidth, which has not been adequately monitored. Over the past year, many major SaaS products, including Microsoft 365 and Salesforce, introduced AI functionalities, leading to increased network traffic that is not accounted for as “AI traffic” in bandwidth assessments.
Half of U.S. employees now utilize AI at work, a significant rise from 21% in 2023, with 28% reporting use a few times weekly. Much of this activity is integrated into existing business applications, generating AI-related traffic that often goes unnoticed in budget plans. Analytics from Microsoft indicate that Microsoft 365 Copilot requires specific WebSocket connections to maintain functionality, which many IT teams may not have anticipated when these features were rolled out.
Cisco’s Nik Kale highlighted that retrieval augmented generation (RAG) architectures generate substantial data traffic as it traverses various network regions and object stores. He noted that machines create 100 times more requests than human users, raising concerns about network capacity and traffic management. These volumes, combined with the hidden nature of AI traffic, present a challenge for IT teams aiming to manage bandwidth effectively.
A Broadcom report identified that 95% of network teams lack visibility into major segments of their networks, resulting in only 49% believing their systems can support AI’s bandwidth needs. This gap in oversight affirms that many organizations remain unprepared for the growing demands placed on their networks by AI and cloud migrations. IDC research predicts a 49% rise in cloud connectivity and a 51% increase in edge bandwidth demand over the next year.
To address these issues, IT professionals are encouraged to utilize tools like ntopng and PRTG for enhanced traffic analysis. They can look for increased HTTPS traffic to specific SaaS endpoints, new WebSocket connections, and API call patterns that rise outside regular business hours, which may signal hidden AI usage. Establishing a fresh bandwidth baseline is crucial to understanding current network utilization, as previous measurements may no longer reflect actual traffic due to the introduction of AI features.
Furthermore, an IDC survey revealed that 44% of enterprises plan to boost connectivity budgets by over 10% by 2026, indicating that reevaluations of current network capacity and requirements are necessary. Quality of service (QoS) policies may also require adjustments to ensure that AI-related traffic does not conflict with essential voice and video services. IT teams should proactively check network requirements for newly introduced AI features to mitigate performance issues and enhance network reliability moving forward.
Understanding AI traffic patterns aids in improving network diagnostics and planning. Knowledge of hidden AI use represents a critical step in managing and optimizing bandwidth amid evolving technological demands.
