Agents vs. Cats: Where I’d place my bet on AI’s mobile traffic implications

In my last post on mobile data traffic trends, I made a fairly non-controversial claim: nobody knows exactly what the impact of AI will be. Will AI compression tamp traffic down? Will AI workloads send traffic skyrocketing? Will traffic volumes be relatively unchanged but with different characteristics? 

In the spirit of Pascal’s Wager and in recognition of the undeniably critical role mobile networks play in keeping societies functioning, I’m going to assume that the traffic levels won’t be going down and, at a minimum, we will be looking at different types of traffic.

Two scenarios dominate thinking on this front. 

• Agentic AI Requirements. In a world where we’re all employing multi-agent workflows, the expectation is a continuous flow of inference requests, API calls and telemetry as agents connect with infrastructure and one another. Even where the impact to end-user traffic is minimal, the background noise of agentic chatter complicates network monitoring and policy enforcement while bringing new requirements in terms of short-lived connection scale, uplink capacity, and determinism.
• AI-Driven Content Growth. I’ve heard plenty of reasonable people suggest that Gen AI apps represent today’s version of the video streamers which drove traffic across telco networks without delivering meaningful revenues…for anyone other than the video providers themselves. Our work with operators suggests that this isn’t yet the case. AI’s potential impact on adjacent content, on the other hand, is another subject altogether. Think upscaled video and gaming content, AI-generated video, and AI-enhancements which push people to simply consume more of what they’re already consuming. 

Now, let’s return to the title. Where would I place a bet on how AI will most impact network traffic? Agents requiring a re-thinking of how we dimension network scale or a new-generation or cat videos which force operators into adding more capacity?

I truly love the idea of a world where operators enable agentic workflows, monetizing diverse and demanding requirements. I also loved the idea of a world in which operators enabled all manner of network slices and monetized those too. Circa 2025, however, we’re just now getting past the pilot stage of slicing deployment in part because demand didn’t explode as many had hoped. Meanwhile, Wired is reporting that Open AI plans a standalone app based on its video generation model (Sora 2) featuring a TikTok-esque video feed populated by AI-generated content. Whether or not users would flock to the app, it’s a clear signal of how companies will be putting AI to use. 

It might seem boring – intellectually lazy, even – to suggest that AI will simply amplify current usage and traffic models. But those are usage and traffic models we understand. Presuming we know how agentic AI traffic will develop while assuming innovation will not arise to address any agentic surge is pure hubris. I’m optimistic about the potential value agentic workflows can offer and telcos’ ability to monetize this value in the long-term. In the near-term, people’s unending appetite for cat videos – especially if engineered to be even more engaging – is something I can bet on.