By Light Reading
Publication Date: 2026-03-20 08:30:00
First came AI-RAN. Now there’s “AI Grid.” Bandied about at Nvidia’s cult-like GTC gathering in San Jose this week, it marks the chip leviathan’s latest attempt to crack a telecom market that so far has shown limited serious interest in its AI-RAN pitch (where RAN stands for radio access network). Grid is the same word telcos use to describe their countrywide forests of mobile masts, but Nvidia is zooming out to include much bigger facilities in its viewfinder.
Optimistically, Nvidia’s AI Grid could be a dense thicket of basestation cells. More realistically, it will be a much sparser collection of small data centers and refurbished central offices. Nvidia even refers to a real-estate footprint of “about 100,000 distributed network data centers worldwide” in its blog about the AI Grid. That’s minuscule next to the 7 million mobile sites that have now sprouted across the planet.
If it sounds like a retreat from AI-RAN, Nvidia will undoubtedly carry on trumpeting the concept. As for AI Grid, it seems to make a lot more economic and technical sense. But it’s nothing new. Germany’s Deutsche Telekom is stuffing Nvidia’s graphics processing units (GPUs) into a data center in Munich. Yago Tenorio, Verizon’s CTO, has talked about putting GPUs into core network facilities. Greg McCall, his counterpart at BT, reckons the UK operator could potentially house GPUs in some of its fixed-line exchanges. Akamai, an AI Grid customer named in Nvidia’s blog, is not even a telco.
