AI Networks Face Triple Threats: Capacity, Licensing, and Supply Chain

Policy experts from institutions such as Lumen and AT&T emphasize that as AI traffic grows, capacity demands, licensing issues, and supply chain constraints are becoming increasingly prominent.

At the Aspen Forum of the Institute for Technology Policy Research held in Aspen, Colorado, a group of service providers and policy experts discussed the challenges faced in the design and construction of AI networks – capacity, latency, policy, and rising supply chain costs all need to be addressed.

The licensing issue urgently needs to be resolved

After recently selling its residential fiber optic business to AT&T, Lumen Technologies is facing licensing issues as it expands its network to support billions of dollars in transactions and various AI use cases with ultra large enterprises. Lumen’s Chief Public Policy Officer Melissa Mann stated that Lumen will have approximately 17 million fiber miles by the end of 2025 and is expected to expand to 58 million fiber miles by the end of 2031. Mann stated, “This is not just an engineering issue, but also a policy issue that concerns our ability to meet these needs.” She pointed out that it is uncertain whether Lumen can obtain all necessary permits within the timeline expected by the mega enterprise. If we want to truly achieve this goal and double the fiber optic capacity of the entire industry, we must solve the licensing problem.”

Larger capacity, lower latency, and programmability

The increasing popularity of AI, especially proxy AI, is expected to change data traffic patterns – communication between robots will increase the time networks spend performing tasks and reasoning. For example, Cisco predicts that network traffic related to AI will double in the next three years, and possibly even more. This is related to the overall growth of network traffic. Giulia McHenry, Senior Vice President of Public Policy at AT&T, stated that the operator’s total data traffic has increased by 15% since 2023, although not entirely related to AI. “But we are ensuring readiness for AI to traverse our network.”

Mann pointed out that up to 50% of the Internet traffic on Lumen’s network is driven by autonomous agents. She added that providing services with low latency is becoming a basic requirement rather than a special feature. “Delay is no longer a preference issue. Many AI use cases and applications have minimum requirements for latency.”

As large enterprises seek to rapidly expand network capacity and control capabilities, the demand for programmable layers in networks is also increasing. Mann stated that 90% of Lumen’s customers now use multiple AI providers and cloud service providers. Empowering them with more control is the main driving force behind Lumen’s recent acquisition of Alkira – the company enables partners to orchestrate and transfer their data to different cloud and AI providers through a single management platform. She explained that this means that if businesses notice a surge in energy prices in Virginia, they can shift their workloads to other areas with lower energy costs and sufficient capacity.

Supply chain constraints continue to drive up costs

AT&T’s McHenry stated that supply chain costs are “rapidly rising” and pointed out that the amount of fiber used in a large data center may be equivalent to the total amount of fiber laid by the company in a year. Harold Feld, Senior Vice President of Public Knowledge, a consumer advocacy organization, pointed out that supply chain constraints and capacity demands may also affect the BEAD network construction project. Feld said, “In the face of shortages, we have to make a choice between building BEAD and building super large scale enterprises because resources are limited.” He pointed out that if ultra large enterprises exert pressure on pricing, contractors will face pressure. “Contractors who cannot afford the price can only wait longer before starting work.”

Mark Walker, Vice President of Technology Policy at CableLabs, stated that supply chain constraints, particularly in terms of memory, also affect broadband customer terminal devices. Walker said, “As we build more network miles and upgrade our networks, the increase in memory costs is directly transmitted to capital costs and the ability to provide services.” Feld added that if broadband service providers are forced to pass on these hidden costs to customers without significantly improving their services, customers will feel dissatisfied.

After the FCC banned new foreign made routers, modem, gateway and router manufacturers had to take additional measures to obtain conditional approval for models requiring modifications to memory and other components.