Generative artificial intelligence (GenAI) applications are placing new demands on enterprise wide area networks (WANs), especially at the edge. AI has the potential to disrupt enterprise network performance with its increasing volume and unique traffic patterns.
The rapid adoption of AI is transforming enterprise networks, shifting traffic patterns and increasing bandwidth demands in ways traditional WAN and edge network architectures were never designed to handle. As the volume and complexity of traffic outside the data center increases, enterprises are rethinking how to design and manage networks at the edge. Granular visibility combined with embedded intelligence can enable businesses of all sizes to dynamically manage these critical networks.
AI Challenges at the Edge
GenAI refers to advanced machine-learning models that understand text, audio and video context based on patterns learned from vast datasets. And it’s everywhere. As of March 2025, Chat GPT reports 600 million monthly users, MetaAI has 500 million and Google’s Gemini has 350 million.[1] GenAI applications tend to have more upload-intensive traffic, such as video, detailed images or code. Yet these data flows are unpredictable, with upload-intensive traffic followed by processing pauses and sudden responses.
Rather than a typical quick request via browser that returns a website or video, the opposite is true for many GenAI use cases. Sometimes large data uploads, like a video scan of an error code on an appliance, receive a short answer. Other times large uploads require even larger downloads, such as the repair manual for the appliance.
As a result, GenAI applications demand high-bandwidth, low-latency connectivity in both directions at unpredictable intervals. The result is a tremendous strain on the edge network and potential risk to the business. Depending on the request, an edge application or agent may need to communicate with multiple distributed agents to complete the task. Delays or bottlenecks in communication at any point along the way can risk the completion or accuracy of the response.
As GenAI adoption grows, networking challenges become more pronounced. The shift to upload-heavy workloads, the increasing complexity of GenAI data flows, and the sheer scale of data movement are already straining traditional WAN and SD-WAN architectures. For its critical and complex data processing, AI demands low latency, less jitter and strong security. AI traffic is unpredictable and creates a host of challenges to existing network infrastructure.
- Changing traffic patterns – While still a small portion of total network traffic, Delante reports that the volume of AI traffic has grown by 500% and that traffic behaves differently.[2] Compared to the network patterns of traditional applications, GenAI applications generate entirely different, unpredictable traffic behaviors. Some data is processed locally at the edge, while that data and more is sent to a data center for further processing by a large language model, training AI models or made available to other applications. AI traffic bursts happen randomly and, without warning, can disrupt the network.
- Latency – Slow or delayed connections can stop the processing of AI workloads, particularly distributed applications. That upsets users and customers. Latency is a measure of responsiveness which affects performance and quality of GenAI outcomes. AI workloads run multiple models in parallel and any delays in accessing data or other models can lead to inaccurate results, response delays and failures. Latency is annoying when watching a video but potentially damaging when conducting a financial transaction or responding to an emergency.
- Scalability – The large, unpredictable increase in traffic from GenAI can be devastating to existing WAN, edge networks and cloud on-ramps. While SD-WAN helps enterprises manage traffic and optimize bandwidth to reduce bottlenecks and improve performance, existing deployments were not designed to accommodate complex GenAI traffic. As WAN and edge networks expand, SD-WAN solutions must scale as well to establish an overlay of encrypted tunnels to effectively navigate the network.
- Security – Security is always a risk and the edge network is vulnerable. The 2025 Verizon Data Breach Investigations Report found, “The percentage of edge devices and VPNs as a target on our exploitation of vulnerabilities action was 22%, and it grew almost eight-fold from the 3% found in last year’s (2024) report.”[3] GenAI applications demand multiple peer-to-peer connections, and those connections must be secured. As the volume and complexity of these connections grow, SD-WAN solutions can establish an overlay of encrypted tunnels to ensure security from end to end. Edge security, such as firewalls, requires reinforced centralized cybersecurity solutions to keep up with the increasing volume and complexity of GenAI traffic.
Deploying GenAI applications requires existing enterprise networks and SD-WAN solutions to evolve in order to identify, understand and dynamically respond to AI traffic requirements.
Supporting AI Workloads and Optimizing Performance
AI traffic needs a network that can ensure high quality of service, even when performance is degraded. The VeloCloud SD-WAN platform and VeloCloud Robust AI Networking (VeloRAIN) architecture are designed to make dynamic, intelligent routing decisions that improve connection quality and enhance user experience. The platform works across any wireless or wireline network, including fixed wireless and satellite networks where high latency and packet loss are common.
VeloRAIN boosts the power of VeloCloud’s existing SD-WAN traffic steering and WAN optimization technologies with its own AI and machine learning technology to dynamically optimize the network for edge AI application demands.
Traffic Prioritization and Optimization
Effective traffic management starts with understanding each application running across the network, its business objectives and the service levels required to meet those objectives. The challenge when profiling AI traffic using traditional methods is encryption. The ability to identify and understand encrypted traffic is critical to prioritizing and optimizing AI traffic. VeloCloud SD-WAN can identify more than 4,300 business and consumer applications, AI and not, creating policies that determine how to prioritize each using the SD-WAN overlay.
By continuously monitoring network conditions across the end-to-end network including 5G fixed wireless and satellite, as well as existing MPLS and broadband networks, VeloCloud Dynamic Multipath Optimization (DMPO) understands the real-time performance of each network, to determine where to route both upload and download traffic for optimal performance.
With Dynamic Application-Based Slicing (DABS), enterprises maximize bandwidth utilization while maintaining the performance of critical workloads. DABS operates at the application layer, intelligently routing traffic across all available networks, regardless of whether they support network-layer slicing. Traffic is also prioritized based on identity and role, giving organizations more granularity when managing the network.
Expanded Visibility and Control
The AI capabilities in DMPO train a large language model on packet capture data and create a chat-like experience for users to get real-time information about the network and its performance. VeloRAIN uses AI trained on anonymized data from VeloCloud SD-WAN edge appliances in real-world deployments to troubleshoot incidents, remediate issues, adjust policies and otherwise automate network management.
Security is paramount and VeloCloud Secure Access Service Edge (SASE) powered by Symantec leverages best-in-class VeloCloud SD-WAN, VeloCloud SD-Access and Symantec Security Service Edge (SSE) to connect users located anywhere to applications at the edge, cloud and data center, in a reliable, secure and optimal manner. The solution delivers an improved user experience, reduces operational complexity and helps mitigate compliance risk.
VeloCloud SD-WAN with Dynamic Multipath Optimization, underpinned by the VeloRAIN architecture, enables AI-powered profiling, secure bandwidth optimization and automation. These intelligent networking technologies help organizations deploy AI today and know that networks will be prepared for the increased demand of tomorrow.
The ability to prioritize, optimize and secure workload connectivity for both existing applications and emerging GenAI applications will enable businesses of all sizes to better utilize existing network capacity, while optimizing performance of all business-critical applications.