Network planning has traditionally focused on a familiar problem: ensuring that people can quickly access digital content. Mobile operators have invested heavily in downlink capacity because video streaming, social media and cloud applications have generated enormous demand for data moving toward users.

Artificial intelligence could complicate that assumption. As AI systems become multimodal and increasingly connected to cameras, microphones, vehicles, industrial equipment and robots, devices may become major sources of network data. Instead of simply requesting information, they will continuously send observations to AI models for interpretation.

The change is not yet visible at full scale. John Saw, T-Mobile’s president of technology, said at Mobile Future Forward that the company has not seen much impact from AI on network usage so far.

Current effects are concentrated largely in transport networks and around major data centers. Saw expects that to change as AI workloads move closer to users and devices.

His comments reinforce the importance of 5G-Advanced research from the 3rd Generation Partnership Project, particularly improvements involving uplink performance, latency and network flexibility.

Physical AI Causes Issues with Network Planning

Ericsson’s 2026 Mobility Report examined traffic growth at 55 service providers during 2025. Uplink traffic grew faster than downlink traffic at 43 of those providers. At 17 operators, uplink growth exceeded downlink growth by more than 1.5 times.

Some of that increase is attributable to current applications, including communications, collaboration tools, user-generated content and cloud storage. These services provide an early indication of a broader shift. AI could add a more demanding category of upstream traffic by requiring devices to transmit live images, video, audio and environmental context.

Ericsson has estimated that cloud-based AI and augmented-reality smart glasses could require between 1 Mbps and 10 Mbps of uplink throughput system delivery.  In some cases, upstream traffic could be eight times greater than downstream traffic. 

The new model is very different from conventional mobile video, which remains overwhelmingly download-oriented.

Ericsson’s medium AI adoption scenario projects that uplink traffic could reach three times 2025 levels by 2031. Under a high-adoption scenario, the increase could reach fivefold. Total mobile data traffic is expected to grow by approximately 2.2 times over the same period.

Operators need to consider traffic direction, as well as, total volume.

Speed and Security are Important Parts of Network Planning

Physical AI may place the greatest pressure on network architecture that the internet has ever experienced.

. Robots and autonomous machines need to process large amounts of information while operating under strict power and computing constraints. Sending every task to a distant cloud could create unacceptable delays, particularly for industrial applications, transportation systems and safety-sensitive machines.

Edge computing can reduce that delay by placing processing resources closer to the device. The Ericsson discussion of edge computing and distributed intelligence outlines how computing can move closer to where data is created.

AT&T Business Product executive Shawn Hakl has described an environment in which users may interact with six to 12 AI models distributed across devices, enterprise locations, edge facilities and multiple cloud platforms. This creates a more complicated communications pattern. 

Data may move between sensors, local systems, network nodes, models and data repositories before a result returns to the user or machine.

The result is a network that must support more than speed. It must provide predictable latency, resilience, security and the ability to route workloads dynamically. Operators may also need stronger coordination between wireless access networks, transport infrastructure, cloud providers and enterprise systems.

Operators Must Invest Carefully

The case for preparation should be balanced against current usage. The GSMA’s mobile industry research indicates that direct generative AI represents only a small portion of mobile traffic today. Ericsson also found that generative AI accounted for just 0.06% of total network traffic in one mature mobile market measurement.

Those figures suggest that operators should avoid assuming that every AI forecast will become reality. Overbuilding for applications that remain experimental could place unnecessary pressure on capital budgets. At the same time, underestimating the transition could leave networks poorly prepared when AI-enabled devices gain broader adoption.

Operators can expand uplink capacity where demand is emerging, and test edge-computing deployments with enterprise customers.  5G-Advanced capabilities will improve network flexibility.

They can also study traffic patterns generated by smart glasses, industrial sensors and connected vehicles before those categories reach mass scale.

Network planning for the AI era will require more than additional towers or spectrum. It will require a broader understanding of where intelligence resides, where data is created and how quickly machines must respond.

The next generation of networks may be judged not only by how effectively they deliver content to people, but also by how reliably they carry the physical world to machines.

For additional telecommunications and technology analysis, readers can visit the USA Herald author page and search results for Jackie Allen. Related USA Herald technology coverage includes: AI, legal issues and digital infrastructure.