- Traffic used to peak and dip predictably by time of day, but AI has smoothed that curve into near-constant demand
- Upstream bandwidth is growing faster than downstream, reversing a 30-year design assumption built around heavy downloads
- The shift is due to AI inference moving to the edge, particularly video processing
For 30 years, network traffic followed a rhythm you could set a clock by. A lull in the morning, a peak through the business day, demand falling off in the evening. And it flowed overwhelmingly in one direction: down, toward users pulling web pages, video and files. Networks were engineered around both assumptions. AI is breaking them.
The daily curve is flattening
For example, MetTel has seen double-digit traffic growth for eight to nine years, but lately the shape of that traffic has changed. The familiar daily curve is leveling out.
"We have seen that smooth out considerably," said Ed Fox, CTO of MetTel, in an interview for a new Fierce Network Research report, "AI and the automated network: Designing telco infrastructure for the age of inference."
AI is driving the change, because AI doesn't take lunch breaks. Human users generate requests intermittently and wait for responses, producing the peaks and valleys networks were built to anticipate. AI operates continuously, at machine speed, around the clock. As more traffic originates from machines rather than people, the daily curve loses its shape and demand approaches a constant load.
That has real engineering consequences. Networks provisioned for predictable peaks — with headroom that sits idle overnight — face a different problem when demand never recedes. Capacity planning, power management and cooling were all built around a rhythm that's disappearing.
Upstream is growing fast
Upstream bandwidth is now growing disproportionately faster than downstream. At MetTel, total bandwidth grew 30-40% annually, and last year 75-80% of that growth was upstream. For an industry that spent decades optimizing for heavy downstream and light upstream — asymmetry baked into everything from broadband tiers to consumer plans — that's a striking reversal.
The source of the upstream surge is the edge. AI video processing at retail locations, camera-equipped wearables, and operational technology at oil and gas companies using cloud-based video analysis for asset management — all of it generates data that flows up and out of those environments, not down into them. As industry pushes AI inference closer to where data is generated — in factories, in the field, in the store — the return traffic is upstream by nature.
A pattern other operators are seeing
This tracks with what others across the industry report. AI workloads are also more deterministic and less bursty than traditional application traffic, according to T-Mobile's Salim Kouidri, SVP of field engineering — another break from the patterns networks were tuned for. And uplink traffic from physical AI could outstrip mobile network capacity by 2029, with demand rising 20-25% annually through the 2030s, according to Mobile Experts research. The upstream story is early, but the trajectory is steep.
What it means for network planners
These changes are showing up in operators' traffic patterns now, and changing key decisions: where to add capacity, how to engineer for symmetry, how to architect the edge. The networks that handled the download era gracefully were built on assumptions AI is overturning.
Download the full report here: "AI and the automated network: Designing telco infrastructure for the age of inference," and watch the companion virtual summit on demand: "AI and the Automated Network".