Nokia projects that artificial intelligence will account for 44% of global network traffic by 2035, up from a small share today, driven mainly by agentic AI and physical AI rather than today’s more familiar generative AI chatbots and assistants. The forecast, part of the company’s latest network traffic outlook, points to a shift in where that traffic flows: increasingly upstream, from devices and sensors toward data centers, rather than downstream to end users.
Nokia’s AI traffic forecast in 30 seconds
- Nokia projects AI will generate 44% of global network traffic by 2035.
- Agentic AI traffic is expected to grow 160-fold compared with 2026.
- Physical AI traffic is expected to grow 105-fold over the same period.
- More than 90% of physical AI traffic will flow upstream, from sensors and devices to data centers.
- Edge and metro networks will carry 36% of that AI traffic, rather than it all converging on core data centers.
The forecast lands days after Nokia’s own previous edition of its network-traffic report, which projected AI reaching nearly 30% of global wide-area network traffic by 2034. The new figures push that share higher and extend the horizon to 2035, reflecting a faster pace of AI adoption across both the consumer and industrial worlds than the company’s earlier model assumed.
Agentic AI: traffic generated by autonomous software
A large part of the projected growth comes from agentic AI: software agents that act autonomously on a user’s or organization’s behalf, chaining multiple steps and tool calls to complete a task rather than simply responding to a single prompt. Nokia expects traffic generated by this category of AI to grow 160-fold between 2026 and 2035.
Unlike a single chatbot query, an agentic workflow can involve many internal exchanges between models, tools, and data sources before it produces a final result for the user. Each of those internal steps adds to network traffic, even when the end user only sees one request and one response, which is why Nokia expects this category to scale so much faster than overall AI usage.
Physical AI: sensors and machines generating upstream traffic
The second major driver is what Nokia calls physical AI: artificial intelligence embedded in physical systems such as robots, autonomous vehicles, industrial sensors, and cameras, which perceive and act in the real world rather than just processing text or images on request. Nokia projects traffic from physical AI will grow 105-fold over the same 2026-2035 period.
Physical AI traffic behaves differently from the traffic patterns networks have historically been built around. Nokia says more than 90% of it will flow upstream, from devices and sensors toward data centers and processing systems, rather than downstream from servers toward end users as with video streaming or web browsing. That reverses the traffic balance that network capacity planning has traditionally assumed.
Why the traffic won’t all land in core data centers
Nokia’s forecast also addresses where this AI traffic will actually be processed. The company expects edge and metro networks, rather than large centralized core data centers alone, to carry 36% of AI-related traffic by 2035. Processing data closer to where it’s generated can reduce the latency and backhaul costs of sending every sensor reading or agentic request all the way to a distant data center.
That distribution has implications for how operators plan network investment: rather than concentrating capacity upgrades solely on links into core data centers, Nokia’s figures suggest operators will also need to size edge and metro infrastructure for sustained, asymmetric AI traffic growth.
An AI grid built for traffic, not just compute
Nokia frames this shift around the concept of an AI grid: network infrastructure purpose-built to move the kind of data AI systems generate and consume, in both directions, across core, metro, and edge locations, rather than infrastructure designed primarily around downstream content delivery. The company positions its own networking products as part of that emerging AI grid.
Fixed wireless access (FWA) is one of the areas Nokia highlights as a beneficiary of rising data demand, alongside the broader AI-driven traffic growth, as operators look to deliver more capacity to homes and businesses without always running new fiber.
Nokia says the projections are based on its own internal modeling of AI adoption curves, device and sensor deployment trends, and existing network traffic data, rather than a single external market study. As with any multi-year forecast, the actual pace of AI adoption, and therefore of traffic growth, could move faster or slower than the model assumes depending on how quickly agentic and physical AI systems are deployed at scale.
Pallavi Mahajan, who leads network infrastructure strategy at Nokia, said the shift toward agentic and physical AI represents a fundamentally different kind of network demand than the generative AI traffic operators have planned for so far, one that is more continuous, more upstream-heavy, and more distributed across the network rather than concentrated in a handful of hyperscale data centers.
Frequently asked questions
What share of global traffic will AI generate by 2035?
Nokia projects that AI will account for 44% of global network traffic by 2035, up sharply from its current share.
How much will AI agent traffic grow?
Nokia expects traffic generated by agentic AI, autonomous software agents that complete multi-step tasks, to grow 160-fold between 2026 and 2035.
Why does AI need more upstream traffic?
Physical AI systems such as sensors, cameras, and robots generate data that flows from devices toward data centers for processing. Nokia says more than 90% of physical AI traffic will move in this upstream direction.
What is the AI grid infrastructure?
It’s Nokia’s term for network infrastructure built specifically to carry AI-generated traffic across core, metro, and edge locations in both directions, rather than networks designed mainly for downstream content delivery.
This article is also available in Spanish: «Nokia», on revistacloud.com.
