Cisco says AI is on track to become a dominant driver of Internet traffic growth over the next decade, even though consumer AI traffic remains a relatively small share of total traffic today.

In its first report focused on AI's impact on wide area networks, the company projects that AI and agentic AI will increase consumer-driven network traffic by about 6.6 times by the mid-2030s as consumer adoption approaches what it called near-universal usage.

Cisco said that increase would represent about 63% more growth than in non-AI scenarios.

WAN traffic patterns

The study was designed to examine AI's effect on the WAN rather than on GPUs or data center fabrics, and to inform network design and capacity planning.

Cisco said it combined real-world traffic analysis using Cisco Crosswork Assurance User Experience, third-party industry data and Cisco-controlled lab tests of AI agents.

Cisco described current AI inference traffic as negligible compared with categories such as video streaming, but said it expects AI inference traffic to account for 25% of total network traffic by 2035.

It also said AI traffic should become a meaningful component of overall network traffic by that point.

According to the study, AI traffic behaves differently from conventional web traffic.

AI inference flows are about two times longer than typical web transactions, require more upstream capacity and operate at software speed rather than human speed.

Infrastructure and operator response

Cisco said AI agents that carry out a wide range of tasks are emerging as power users, with as much as 450% more total traffic generated per task when the work is performed by an agent.

It said that shift will push operators to deploy flow-aware network and security systems as AI traffic increases.

The company also found that about 9% of AI inference flows carry more upstream than downstream traffic, versus about 0.5% for typical web traffic, and said that gap is likely to widen as agentic AI usage grows.

It also expects network latency to become more important over time.

While fiber networks already offer relatively symmetrical data rates and low latency, Cisco said cable operators are also addressing both.

It pointed to DOCSIS mid-split and high-split upgrades to dedicate more spectrum to upstream traffic, and said Tier 1 cable operators including Charter Communications have started introducing low-latency capabilities into their DOCSIS networks.

Cisco's Javier Antich, principal product management engineer in the CTO office of its provider connectivity group, and Gurudatt Shenoy, SVP of product management for provider connectivity, said AI traffic is already reshaping infrastructure needs by changing not just traffic volumes but the shape of traffic.

Separately, Comcast Chief Network Officer Elad Nafshi said at the Cable Next-Gen event in March that 97.1% of AI traffic on Comcast's network is still text-based, compared with 2.6% for images and 0.3% for video.