Investing in AI Hardware Presents Multiple Challenges
“In the future, every single base station, every single radio network would become an AI powered radio network,” – Nvidia founder and CEO Jensen Huang
During Nvidia’s May 20 earnings call, CEO Jensen Huang made many bold predictions, including the one above. He also explained to investors that Nvidia is now reporting its data center business in two distinct segments – hyperscale public cloud providers who get custom-designed silicon, and everybody else who gets Nvidia’s off-the-shelf chips.
Wireless carriers typically fall into the “everybody else” category, even though they may at times collaborate with Nvidia to design specialized chips. This segment is competing for a limited supply of hardware as prices for memory skyrocket.
Among the leading U.S. wireless carriers, T-Mobile appears to have the most aggressive roadmap for AI-RAN, but all carriers will need to develop more autonomous networks in order to efficiently manage the traffic AI will generate. One challenge will be timing the investments they need to make. Lead times for some GPUs can already exceed a year, and waiting too long to invest in AI could leave networks unprepared for what’s ahead. But moving too fast could lead to capital expenditures that may not deliver near-term ROI.
David Eckell, National Business Development Manager for Enterprise and Data Centers at Graybar, notes that the traditional carrier capex cycle is 8-10 years, based on the cadence of 3G, 4G, 5G and now 6G, which is not expected to launch commercially until 2028. Server hardware cycles are much shorter, and this could mean that investments in AI-RAN would need to be monetized more quickly if carriers want to justify the expense of keeping their AI hardware current. To justify that expense, they need to understand customer use cases. A good portion of the investment in 5G ended up enabling networks to support growing data traffic, but did not generate blockbuster returns. Operators want AI-RAN to take a different trajectory.
Virtualized base stations using high performance chips for AI-RAN can in theory support other workloads as well. This possibility has created excitement around the idea of AI at the network edge as a service operators can monetize. According to Eckell, demand for these services has not materialized yet. He shared his perspective during a panel discussion at Connect (X) entitled “Driving Towards AI with Telco Edge Infrastructure.”
“We’re not seeing the money flow,” Eckell said. “AI doesn’t need the edge. The edge needs AI.” Eckell does not anticipate a near-term rush of investment from hyperscalers or enterprise customers looking to invest in AI at tower sites, but he does foresee potential opportunity for the telco industry in adjacent spaces. He sees metro edge gaining traction through CDN‑style architectures, regional aggregation, and metro compute.
By Martha DeGrasse, Inside Towers Contributing Analyst
This article represents the opinions of veteran telecom industry editor and journalist Martha DeGrasse, an Inside Towers Contributing Analyst with features appearing monthly. DeGrasse owns Network Builder Reports and contributes regularly to several publications. She was formerly a writer and editor with RCR Wireless and a TV business news producer.
