How AI is fundamentally transforming the internet’s physical architecture

Originally published by Fast Company

What one exec describes as “the biggest deployment” in telecom history is underway to provide the bandwidth required to run the data- and power-hungry technology

Lost in the headlines trumpeting the astounding size and price tags of tomorrow’s data centers—$200 billion? $500 billion??—is the question of how to connect them all. While AI’s burning needs for ever more data, compute, and energy dominate tech industry discourse, one piece of the infrastructure puzzle has been all but taken for granted: bandwidth. Not anymore. A comprehensive transformation of the internet’s physical architecture—from the fiber-optic cables to the future sites of data centers—is currently underway, driven by the unprecedented demands of AI workloads.

This overhaul of the internet’s plumbing requires more than simply building bigger pipes. The networks underpinning cloud computing for the past 20 years are fundamentally unsuited to AI’s unique demands. The point-to-point infrastructure that served web and cloud applications so well are proving inadequate for the complex mesh of connections AI requires, in which hyperscalers’ data centers must shuttle massive volumes of data between their customers’ clouds and locations and themselves, with the direction and throughput liable to change on a dime depending on business needs and strategy. A network comprised of a few large threads crisscrossing the country isn’t going to cut it; what’s needed is a fabric binding every node together.

“How are we going to build the mass-scale, low-latency, highly-reliable connectivity to support the future of AI?” asks Kye Prigg, executive vice president of enterprise operations at Lumen Technologies, where he oversees the telco’s network expansion. Lumen’s answer is to lay more than 20,000 miles of state-of-the-art fiber-optic cable, and effectively double its intercity miles across the U.S. Hundreds of construction crews will build hundreds of future relay stations in what Prigg modestly describes as “one of the biggest—if not the biggest—fiber deployments going on in telecom worldwide right now.”

A new approach to network design

Unlike the broadband boom of a generation ago, today’s buildout is designed to meet the needs of a fundamentally different internet. Whereas traditional cloud computing typically manages information moving between organizations’ own systems and individual data centers, AI workloads demand intense communication between companies’ respective clouds and even between hyperscalers. “With cloud-to-cloud, you’re going to have heavy data flows across networks,” Prigg says. This shift not only requires increased capacity, but also a new approach to network design optimized for point-to-point traffic at epic scale.

This reboot starts with the cables themselves. Last year, Lumen partnered with Corning to utilize their fiber-optic cables with dramatically higher density than previous generations. While older versions typically contain fewer than 200 glass strands per cable, new ones will pack 800 to 1,700 strands into smaller circumferences engineered to fit standard pipes. This will not only lower network latency—crucial to real-time responsiveness—but also accelerate rollout through existing infrastructure.

Wavelengths on demand

Lumen is looking to deliver even more customer benefits from its optical relays, which will be located every 50 miles along major routes. Many of these “mini–data-center signal boosters,” as Prigg describes them, will have technology available to provision and reprogram bandwidth on the fly, enabling customers to connect to a vast ecosystem of data centers and cloud service providers and thereby easing complexities of multi-cloud environments and AI applications.

The company is also preserving a significant amount of dark fiber for itself. Having this capacity, Prigg explains, means “an enterprise that wants to light up fiber can get it virtually on demand—we’re talking about wavelengths on demand.” This represents a departure from the industry’s practice of waiting for orders before breaking ground. “That typically takes months and months,” he adds, “so we’re going to have it there already, good to go from Point A to Point B to Point Z.”

Which is a good thing, because—in another break from cloud computing—every hyperscaler’s demands are different in their own way. “They want different data centers to have different connections,” Prigg says. “Some of them want to purchase bandwidth from us and light it up themselves, and some of them want us to do the whole thing, because they don’t see running an optical network as a core competence. There are all different flavors.”

A race to power

Finally, the geography of this new network is being shaped by an unexpected factor: access to energy, rather than right-of-way. “It’s a race to power, really,” Prigg says. “We’re starting to see particular corridors opening up where there is access to excess power in the grid, and these are the locations that are being looked at for these future data centers.”

With “significant deals in the pipeline,” in Prigg’s words, beyond the initial $8.5 billion in AI-related deals the company has already secured, Lumen is building capacity far beyond current requirements—a massive bet on AI’s transformative potential reminiscent of the broadband boom that laid the groundwork for the cloud. But that may be the only thing they have in common. “This is the biggest one I’ve done in my 30 years in the industry,” he says.

^{This content is provided for informational purposes only and may require additional research and substantiation by the end user. In addition, the information is provided “as is” without any warranty or condition of any kind, either express or implied. Use of this information is at the end user’s own risk. Lumen does not warrant that the information will meet the end user’s requirements or that the implementation or usage of this information will result in the desired outcome of the end user. All third-party company and product or service names referenced in this article are for identification purposes only and do not imply endorsement or affiliation with Lumen. This document represents Lumen products and offerings as of the date of issue. Services not available everywhere. Lumen may change or cancel products and services or substitute similar products and services at its sole discretion without notice. ©2025 Lumen Technologies. All Rights Reserved.}