This is according to the new study No time to wait: The accelerating impact of AI on campus and branch networks published by Cisco in collaboration with Foundry[i], which reveals how the rapid rise of large language models (LLM) and the growing wave of agentic AI are putting unprecedented pressure on enterprise campus and branch networks, while attack surfaces are also expanding beyond what defenses can handle.
One-third of the organizations surveyed globally already have large-scale, enterprise-wide AI agent implementations, and 85% anticipate moderate to significant expansion of their use in the next 24 months. These organizations also expect AI's impact on traffic to more than triple in the next three years, with a staggering 235% increase.
This is because, unlike human users, AI agents operate at machine speed, triggering dozens of API calls, database queries, and model inferences in a matter of seconds. They generate dense traffic in both directions (lateral communication between devices or servers, necessary for AI agents to exchange data) that traditional networks were never designed to handle.
“In the realm of generative AI, traffic is much more north-south. In the realm of agent-based AI, it will be much more east-west… Typically, networks are designed for constant traffic… suddenly, three agents try to communicate with each other and solve a problem. How can we support increased east-west traffic?”comments an AI strategy manager from a US technology company interviewed for the study.
These same agent-enabled AI workloads, which have the potential to transform businesses, are also particularly fragile. Advanced AI users note that AI workloads are extremely vulnerable to network issues: they are more sensitive to reliability and uptime (80%), bandwidth (75%), latency (71%), and packet loss (62%) than traditional applications.
In two years, network capacity will reach its limit.
Less than a third of mature AI users say their networks are fully prepared for projected AI growth. Overall, 76% of respondents admit they need upgrades, and 73% say they have reached, or will reach, campus and branch office capacity limits within the next 24 months. Crucially, Wi-Fi is becoming a major bottleneck for AI, with more than half citing it as the area driving the largest increase in capacity requirements.
There is also still a disconnect between forecasts and reality, as three-quarters of IT leaders agree that they have more confidence in their organization's AI strategy than in the network's ability to implement it. But while 91% cite budget constraints as a barrier, almost all companies plan to modernize their networks.
Attack surfaces are already expanding.
AI has also created a complex security environment, with the vast majority reporting difficulty keeping up (92%) and AI already causing some damage (90%). More than two-thirds also believe that AI-related threats are evolving faster than their ability to adapt, and that failing to upgrade networks in the next two years will only increase security risks. Meanwhile, the 'observability' gap is widening, as traditional monitoring tools struggle with intermittent, east-west agent flows.
The findings make it clear that network resilience, observational capabilities, and adaptive security are not complementary actions in the age of AI, but essential. The network has survived decades of transformation, from the dot-com bubble to the cloud, adapting and evolving to meet the needs of the moment. Organizations that view network modernization as a prerequisite for their AI strategy, rather than a parallel process, will be involved in defining the next generation of enterprise AI.
