Explainer: Network traffic is fundamentally changing in the AI supercycle
Global network traffic is surging, driven by changes in consumer behaviors, enterprise processes and AI. But while it’s tempting to simply think about an explosion of data, our Global network traffic report shows that, more critically, AI is changing the nature of traffic and requires connectivity to advance in parallel.
In 1519 Ferdinand Magellan set sail from Spain to find the Spice Islands. This took nearly three years on a wooden sailing ship and saw the first circumnavigation of the globe. Over the next few hundred years travel advances came thick and fast. And these advances didn’t just increase the volume of global traffic but changed its entire nature.
Today a politician can jump on a plane, attend an emergency gathering of international leaders and change the world forever. A blue fin tuna caught in Japan’s seas can be on sushi platter in Los Angeles in 24 hours, but lemons still often travel thousands of miles by cargo ship. The possibilities are endless, and they’re calibrated for specific results. Now the arrival of the AI is having a similarly transformative impact on global network traffic.
“Global network traffic is changing in character, not just in volume,” explains Nokia’s Global network traffic report, produced with Bell Labs Consulting.
This is due to fundamental shifts in the way people and businesses use connectivity. Some shifts are directly driven by AI, some indirectly, and some are not dependent on AI, but together they’re reshaping how networks must be planned and operated.
Moderate estimates suggest that about 70% of global WAN traffic will still come from non-AI sources by 2034. Yet AI traffic will represent the biggest area of growth (23% vs. 15% CAGR) during this time.
The important takeaway, however, is the transformative way traffic is moving across the globe. At present downlink dominates due to content consumption at scale, but we’re moving past the age where consumers simply stream TV in real-time. Information is no longer just for human consumption. It is gradually providing more context for artificial intelligence to collate, recommend and make decisions.
For network operators, the key will be to stay aware of these changes and respond to what they mean, in terms of both challenges and opportunities. Automation will clearly be a part of the solution: AI-driven operations will help networks adapt to shifting traffic patterns, while advancing connectivity in turn will enable the next wave of AI-powered experiences and processes.
How three trends are driving global traffic through 2034
At present, only some network traffic is related to AI, but this is changing. And three key trends are gradually transforming our interactions with the digital world.
1. Immersive experiences drive two-way interaction with the network
Expectation for immersive experiences began with streaming videos. But they’ve evolved to include more interactive engagement, such as multiplayer VR gaming, sophisticated 3D manufacturing modelling and collaboration. These experiences are shifting behavior towards more active participation.
Impact on traffic: For operators, immersive and interactive services make networks more sensitive to delay as real-time interaction becomes central to the experience. This means success depends on delivering consistently smooth performance through predicable latency and jitter, not just adding more bandwidth.
2. Enterprise and industrial operations move to the edge
The use of real-time data in industrial and enterprise settings has been escalating for years. This stems from digital twins modelling things like cars and biological processes, such as the cardiovascular system. AI also drives the physical robots necessary to make all those new cars and personalized drugs modelled by this data.
And the sheer size of some of these tasks means that some elements must be handled by servers on-site – on the edge – while others need to be pushed out to data centers further afield. The use of AI is also increasingly necessary in the management of all this.
Impact on traffic: For communications networks, this means that traffic becomes more distributed and variable, with sudden bursts of activity. This requires careful management of the traffic from edge-to-cloud.
3. AI inference generates traffic autonomously
AI is beginning to directly generate its own traffic. This comes from a wide range of consumer sources like AI chatbots, AI-assisted cloud applications – such as fitness or shopping services – as well as AI-powered XR experiences. It also comes directly from enterprises via content creation, predictive maintenance, autonomous operations and analytics.
Indirect traffic, like video streaming recommendations and online marketplaces, is also rising in parallel. And all these sources are coming together to fundamentally change the nature of traffic.
Impact on traffic: Our report suggests that the growing volume of AI-driven autonomous activity will mean 37% of total network AI traffic will be machine-generated by 2034. When this is combined with other traffic sources, it is expected to drive a threefold increase in inter-datacenter traffic and will significantly elevate the importance of resilient, efficient, and secure connectivity.
ChatGPT, agents and robots will hit networks by 2034
Global predictions can only go so far. So, to augment its global findings, the research team at Bell Labs Consulting applied its traffic predictions to a reference network of a telecom provider in an advanced market. This model takes a local example and simulates how different types of AI traffic will change through to 2034. The impact of these three different types of AI also roughly maps to wider trends.
Generative AI means more uplinks
Generative AI is uplink-heavy by design, with workloads driven by multi-model, media-rich content submissions for inference. AI-enabled IoT will further amplify this uplink of traffic through the transmission of video streams and sensor data.
The news still emphasizes Generative AI, with engines like ChatGPT creating new materials – like written content and video. And it is taking on a life of its own through recommendations, personalization and real-time experiences.
Yet Generative AI is only the tip of the iceberg, and over time, it will work with Agentic AI and Physical AI to drive a broader AI impact. For networks operators, Generative AI means a surge in demand for uplinks and requires a rethink of network capacity to match the changing shape of traffic.
Agentic AI means more ‘bursty’ data
With Agentic AI, machine interactions will substantially increase backbone traffic. This machine traffic is often ‘bursty’ in nature due to periods of high activity followed by periods of low activity.
Agentic AI is a system that can plan and act to achieve predetermined goals. This type of artificial intelligence may drive customer service or virtual assistants, while Generative AI can act as a natural “brain” behind the scenes.
Over time, Generative and Agentic AI capabilities are expected to converge further. But for network operators, this evolution highlights the need to manage ‘bursty’ irregular data patterns.
Physical AI faces high latency
Physical AI systems, which sense and act in the real-world, are expected to be predominantly uplink-driven, with most traffic originating from sensors, cameras, and other upstream data sources. This will make latency paramount.
Physical AI, where artificial intelligence powers physical machines, is the third part of the AI puzzle. This has the potential to make latency for AI bigger than an organization’s latency budget and drives a need to run AI workloads at the edge. The upshot of this is that areas with large amounts of AI-driven robotics – like industrial sites – will need to juggle workloads onsite and offsite.
Physical AI will likely run in parallel with other types of AI. Automated delivery robots, for example, are already operating in some US cities and can be used to provide allergy advice on takeout orders delivered to your doorstep. These sidewalk carts can therefore utilize all three types of AI and offer a glimpse into our connected future.
AI on the network in 2034
- Generative AI will drive a surge in uplink demand creating a need for capacity to match
- Agentic AI will mean ‘bursty’ irregular data patterns necessitating pre-scheduling tasks, using network slicing
- Physical AI will make latency for AI bigger than latency budget, driving a need run AI workloads at the edge
Connectivity is advancing towards AI-native networks
The arrival of AI is like the arrival of the airplane. It is not just about an increase in global traffic volume. It is a fundamental rethinking of how traffic moves across the globe.
AI is an evolution that will build into a revolution, driving a supercycle of change like the internet before it. The full scope of AI will not happen straightaway, but gradually, and will require connectivity to advance in parallel. Communication networks are already optimized and powered by AI, but ultimately, they must become AI-native networks.
