How does this distributed and cloud-based AI-RAN work, considering that on average, 50% of the computing power remains for things like token generation? Why isn’t anyone comparing this to, say, SpaceX’s edge computing in space? Which approach is smarter or more cost-effective?
A base station site can probably accommodate several dozen cards if needed. I can’t estimate how that compares to Nvidia’s new racks in terms of computing power, but I wouldn’t be surprised at all if major AI-RAN deals in the Middle East are announced in the next review. It’s probably smarter to distribute computing than to build vulnerable data centers.