Cisco: AI Traffic Is Radically Reshaping WANs

Why It Matters

AI‑driven traffic will become a core network load, reshaping capacity, latency, and security strategies for service providers and enterprises alike. Ignoring its distinct characteristics risks performance degradation and lost competitive edge.

Key Takeaways

• AI inference traffic could reach 25% of WAN volume by 2035
• AI flows last twice as long and flow ten times slower
• Upstream traffic share rises to 9% for AI flows vs 0.5% web
• QoS and security policies must adapt to longer, asymmetric AI flows
• Service providers need larger flow tables for growing AI inference sessions

Pulse Analysis

Cisco’s new "AI Impact on Wide Area Networks 2026" report quantifies a seismic shift in WAN traffic patterns as autonomous AI agents proliferate. While baseline enterprise traffic is expected to grow 2.5‑fold over the next decade, the addition of agentic AI could push total traffic to nine times current levels. By 2035, AI inference alone may account for roughly 25% of all WAN bytes, a dramatic rise from today’s negligible share. The study’s blend of provider‑level measurements, third‑party data, and controlled lab experiments reveals AI‑generated flows that are up to 450% larger than human‑originated traffic, underscoring the urgency for network planners to reassess capacity forecasts.

Beyond sheer volume, AI traffic exhibits distinct technical traits that challenge legacy network designs. Inference flows persist twice as long as typical web transactions yet transmit data at rates ten times lower, creating prolonged, low‑throughput streams that strain flow‑aware devices and expand state tables. Moreover, AI traffic is markedly more upstream‑heavy—about 9% of AI flows are upstream‑dominant versus just 0.5% for conventional HTTP—forcing a rethink of radio and back‑haul provisioning. These asymmetries, coupled with latency‑sensitive LLM inference that can span hundreds of milliseconds to several seconds, demand differentiated QoS policies, enhanced observability, and tighter path security to preserve end‑user experience.

The business ramifications are equally profound. Gartner predicts 40% of enterprise applications will embed task‑specific AI agents by 2026, and by 2035 AI‑enabled software could generate $450 billion in revenue, roughly 30% of the enterprise software market. Service providers that pre‑emptively upgrade routing architectures, expand flow‑table capacities, and embed AI‑aware traffic engineering will capture new revenue streams and avoid costly performance bottlenecks. Enterprises, meanwhile, must treat network design as a strategic component of AI deployment, aligning capacity planning with AI workload forecasts to ensure that the “nervous system” of connectivity can keep pace with machine‑speed decision making.