Once again you make more money selling shovels then mining yourself.
BFL could do that and make a couple million USD over the next year or they could sell the rigs and make a couple million USD on day 0 and now the miners take all the future risk.
6 months later they can then can cut their prices 50% and sell another couple million of the same rigs at half the price.
6 months later they can then they can cut their prices 50% again and sell another couple million USD worth of rigs at 25% of the original price.
6 months later they can then they can cut their prices 50% again and sell another couple million USD worth of rigs at 12.5% of the original price.
If some competitor comes along they will be looking to maximize the return on their huge NRE so while each company may have to reduce their prices somewhat a price war hurts both companies and it is the miners who benefit. The markup on the actual chip (not design, NRE, etc) but the actual chip is 10,000%+. So years and years from now BFL could be selling the same chips (now maybe 1/20th of the original price on day 0). Eventually they move to a new process (i.e 45nm) and start the game all over again.
It is much more profitable to sell shovels and anyone who has enough funds to be in that position already knows that.
Market-wise your points are valid, assuming they do have that long term foresight, but you've missed one of my arguments - the initial shock of 50-100 times more hashpower is unprecedented.
My assumption was to initially use the hash-power to cover their NRE asap (we're talking 2-8 weeks after 25btc drop to avoid public outrage). They already have the money from preorders, remember? The chips must be manufactured in huge batches to be cost effective.
Another point is to drive the cost of mining in a controlled manner. Under your scenario, they'll equip 10% of miners and create huge demand ripple in the 90% who are out of the league now. My naive economic ignorance calls for *huge* price increases, to avoid more obscure pre-orders. This might be indeed more profitable than stalling the delivery by their mining, but only under the assumption their production will be *continuous* which I somewhat doubt.
Oh, or maybe some more preorders and months-long deliveries, for you know what :)
Either your or mine way, you'll end up with something akin to Intel (historically), until AMD and Motorola shows up :)
Note that I'm not implying that intentions of BFL are evil, on the contrary. Who's first to the market, and with good PR, wins, as it should be.
A lot of people have problem with entire ASIC scenario though - because up until now bitcoin was more or less commodity hardware (FPGAs are reprogrammable after all :) and thus not that *much* of waste of resources.
(Pure energy-efficiency appears to be opposite, carbon footprint of manufacture is unknown...).
Now the market is cornered into producing pretty-much-useless-except-bitcoin-which-also-get-obsoleted-fast-by-innovation machines, just because the task to be solved is strictly matter of dedicated silicon, rather than household items which can serve multiple purposes.
scrypt/bcrypt is *far* from perfect, but solving tasks suited for CPUs with high memory bandwith seems to be an alternate step in a nice direction - far in the future it's nice to have competing big CPU manufacturers, rather than clendestine asics (which are *years* behind CPU architectures anyway). It moves forward whole general computing, not just bitcoin mining.
