Pretty much. And, I don't think anybody questions the fact that "the cloud" can provide large computation capacities. Hell, when I need to run a heavy numerical simulation, I run it on a remote cluster, which is comparable to some extent.
The issue is there's a lot of hand-waving regarding all that supposedly new, extra stuff that'll make games better somehow. The convenience for devs/publishers to have direct and easy access to an allegedly cheap, scalable bank of servers is great. Good for them. That's not what I'm personaly interested in, and that's not what microsoft has been selling with talk of the "infinite power of the cloud" and "4X the computational power when your xbone is plugged into the cloud."
Offloading AI to servers is nice. People were also doing that in early 2000's with bots on dedicated CS servers. Is "the power of the cloud" going to turn next gen AI into something more? Is computational power what's holding back AI in modern games? Maybe. I'm honestly sceptical. Would love to be proven wrong and see actually competent AI in games (for a change. Yes Ellie, I'm looking at you sweetheart).
There's been a lot of talk about computing illumination in the cloud. It is computationaly heavy, doesn't necessarily need frame-by-frame refresh, and results can be downloaded back as reasonably small chunks of data. It's however also been pointed out that these computations could be performed before release, and that this small chunk of data could be stored on disc. Streaming from disc being a hell of a lot more rapid than downloading from servers. Not to mention, performing the computations once and for all ahead of time would seem to make more sense than recomputing the same sets of data for each user, on servers you're renting. In addition, the lighting thing has, as far as I recall, only been mentioned as "something that maybe could work down the line, theoretically, but we're not doing it right now."
So, in terms of visible, measurable-by-the-player benefits of the cloud to games, we're left with: dedicated servers, running AI, possibly running physics (I don't believe specific details have emerged on that yet?), maybe computing illumination down the line ("but we aint doing it now"), and the prevalent "maybe other cool stuff, people will eventually figure it out lol."
In terms of "infinite power of the cloud," that's a little thin. It also is all wrapped in a vague, magic mystical veil of confusing mystery, see the OP,
I have no doubt that in theory, you could do a lot of stuff. What I'd be interested in knowing however, is what can I reasonably expect the impact to be, in practice, for games I'll play.