I have a question about CPU mining. I've been browsing thru bitcointalk again due to the links you've been sending. And I've seen two things I'm curious as to your take on when it comes to CPU mining.
"Well, CPU minable coins, such as our XMR, have the problem that we have to worry about not just cryptocurrency mining power, but all CPU power in the world that could mine. What if all idle CPU power in the world suddenly started mining Monero? What if someone subverted AWS and pointed all Amazon's hardware at Monero? Could they mount a 51% attack? (Extermely unlikely, I know, that anyone will subvert AWS, but the point is we now need to worry not about someone having 51% of the SHA-256 ASICs, but simply somone having more CPU than the Monero miners). "
"I don't get why people worry about ASIC centralization when the two organizations in the world with the most computing power are Google & the NSA"
Indeed those are serious concerns. But Google and the NSA don't have spare capacity. They would stop rendering existing services to use that CPU power for attacking a coin. Much more of a concern is if the NSA is serving national security letters to specialized or closed source ASIC manufacturers (e.g. any ASIC for Monero is going to be a very specialized likely closed source design, because
it is so complex) and requiring them to sell xx% of their production to them (much more realistic for the NSA to lock up some % of control that way than redeploy their huge infrastructure which is already dedicated to other tasks). Also assuming the $3 - 5 trillion black budget is correct, the NSA has unlimited funds and so we need to very concerned about ASICs that could be (covertly) locked up by them, unlike generalized CPU power that can't be.
Note I made the point that botnets pull from the supply of computers as legitimate mining, thus if mining demand is high enough, the price of botnets need to rise asymptotically to approach the cost of renting the same hardware, so botnets are not an extreme threat at the asymptote where most people in the world mine.
This is why I would want any ASIC to be very modular for generalized use (e.g. SHA2 or the AES round circuit) and easy to reproduce by a wide array of vendors. Ideally you want that any specialized ASIC can't best the modular ASIC by an order-of-magnitude, and it would be really sweet if not more than doubling of performance from the highly specialized ASIC. Amdahl's Law applies but is not sufficient by itself for complete characterization of the solution space.
And the most CPU power in the world resides with the users. They just aren't organized...