Dean Bubley: Does AI Drive Demand for Mobile Spectrum?

We’re confronted by stories about the huge impact of AI. We know it will mean more GPU chips, more energy for powering data-centers, and more fiber to connect them. We hope it will create more productivity and innovation. It’ll need more regulatory oversight and guard-rails, as well.

But some in the mobile industry have gone further. They are asserting it means we need more spectrum for public cellular carriers. On a recent Broadband Breakfast webinar, a CTIA speaker suggested that a lack of suitable 5G spectrum might somehow hinder AI development and usage.

More traffic?

There are four ways AI could potentially increase traffic on wireless networks:

• More use of existing mobile applications: AI-driven experiences might increase data consumption, such as by creating more addictive, highly- personalized scrollable timelines or compelling video content for social media on smartphones.
• New applications based around AI: New apps such as personal assistants
  (agents, in AI language) or virtual “friends” could increase traffic, especially if they generate lots of audio or video content.
• New AI-powered connected devices: Innovations such as factory robots, or new remote-driven vehicles, might especially send more uplink data traffic, suitable for AI-led analysis or control.
• More AI-led processing of local sensor or camera data: These could send more data to the cloud over wireless networks. For instance, AR glasses or
  industrial production-line quality inspection cameras could benefit from AI.

All these sound possible, even likely, generators of data traffic over wireless.

But almost none of these aspects of potential AI uses imply the need for more exclusive, high-power, wide-area spectrum for mobile operators. In fact, most of these applications’ traffic will be generated indoors, or on enterprise sites – where Wi-Fi or local / private networks using bands such as Citizens Broadband Radio Service (CBRS) will play a greater role.

Other use-cases already exist, such as AI recommendations for content to watch, or AI-curated social media timelines, so incremental changes and personalization are unlikely to move the needle much further.

There could be a few exceptions such as autonomous vehicles on public highways, but these will do most processing locally onboard, especially data-heavy image and radar/lidar output. Relatively little will transit the network in real-time – except for critical data and telemetry. Bulk data can be sent overnight while parked or in a garage or at a charging station for re-training the model, either over uncongested off-peak mobile networks or local Wi-Fi.

Overall, any extra AI-generated wireless traffic presents at least as much an argument for more unlicensed and shared / dynamic spectrum as it does for exclusive-licensed bands for MNOs.

But maybe it means less traffic?

There are various scenarios where AI might actually reduce network traffic.

In particular, there is a rapid growth of interest in “edge inferencing” and on-device AI. These have the potential to alleviate network traffic levels, by processing data and tasks such as image recognition or speech processing locally, rather than relying entirely on sending data to and from the cloud. Many new phones, laptops and other systems are now building in AI-capable chipsets.

In addition, looking at the mobile market today, 100s of millions of people are using mostly text-based AI applications. This is in stark contrast to the mobile industry’s forecasts of a few years ago, which expected much the same group of leading-edge consumers to be using cloud-gaming or AR/VR, with much heavier usage of the network. Instead, their phones now have apps installed from ChatGPT, Perplexity, Gemini, DeepSeek and various others.

Considering more recent and emerging innovations, new “agentic AI” tools and “reasoning” applications get the AI model to do lots of work “inside the cloud”. They link many separate functions, to complete complex tasks for an end-user behind the scenes. There may well be less data sent across the “last mile” access network, while backbone connections between datacentres are more heavily used instead.

Another theme to watch is AI-led “semantic compression” which could transmit images or video much more efficiently, using the meaning and description of content, rather than just mathematically encoding the bits. Researchers suggest this could yield 10x or even 100x improvements in data required for some applications.

AI means more efficient use of network and spectrum capacity

At the same time, AI is also helping to improve wireless network performance and
efficiency in many ways. It is the central plank of initiatives such as AI-RAN, as well as being a core area of study for new 3GPP 5G and 6G standards. By dynamically adjusting network parameters and predicting traffic patterns, AI can enable mobile networks to handle increased traffic loads, without necessitating additional spectrum.

As mobile industry body 5G Americas itself notes “artificial intelligence is reshaping cellular networks, driving efficiency, adaptability, and innovation”. In other words, AI will mean mobile networks should be able to transport more data, with the same infrastructure and spectrum. AI will itself help the cellular industry transport more data with fewer resources.

AI will also play a role in future versions of dynamic spectrum-sharing, allowing more efficient use of scarce spectrum resources. This is an area where the US has a notable R&D lead at present.

Conclusion

AI will undoubtedly change wireless networks, and in some cases the volume or pattern of data traffic. But it is currently very unpredictable exactly where and how that might manifest. Default assumptions forecasting materially increased traffic on mobile networks may very well be misplaced. Indeed, there are even reasons to suggest it could reduce traffic levels, or improve efficiency of spectrum use.

We should definitely keep our eyes open about AI’s impact on wireless – but there is no direct causal link between maintaining AI leadership and a requirement for more spectrum.

Dean Bubley is the Founder of Disruptive Analysis. He is one of the leading analysts covering 5G, 6G, Wi-Fi, telco business models & regulation, and the emergence of technologies such as quantum networking and AI. This Expert Opinion is exclusive to Broadband Breakfast.

Broadband Breakfast accepts commentary from informed observers of the broadband scene. Please send pieces to commentary@breakfast.media. The views expressed in Expert Opinion pieces do not necessarily reflect the views of Broadband Breakfast and Breakfast Media LLC.