They'll recoup the value of the entire thing, with extra profit, if even just one of the YC startups using it has explosive growth. It's a smart investment. Even without the money, it's a great technology showcase. YC now gets to give its startups the Oscars giftbag treatment. :)
In a way, it's like investing in YC startups then. Give a sizable but not unbearably so amount of money to a large set, hope that statistically these companies have a chance to experience hockey stick growth, and have that one black swan cover all your costs across the whole group.
3 years of technical investment is pretty much enough to stop you leaving unless something goes very wrong. This $500k is a small price to pay for a client who goes on to be the next AirBNB or Dropbox. Even just the prestige of having a company like that to draw on for testimonials would justify the spend.
However...
For a startup to choose to build on Azure they'll need staff with MSFT stack experience. 5 or 6 of those will burn through $500k in no time compared to the equivalent on a more open platform.
Whether or not to use Azure is certainly NOT obvious.
I have made this comment before but I think that it bears repeating: Azure has good Linux support. I find little difference running Ubuntu VPS's on Azure that on other good platforms like AWS, Google, BlueMix, etc.
Does the Linux image run virtualized over an MS kernel? I often laugh running MS code in a VM over __nix because when the Guest crashed, my host OS is still running. I cant imagine running a __nix Guest over a MS Host though.
I dont have much experience with MS servers, as I like to get things done, but how are they when it comes to stability?
For the future YC batches - depending on what your company is going to be doing, it would make very little financial sense to NOT host on Azure.