Nokia predicts AI will make up almost half of all network traffic by 2035.

In its Global Network Traffic Report 2026, the Finnish giant forecast that global network traffic will grow 28% annually, nearly quadrupling to 3,789 exabytes (EB) per month halfway through the next decade. AI-native traffic will account for 44% of the total, representing approximately 1,681 EB per month.

Traffic generated directly by AI applications and workflows is expected to increase 60-fold from this year’s levels, growing at a 58% compound annual growth rate (CAGR) to reach 987 EB per month, representing 26% of global traffic. Another 694 EB per month, or 18%, will come from AI-influenced traffic, such as increased video consumption or online shopping. Remaining traffic will be of the traditional, non-AI sort.

Enterprise access AI traffic is forecast to grow at a 59% CAGR, faster than consumer AI and overtaking consumer AI later in the forecast period. Industrial IoT applications will generate approximately 70% of enterprise AI traffic by 2035, driven by devices like cameras, autonomous vehicles, drones, and sensors.

This subset also includes humanoids, with Nokia detailing physical AI traffic generated by robots, robotaxis, and the like will increase 105-fold at a 68% CAGR to approximately 67 EB per month by 2035.

Agentic AI will become one of the fastest-growing traffic categories, increasing approximately 160-fold to 585 EB per month by 2035, representing a 76% CAGR.

Nokia also expects around 1,134 EB per month of AI traffic will enter access networks. Edge and metro core will service 36% of all AI-native traffic, with the former carrying 9% of AI-native traffic, and the latter handling 27% of AI-centric traffic. The rest will be serviced by regional core or central cloud.

The vendor did warn that the uplink-heavy nature of AI will pose a considerable challenge to capacity.

“As uplink demand rises it will place increasing stress on the cellular network,” Nokia noted. “Uplink share describes the traffic mix on a cell, not the pressure placed on the network. What matters is how much of a cell’s uplink capacity this traffic uses during busy hours. Above 70% utilization, we see latency and jitter start to rise, impacting many applications including AI.”