This deal makes things worse for non accelerated/VC'd startups because now they're competing with companies who not only have funding but also have $100K in free hosting.
If you're trying to validate an idea, not having to worry about hosting costs for the first year or so could be a big deal. Especially if you're planning to sell out later to a larger company who has so much infrastructure already that hosting is practically free.
Hosting costs for validating an idea run about $10-20/month, not $10-20K/month. Get a shared webhost or VPS and run things on a single box. Even if you are using every programming productivity trick in the book (dynamic languages, 3rd party libraries, RDBMSes, etc.), there is no reason whatsoever why serving 20-100 users should cost $100K. Google had indexed the whole web and was serving most of the Stanford campus before they got their first check for $100K.
Do you think you could run a service equivalent to say Google Maps on $20/month hardware under the sort of load that might come from being on the front page of HN and reddit on the same day (Heck, even HN alone)?
Also consider if you are serving HD video or doing financial number crunching over a large data set in the background.
Remember that we're talking MVP here, not production service. If your service makes it to the front page of HN and then crashes under load, you go to YCombinator or some other accelerator, say "We built a service, it was so popular that it got to the top of Hacker News and then crashed under load. Here's the demo link, and we have a list of X thousand users who signed up to learn more before it crashed." You will be accepted, and then you will have access to the $100K in Google Cloud's program.
The one exception is if they feel that you lack the technical talent to build a scalable system even with hundreds of thousands of dollars in funding. In that case, they will tell you "Go find an ex-Googler or ex-FBer to be your technical cofounder and come back to us." In that case, $100K won't make much of a difference, you lack some critical skills to employ that $100K well. If that is your situation, well, I happen to be an ex-Googler with experience scaling Google Search who is in the market for a new project. I have my own ideas but would be happy to abandon them for something with demonstrated traction, so I would be happy to entertain e-mails (my address is in my profile) with a demo link and a spreadsheet with X thousand user signups.
If you need to run your service for longer than the few hours it takes to burn through all of the CPU/disk/bandwidth you can afford to be useful, or want to skip the incubator step it could certainly be useful to be able to apply directly.
It could also improve the value proposition for lower tier accelerators that might have a mentor network but lack the resources to support the infrastructure requirements of the most promising participants. If implemented properly, it should improve the visibility of regional players.
