Nokia Forecasts AI Will Drive 44% of Global Network Traffic by 2035

The Finnish telecoms equipment maker's Global Network Traffic Report projects that total traffic will nearly quadruple from 2026 levels to almost 3,789 exabytes a month over the next decade. For operators and data centre builders, the more consequential prediction is about direction and location rather than volume. Nokia expects machines to send much more data up into networks than people do, and more AI processing to move out of central cloud facilities and closer to users.

The fastest growth comes from two categories that are small today. Nokia projects that traffic from agentic AI, software that carries out tasks on a user's behalf, will grow about 160-fold between 2026 and 2035, from a small base. Traffic from what it calls physical AI, meaning robots, vehicles and other systems that interact with their surroundings, is forecast to rise about 105-fold, and more than 90% of it is expected to travel over the uplink.

That would reverse the pattern most networks were designed around, in which consumers mainly download video and web pages. In the cells with the heaviest AI use, uplink could account for 24% of total access traffic, the report found. That figure applies to the most AI-intensive areas, not the global average.

Nokia also expects a shift in where AI work is done. The central cloud will still handle most AI traffic, but edge and metro core facilities are forecast to serve 36% of AI-native traffic by 2035. The company attributed this to demand for lower latency, more efficient models and data sovereignty rules, which push inference nearer to users and devices in an arrangement it calls the "AI grid".

Separately, the report predicts that fixed wireless access, in which homes get broadband over mobile networks, will generate more than a third of consumer cellular traffic by 2035.

Pallavi Mahajan, Nokia's chief technology and AI officer, said: "AI is changing what we need from networks, not just how much traffic they carry." She said network readiness should be a central part of AI planning, adding: "The next phase of AI will depend as much on the network as on compute."

Nokia said the forecast comes from a bottom-up model of consumer, enterprise and AI traffic that combines application usage, connected-device activity and agentic workflows with third-party research, expert review and testing in its own labs. The figures are modelled projections, not measured traffic. Nokia also sells the routing, optical, mobile and data centre networking equipment that this kind of growth would require.

The forecast is well above other industry estimates. Cisco has projected that AI inference will make up about a quarter of total network traffic by 2035. An earlier edition of Nokia's traffic report drew criticism from telecoms analyst Dean Bubley, who called its estimates of indirect AI traffic questionable. How much of the forecast uplink surge turns up in operators' capacity budgets will depend on how quickly robots, connected vehicles and AI agents are deployed at scale.