There are tens (hundreds maybe) of thousands of enthusiasts and engineers out there with computers more powerful than the Azure servers, and nobody has proven that remote interactive physics is feasible over the internet.
And if it's not interactive (iow : pre-rendered events), what's the point? That data could just as easily be stored on a disc. For the destruction to be unique it has to sync up with your game and be immediately variable. With pre-rendered events there really isn't much limit (other than what your GPU can draw of course).
Also, it's now halfway through 2014, we're talking hypotheticals about a 2016 title, and yet there are no true proof of concept demos? Hell, with the ID@Xbox and other smaller-dev companies supposedly having free Azure access, it would be cake for them to do a small test out there to show what they're talking about.
I'll tell you what they're talking about : stuff that's barely feasible on local gbit lan with stupidly high resource usage.
A Physx hardware card requires a fair amount of bandwidth to run for crap's sake :
http://physxinfo.com/news/880/dedicated-physx-gpu-perfomance-dependence-on-pci-e-bandwidth/
We're talking about bottlenecking happening even at the PCI-E 2.0 X1 level (500 MEGABYTES/SECOND EACH WAY). And that's with near-zero latency. I don't have to tell you that an internet connection is at best common case 10-20x slower (8 Megabit = 1 Megabyte, so a 500 MB/sec internet would be labeled as 4000 Mbit! With a ping of .1ms lol.
Keep believing PR spin and all the silly buzzwords you want, but remember, every corporation in tech has been known to have this craptalk through the ages. If you believed everyone, we would have realtime ray-tracing a decade+ ago in realtime consumer gaming graphics and all kinds of other silly stuff.
Even Intel, the kinds of processors, had promised all kinds of amazing stuff with a project called Larrabee, and that went into the ether.
I'm not saying that the R&D and lessons learned along the way won't be useful integrations along the way, but this story : 'Cloud physics will allow much more advanced local interactive destruction effects/etc' is just not feasible on so many levels that it boggles the mind. That ANYone believes it is stunning. It's a thousand times more ambitious than streaming a game over the internet, and I don't even believe that's a great idea overall anyway.