Some 20 years ago, the fundamental nature of broadband networks shifted because of a single change in consumer behaviour to streaming video, redefining what subscribers expected from their connection and what operators had to build to support it. Now, research from wireless technology management firm Plume suggests that agentic artificial intelligence (AI) may be driving a similar turning point.
In Plume research, The upload era: How the agentic wave is reversing the direction of home internet traffic, the company said “good broadband” no longer means only how fast content arrives – it also means how reliably critical data now leaves the home. When a stream buffers, people wait; when an agent fails, work can stop.
That said, Plume raises questions about upstream capacity and network visibility among other issues. What comes next, it stresses, will be measured one router at a time
Putting the research into content, the study noted that despite broadband infrastructure traditionally being built on the assumption that homes download far more traffic than they upload, the internet was now starting to flow in reverse. In addition, it said the question of whether large language models (LLMs) would take hold in the home has been answered, with adoption beyond doubt.
In the post, the company cited other recent research with Parks Associates showing AI has firmly entered the mainstream, with 59% of US internet households now using AI tools, up from just 40% two years earlier.
Exemplifying this shift, Plume noted that while, for example, Netflix requires much more download bandwidth than upload to deliver a quality streaming experience, new network data has shown AI agents flip that equation further, with some agentic traffic sending much more bandwidth upstream compared with downloads.
While this traffic is concentrated among a small group of power users, Plume emphasised it is the direction that matters. Most homes now use AI, and some agentic workloads send far more data out of the home than they take in – the reverse of streaming. That, says Plume, has real implications for how operators build and sell broadband.
Specific telemetry showed subscribers were clearly adopting AI, and what was key to usage was not how long a home uses an LLM, but how often.
Some 86% of US-based Plume-enabled homes use an LLM at least once a quarter, and two-thirds use one regularly. Homes using an LLM 10 or more days a month had at least a 93% likelihood of remaining active the following month, and almost never cut back from there.
The threshold was 10 days. Beyond that, Plume found usage may grow, but did not shrink. Those premises represented 47% of all homes and 95% of all LLM minutes.
Assessing the key issues from the data, Plume was adamant that even though the data showed what tomorrow’s broadband workloads will demand, those leading the transition were still a minority.
Going forward, Plume observed that as AI moves from conversation to delegation, the networks that succeed won’t simply deliver more bandwidth. They’ll understand the applications, behaviours and outcomes that matter most to subscribers. The bottom line was that the agentic wave has begun and that the question was no longer whether it will reshape broadband, but how quickly operators will recognise the shift.
