AI, cloud and mobile set to drive significant growth in uplink traffic
The accelerated development and convergence of AI, cloud and mobile will fundamentally shift future traffic patterns, driving significant growth in uplink demands on mobile networks in the coming decade.
Key insights
As intelligent devices increasingly rely on cloud-based processing, data is flowing in the uplink more than ever before. Yet, this shift is more complex than a simple surge in uplink demand. While AI-driven systems like AVs and AR glasses continuously send data to the cloud, advances in on-device intelligence, compression and smart data transport are reshaping how and when that data moves. The result is a more dynamic balance, where networks must support both the growing appetite for real-time cloud and the efficiencies that keep bandwidth use sustainable. Understanding this interplay is key to preparing for the next wave of connected intelligence.
Figure 23: Future drivers of uplink traffic
Uplink requirements of current AI glasses
To date, approximately 2 million smart glasses from leading manufacturers have been sold in the US – amounting to approximately 1 percent market penetration – with ambitions to sell millions per year going forward. The success driving these sales is in connecting the user to an AI agent that delivers sentient engagements based on video and audio input from the glasses.
Going forward, some models will use AI capabilities right on the glasses and/or tethered devices; however, advanced AI capabilities will need to run in the cloud and – when inference time of the models is low – the uplink network characteristics become critical.
A recently announced smart glasses model has an advertised video capture resolution of 1,440 x 1,920 pixels. Multimodal AI on-demand engagements typically require framerates in the order of 5–10 frames per second (FPS) while being used. Always-on agents, on the other hand, are likely to use lower and perhaps dynamic framerates, such as 1 frame every 5–10 seconds. The on-demand agent can make use of video codecs with a compression ratio of about 0.1 bits per pixel (bpp). At the given resolution and a frame rate of 5 FPS, this yields about 1.4 Mbps in the uplink. It is further assumed that for users of these AI agents, about 20 percent are “power users” at 100 min/day, with the remaining 80 percent being “ordinary users” at 10 min/day. This yields an average of 28 min/day.
The always-on agent will need to use image compression, at about 0.5 bpp; at the given resolution and a frame rate of about 0.1 FPS, this yields about 0.14 Mbps. It is then assumed that the agent is on for about 8h/day.
The resulting increase in the uplink percentage with regards to today’s global average baseline of about 2 GB per month is shown in Figure 24. Per user, the always-on agent consumes a slightly higher uplink than the on-demand agent under these assumptions. For a given device penetration, given the value on the x-axis in Figure 24, some users may adopt an always-on agent whilst others prefer on-demand. The future demand will therefore be between these two curves. This potential growth of uplink traffic underlines the importance of network capacity planning, spectrum allocation and RAN feature developments.
Figure 24: Uplink traffic increase with regards to today’s uplink baseline versus AI/AR glasses penetration
Future demand will be somewhere between the lines for on-demand AI-agents and always-on AI-agents
