Its an interesting choice of battlegrounds. I find the infrastructure vs infrastructure wars fascinating by how Google and Amazon invested in them for their own businesses and are now productizing the 'remainders' and getting less and less for it.
This is a myth. Both companies created their infrastructure businesses as separate entities by leveraging their datacenter expertise, but they were not 'remainders'.
The only way the business model of selling spare resources could work is if you found customers with a different usage pattern to fill the trough. That's hard.
Yes, you can find some customers with time-insensitive batch processing that could be done during the night time. But Google and Amazon already do massive amounts of batch processing, and still won't have a flat utilization level over the day.
Or alternatively you could try to flatten out the diurnal variations by making sure there's no geographic affinity between users and the DCs. But that's suboptimal given intercontinental network latencies.
The Great Recession hit, Google stopped building infrastructure for a time and focused on utilization, which lead to some awesome improvements in cgroups and i/o scheduling and such, and only after that and some early successes of both AppEngine and other projects which were kind of like it but much less well publicized, did the idea start to be considered as something Google might actually productize.
I could completely see them building infrastructure at this stage which is purely for rent, they have the recipe for building data centers pretty well baked. But I have never heard it put that they started out with that in mind.
> Thinking at Google has changed since you left.
Are the prices dropping faster than moore's law? The costs are dropping approximately with moore's law.
Edit: the interesting bit here is just how you lower your prices. In the VPS market, it is traditional to keep charging your customers the same, and to just give them more ram/cpu/disk/network. This is way easier on the bottom line, as your revenue doesn't fall (by much. Most people won't bother changing to a smaller plan, some will.) - in the "cloud" the tradition is a bit more consumer-friendly; the tradition is to just cut the price and hope the consumer will buy more.
Yes, the cost progression of storage is expressed by Kryder's law, and is actually faster than moore's law. (well, in terms of storage space, not storage speed.)
Of course, both are observations and not true laws; things will get interesting if/when the capital and power required per unit compute (or per unit ram or storage) stops dropping.
But, my point is that in the past, the costs involved in providing compute infrastructure has fallen dramatically over time, and most people expect this to continue, at least for a while.
In theory, not necessarily in practice. The Thai floods caused a multi-year pause in Kryder's law we're still recovering from. (I'm not sure how robust Kryder's law is - there's a lot less data and analysis of it than Moore's, since that's the famous one.)