(this obviously assumes no huge increase in the BTC price amongst several other things)
Yeah, that seems possible. In either case, the result to the miner is the same - hardware will have low profitability.
Just a few thoughts here: The cost to run 1TH of 28nm will be about $70/month. Assuming an ROI of 10 months as a standard that was common with video cards, that means hardware costing $700/TH would be the limit. And I bet mfg's would be profitable at that price. But that equilibrium won't happen until about a difficulty of 33 billion (it's currently about 148 million), or stated another way, until a little over 200 times as much hashing power hits the network. Hmmm. Looking at it like that, it kind of seems manufacturers would have a hard time delivering that much hardware anytime soon. Cointerra was talking about dropping 2 PH on the network by year end. That would only be 13 to 14 times the current hash rate. They and other 28nm manufacturers would need to ship ~16 times as much hardware as that to hit 33 billion.
I'm curious as to when miners will collectively seem to not buy hardware. Will it be when the the electrical cost is 20%, 25%, 50%, maybe 90% ... of the monthly coins generated? I don't think it has historically been much more than 10%, but I'm not sure.
Historically it has been a LOT more than 10% before; it's hit 100% before for standard GPU mining in places with mid-priced power, so those numbers aren't crazy at all.
Also, the figures for the network aren't daft either. CoinTerra dropping 2PH was with only 4k chips. 4k CHIPS!! That's the same as in 60 BFL singles! So multiplying their produciton alone up by a factor of 16 isn't at all unrealistic on the face of it.
I'm run off my feet right now, so can't contribute much more to the discussion, but just to say that those numbers aren't silly, and that electricity will always be the limiting factor, I'm pretty sure.
Good points. The only difference however is that BFL SC Single's chips are smaller and simpler architecture (28nm is about 5.3 times denser). So, if you proportion that to 60 BFL Singles x 5.3 = 318 BFL Singles. Still not a lot. And then there are the other 28nm mfg's. So, from that, I'd still guess miner prices will drop a lot and relatively soon.
Density here is not relevant, and the cost of mask sets does not scale linearly with process node - 28nm is a LOT more expensive than 65nm. But once the NRE is paid for, chips are, as they say, as cheap as chips! So once CoinTerra have made the first 4k, the rest are pennies, which as you rightly say, means they can afford to sell cheap.
The take-home point here is that shit is going to go down, and, just as it was before, the winners are going to be those with the most energy efficient miners (or the ones who got the inefficient chips early enough that it didn't matter, but that is only a temporary win).
Agree with the take home point. But wanted to edit the comment anyway. New one would have read:
EDIT: Good points. Seems likely that prices will drop fast and relatively soon. Only thought I can think to add is that Cointerra chips are larger in size per chip.
I wonder how much fabric 4000 Cointerra(CT) chips would cover in terms of BFL chips? If one Cointerra (CT) chip is 100 times faster, you'd need 100 BFL chips. But CT chips are about 5.3 times denser (i.e. 65^2/28^2). So, I guess the fabric size of a CT chip would be 100/5.3 = 18.86 times a BFL chip. So, maybe 4000 CT chips would be more like making (60*18.56) 1132 BFL Singles? Still something they should be able to manufacture in relatively short order.