It probably depends on how much demand there is for their computing power and at what price. Operators might also like edge computing in the sense that it reduces the need to upgrade fixed networks.
However, I would believe that initially, the computing power of the GPUs will be set to a level believed to be sufficient for mobile peak performance needs a few years down the line. Would their computing capacity even match a medium-sized data center? An AI grid is coming anyway, so why not build the infrastructure ready for it if there’s essentially nothing to lose. Nokia’s radios cost the same with GPUs as they did with their own chips, and the higher power consumption can be offset by AI-RAN energy savings and likely by selling this computing power.
There are already interested parties. Let’s see what this looks like in 2027–2028. Maybe it’ll turn out to be a dud, or maybe Nvidia is making sure it succeeds somehow. They are reportedly good at selling their graphics cards. Even if it means paying people to buy them. 