Nokia estimates that wide area network (WAN) traffic will grow between 3 and 7 times by 2034, reaching 3,096 exabytes (EB) a month in its moderate scenario, of which about 921 EB a month would come from artificial intelligence, nearly 30% of the total. These are projections from the Global Network Traffic Report, the study the Finnish company produces with its Bell Labs Consulting division, covering the 2024-2034 period.
Key takeaways from Nokia’s traffic report, in 30 seconds
- Nokia and Bell Labs Consulting expect global WAN traffic to grow 3 to 7 times from 2024 to 2034.
- In the moderate scenario it would reach 3,096 EB/month in 2034; the conservative scenario stays at 2,277 and the aggressive one climbs to 4,878.
- AI would contribute about 921 EB/month, nearly 30% of the total, growing 23% annually.
- AI agents and inter-data-center traffic are the fastest-growing segments.
- The report warns of more upstream traffic and tighter latency requirements.
The document measures traffic that moves through operator and enterprise WAN infrastructure: the public internet, managed and private WAN services, and data movement between data centers. It leaves out traffic that never leaves a campus local network. To build its figures, Nokia combines data reported by operators, public and national statistics, information from GSMA Intelligence and GlobalData, and its own measurements, validating the projections against observations from customer networks. The report carries no explicit publication date; the version available on Nokia’s website is dated 2025, and the explainer page that accompanies it was last updated on March 2, 2026.
The company lays out three scenarios rather than a single figure. In the conservative one, global WAN traffic would reach 2,277 EB a month in 2034, with a compound annual growth rate (CAGR) of 13%. The moderate scenario, which Nokia uses as its reference, points to 3,096 EB/month and 16% annual growth. The aggressive scenario raises the figure to 4,878 EB/month, with 22% annual growth. That range is where the “3 to 7 times” figure the report repeats comes from.
Who generates the traffic in 2034, according to Nokia
Households will still account for the bulk of it. In the moderate scenario, consumer fixed broadband (excluding fixed wireless access, FWA) would contribute 1,806 EB/month in 2034, around 58% of the total. Consumer mobile broadband would add 499 EB/month and FWA another 261 EB/month, while enterprise and industrial traffic would reach 531 EB/month, around 17%. Video keeps its weight: according to the report, it accounts for almost three-quarters of mobile data.
| WAN traffic in 2034 (EB/month) | Conservative | Moderate | Aggressive |
|---|---|---|---|
| Global total | 2,277 | 3,096 | 4,878 |
| Compound annual growth 2024-2034 | 13% | 16% | 22% |
| Consumer, fixed broadband (excl. FWA) | 1,405 | 1,806 | 2,791 |
| Consumer, mobile broadband | 364 | 499 | 633 |
| Enterprise and industry | 333 | 531 | 913 |
AI doesn’t appear as a separate segment, but rather as a layer running through all of them. Nokia estimates that AI-related traffic will grow at an annual rate of 23%, compared with 15% for the rest, and will reach 921 EB/month by 2034. Most of it will come from consumers, directly or indirectly, through assistants, generative applications, and content recommendations. Enterprise and industrial AI is much smaller in volume, around 101 EB/month, but it’s the fastest-growing segment in the moderate scenario, at 48% annually. That includes factories with robotics, sensors, and machine vision, which the study says require reliable connectivity with tighter latency.
The report devotes a section to AI agents, which turn task execution into an autonomous loop and multiply queries, model calls, and machine-to-machine interactions. Nokia expects machine-to-machine (M2M) traffic generated by these agents to go from 66 EB/month in 2025 to 537 EB/month in 2034, at a 26% CAGR. On its explainer page, the company adds that 37% of AI traffic will be machine-generated by 2034.
More upstream traffic, tighter latency, and heavy inter-data-center traffic
Nokia also anticipates a change in the shape of traffic. Downstream traffic dominates today, but generative AI invites people to create and send content, and the report expects household traffic to become more symmetric over time. Some consumer AI, augmented reality, or home management applications could reach 40% upstream traffic, though the document itself notes that their overall impact will stay modest given their low bit rates per user. In the enterprise, the trend runs the other way: enterprise AI, which today is mostly upstream, would shift toward downstream traffic.
The part least visible to the user is what happens between data centers. According to the study, AI inference traffic flowing through links between data centers will grow 20.3% annually through 2034, and the 921 EB/month of user-originated inference would generate about 3,260 EB/month of WAN traffic between data centers, 3.5 times more. Two-thirds of that volume (67.4%) would travel over operators’ transport networks. Nokia concludes that networks will need to prepare for more symmetric traffic, stricter latency, and massive interconnection between clouds and data centers. Back in May, Revista Cloud already looked at how AI traffic is changing the rules of cloud networks.
These figures are worth reading with caution. They’re projections from a network equipment maker with a direct interest in operators expanding capacity. Analyst Dean Bubley, in a review published on February 16, 2026, criticizes the report for explaining too little of its assumptions, considers its AI-driven incremental traffic estimates optimistic, and says it’s missing a downside scenario and a clear comparison with the previous edition. It’s also not Nokia’s first range: the 2023 edition put global demand between 2,443 and 3,109 EB/month as early as 2030, above what the current edition projects for 2034 in its conservative and moderate scenarios.
More traffic doesn’t mean the same number of data centers
For the data center sector, the report points to more facilities, but above all to different facilities: more specialized, with higher power density per rack, more efficient, and, where latency matters, closer to the user. The document itself describes AI traffic as something that doesn’t stay inside one data center, but moves between the network edge, metro networks, the core, and the cloud, with bursts of inference and training at the metro edge.
That growth doesn’t have to translate into a proportional number of new buildings. Chips deliver more performance per watt, models get compressed, and architectures distribute load better, so the same data center can handle considerably more traffic than it could just a few years ago. What is increasing is the pressure on physical infrastructure: available power, cooling for ever-denser loads, land with grid access, fiber connectivity between facilities, and the capacity to operate them. The International Energy Agency (IEA) estimated in its Energy and AI report, dated April 10, 2025, that data centers consumed about 415 TWh in 2024, close to 1.5% of global electricity, and that figure could reach around 945 TWh by 2030. Brussels is already working in that direction, and will give data centers an energy rating.
Nokia’s report doesn’t quantify how many data centers will be needed or how much energy they’ll consume, and its traffic figures don’t allow for a direct estimate either. What it does suggest is that the connection between clouds and data centers, with high capacity and low latency, is becoming a persistent demand, and that’s the point where the plans of operators, cloud providers, and infrastructure developers intersect.
Frequently asked questions
How much will global network traffic grow by 2034, according to Nokia?
Nokia expects WAN traffic to grow 3 to 7 times. By 2034 it would range from 2,277 EB/month in the conservative scenario to 4,878 EB/month in the aggressive one, with 3,096 EB/month in the moderate scenario.
How much of that traffic will artificial intelligence generate?
In the moderate scenario, AI would contribute about 921 EB/month in 2034, nearly 30% of global WAN traffic. It would grow 23% annually, compared with 15% for the rest of traffic.
What is an exabyte?
An exabyte (EB) equals one billion gigabytes. The report expresses all its projections in exabytes per month.
Will proportionally more data centers need to be built?
The report doesn’t quantify that. Better chips, models, and architectures let the same infrastructure handle more traffic, but pressure on energy, cooling, land, and connectivity between data centers keeps rising.
Sources:
- Nokia and Bell Labs Consulting, Global network traffic report (2025): https://www.nokia.com/asset/213660/
- Nokia, Network traffic is fundamentally changing in the AI supercycle (updated March 2, 2026): https://www.nokia.com/artificial-intelligence/explainer-network-traffic-is-fundamentally-changing-in-the-ai-supercycle/
- International Energy Agency, Energy and AI (April 10, 2025): https://www.iea.org/reports/energy-and-ai/executive-summary
- Dean Bubley, Review: Nokia Global Network Traffic Report (February 16, 2026): https://deanbubley.substack.com/p/review-nokia-global-network-traffic
- TVBEurope, on Nokia’s 2023 report edition (October 31, 2023): https://www.tvbeurope.com/?p=114773
This article is also available in Spanish: «Nokia prevé que la IA genere cerca del 30 % del tráfico de red mundial en 2034», on revistacloud.com.
