The Virtuous Circle of Fiber and AI

Artificial Intelligence (AI)’s explosion is one built and dependent upon fiber to continue its rapid growth, both within and outside of existing and new build data centers. Rock-solid, high-speed, low–latency broadband is necessary and vital to feed the ever-growing demand for data to train AI models, enabling businesses and households to unlock new productivity.  

While new software models, chips, and nuclear power plants have received the bulk of attention, fiber deals and investments aren’t far behind. Zayo’s March announcement to acquire Crown Castle’s Fiber Solutions business for $4.25 billion cited the need to invest in fiber to support the growth of AI across the United States. In August 2024, Lumen announced it was working to more than double its intercity network miles to support AI for cloud data centers, enterprises, and public agencies, reserving 10% of Corning’s global fiber capacity for each of the next two years. 

“We’ve made several key announcements over the last nine months, particularly with the hyperscalers,” said James Feger, Senior Vice President, Product Segments, Lumen.  “Microsoft, Google, AWS are all building to get data in and out and between data centers at an aggressive rate.” 

Large scale data service companies are buying 400 Gbps and faster services in multiple quantities across different locations to support large clusters and get their language models distributed closer to their workloads and users, with some very interesting implications beyond straightforward human to machine interactions. 

“Traffic’s traversing between those data centers and what I would call the end user,” said Feger. “To me, an end user doesn’t have to be a human. It can be an application, another AI engine, a variety of things that would be consuming the information out of the language models. We’re starting to see this trend [where] inference and AI is becoming more specialized – deployed into distinct segments or virtualized domains tailored to specific tasks. But they still need to talk to each other, share context and operate with intelligence across boundaries.” 

AI to AI communication or AI rings of processing is something that Feger expects will generate a “massive step change” in traffic, with humans the slow part in the value chain. 

“When AI starts talking to AI and they’re learning from each other at the speed of machine, that’s a very different traffic set,” Feger said. “I don’t know that anybody has a real understanding of how much traffic that can generate. I remember the movie Short Circuit with Johnny 5, ‘Need more input!’ We have set ourselves up for ‘need more input.’” 

The continued investments in both fiber and faster electronics by Lumen and other service providers ensures that broadband won’t be the bottleneck for AI, but power is likely to be an ongoing issue, resulting in a reassessment of network models from centralized data centers to an everything-old-is-new-again revival of edge computing and smaller distributed data centers. 

“To me, it’s the edge out solution,” said Feger. “The term ‘mini data center’ is not accurate. What we’re seeing is purpose-built for edge use cases. Look at what Microsoft announced at their Azure conference for example. You’re starting to see this push into the edge, where massive enterprises get their own [dedicated] versions of these deployments. To me, that’s where the ‘edge’ or mini data center concept exists. It could still be sitting in a massive data center, but that’s where the applications are distributed, how they’re viewed, and where they sit in architecture.” 

Moving compute functions closer to users with available power to run them is cheaper than “pulling the power supply to where the data centers are,” Feger said, with Lumen and its customers looking at new build locations based on where there is available power, since it’s exponentially easier to bring fiber to a new building or campus than to build new transmissions lines or the years it would take to get a small modular nuclear reactor licensed, built, and placed on-site.  

Drinking the AI champagne 

However, Lumen is doing more with AI than simply selling bigger pipes to feed the needs of data centers. The company is embracing the technology for network management and applying it to provide better service, as well as using it for cybersecurity. 

“Internally, we’re driving very aggressively for full AI automation,” said Feger. “People like the AIOps concept, so we’re focusing on reduction in human error, network reliability, faster restoration times.” 

Lumen was reluctant to share details at press time, but Feger indicated that the company would have more public discussions of its AI usage in the near future. Some of its work has centered on improving customer experience and time to restoration of service, building on its ongoing work to build and operate a digital twin of the entire network in support of Lumen’s Quantum Fiber service, as discussed in the Fiber Forward Q1 2024 edition.

AI is driving hyperscaler data centers to increase both bandwidth and fiber connections, according to Lumen Technologies Senior Vice President James Feger. (Source: Lumen Technologies)

“We’re able to simulate configuration, network behavior changes, so we do a lot of pre-work modeling,” said Feger. “AI is looking at this, to test new configurations, do traffic flow analysis and really give us a better level of confidence before we go in and put ‘hands into the network.’ The other piece is using the digital twin for predictive analytics in customer support. We also have a product that leverages AI at the core through our threat research team, Black Lotus Labs. Because so much internet traffic rides through Lumen’s network, we have unique visibility into the internet backbone and we’re using AI algorithms to generate threat intelligence to help detect malicious traffic much faster. The whole concept of Zero Day starts becoming Zero Hours, Zero Minute type of stuff, versus waiting for the end of the day to find out that something’s happened.” 

At the applications level, Lumen sees customers leveraging higher fiber bandwidth and AI for different verticals to deliver more efficiencies that Feger could only speak of generally due to proprietary considerations.   

“We have customers in all verticals that are presenting extremely interesting use cases, which is driving them to consume more services across our fiber network, and utilizing our Private Connectivity Fabric infrastructure,” said Feger. “It’s simple things. A long line at the [checkout] register can be a bad customer experience and can result in lost sales. Think of a retail store looking at their security camera footage [in real time] and counting the number of customers that are showing up to help with checking customers out. There are three or four associates in the back restocking or doing other work. They get an alert and start coming forward automatically to staff the registers rather than having the lines back up and having to page people.” 

Financial services have long used AI for fraud detecting, to deliver better experiences with how customers work with their money, and by providing insights for retirement plans, but today’s tools and agents are becoming much more intelligent, customized, and tailored to the individual, instead of offering relatively generic advice.  

“It’s fascinating going out and chatting with the customers to hear how they plan to use AI and why they are coming to us for different connectivity and security options,” said Feger. “An industrial drill manufacturer now has intelligence inside the drill. Those large drills have compute and sensors on them now. The drill is feeding data back up to the surface to make sure they’re at the right speed, checking for the sediment they are going though, ensuring they are not headed for an expensive blowup.”    

At Fiber Connect 2025 in Nashville, Feger will be discussing the impact of AI at a keynote session on Tuesday, June 3, 2025. It should be no surprise that the Fiber Broadband Association (FBA) recognizes the symbiosis of AI and fiber and its importance. FBA released the Accelerating AI with Fiber: Systems and Strategies white paper in March 2025 and embarked on created an AI committee at the December 2024 Premier Members Meeting. The committee’s first official meeting took place at the end of March and chairs are expected to be announced in the near future.  

The 24-page Accelerating AI with Fiber: Systems and Strategies paper examines fiber’s critical role across three areas of AI-driven innovation: data centers, fiber networks, and homes. FBA expects to release a second report, AI Fiber in Perspective, in the second quarter of 2025. The report will present AI fiber trends using a People, Process, and Technology model and an assessment from growth and industry leaders.