In relation to designing and constructing networks for AI, the challenges look a bit like a multi-headed monster, with capability, latency, coverage and rising provide chain prices all demanding consideration.
A panel of service supplier and coverage specialists debated these challenges and extra at this week’s Expertise Coverage Institute Aspen Discussion board in Aspen, Colorado.
Allowing wants a repair
Coming off the latest sale of its residential fiber enterprise to AT&T, Lumen Applied sciences is going through allowing points because it appears to broaden its community to help billion-dollar offers with hyperscalers and allow a variety of AI use instances.
After exiting 2025 with about 17 million fiber miles, Lumen is projected to broaden that to 58 million fiber miles when it exits 2031, defined Melissa Mann, Lumen’s chief public coverage officer.
“It isn’t simply an engineering query. It is actually a coverage query and our potential to fulfill these calls for,” Mann mentioned, noting that it isn’t clear whether or not Lumen will be capable of acquire all of the permits required to construct as rapidly because the hyperscalers need it to.
“If we’re truly going to do that and double our fiber capability throughout the business, we have to repair allowing,” Mann mentioned.
The panelists did not focus on particular allowing fixes, however specialists talking on a latest episode of Gentle Studying’s The Divide mentioned they consider most of the challenges could be addressed with “shot clocks” (obligatory, time-limited deadlines) and cost-based charges.
Extra capability, decrease latency and programmability
Rising adoption of AI, particularly agentic AI, is anticipated to alter information site visitors as bots discuss to different bots, rising the period of time networks are used to carry out duties and inferencing. Cisco, for instance, forecasts that community site visitors related to AI is anticipated to triple inside the subsequent three years and will improve much more.
That ties into general will increase in community site visitors. Giulia McHenry, SVP for public coverage at AT&T, mentioned the operator has seen a 15% improve in general information site visitors since 2023, although not all is AI-related. “However we’re making certain that we’re prepared for AI to cross our networks,” she mentioned.
Mann famous that as much as 50% of Web site visitors on Lumen’s community is being pushed by autonomous brokers. And delivering service at low latencies is turning into desk stakes, not a particular characteristic, she added.
“Latency is now not a choice. There is a flooring on latency for a lot of of those [AI] use instances and purposes,” Mann mentioned.
As massive enterprises look to rapidly scale capability and management of the community, there’s additionally higher want for a programmability layer within the community.
Mann mentioned 90% of Lumen’s clients now use a couple of AI supplier and a couple of cloud supplier. The flexibility to present them extra management was a major driver of Lumen’s latest acquisition of Alkira, an organization that permits companions to orchestrate and transfer their information to totally different clouds and AI suppliers through a single pane of glass. In apply, that implies that if an enterprise sees vitality costs spike in Virginia, it could possibly shift workloads to a different area the place vitality prices are decrease and the place ample capability is obtainable, she defined.
Provide chain constraints nonetheless driving up prices
In the meantime, provide chain prices are “skyrocketing,” mentioned AT&T’s McHenry, noting that a big information heart would possibly use as a lot fiber as the corporate lays down in a yr.
Provide chain constraints and demand for capability may impression BEAD community buildout initiatives, in response to Harold Feld, SVP at Public Information, a shopper advocacy group.
“We get a alternative between BEAD buildout and the buildout of hyperscalers when there is a scarcity and there is solely a lot to go round,” Feld mentioned.
He famous that contractors come beneath stress if hyperscalers successfully put their thumb on the dimensions on the subject of pricing. “These that may’t afford to fulfill that worth are simply going to have to attend that for much longer for buildout,” Feld mentioned.
Provide chain constraints, significantly on reminiscence, additionally impression broadband buyer premises gear (CPE), mentioned Mark Walker, VP of know-how coverage at CableLabs.
“As we’re constructing these further community miles and upgrading our networks, that improve in reminiscence prices flows straight via to the capital prices and the power to ship companies,” Walker mentioned.
And if broadband service suppliers are compelled to go alongside these hidden prices with out measurably enhancing the service, clients will develop into pissed off, added Feld.
CPE makers are feeling the results of reminiscence shortages in additional methods than one. Within the wake of the FCC’s ban on new, foreign-made routers, the makers of modems, gateways and routers have needed to take further steps to achieve conditional approvals for fashions that require adjustments to reminiscence and different parts.
