I'm thinking that the 100k chip order will last at least 4-6 months, maybe longer. We are only looking at 10-15k to cover the initial preorders that ship "though the end of the year" that cut off in September IIRC.
One of the latest tidbits that leaked from Josh was that Rev 1 (i.e. not this run, January or later) would potentially be packaged in ceramic, so there might be a 25-50% faster unit available from them within a few months. This is mixed news as early adopters will have to trade-up or resell their old unit to get the same speed, but it might or might not be worth it.
We will have Moore's law speed increases for a bit yet which is about 1.5x faster per year, as well as some catch-up that could get closer to modern processes. These are NOT the last ASICs to be produced or sold, unless bitcoin fails entirely, and soon.
I also did a rough model on log10, and it seems like the curve is too steep at the beginning, or the end depending on how many time periods you use. I'm assuming a period of large clearing batches (5k? 10k? chips/week) for backorders from mid-December (network impact date, not ship) to mid Feb. Then smaller batches once they catch up, in the hundreds of chips per week, maybe thousands if the +/- is still high.
This updates my current hypothesis:
I think BFL (and a lot of others) are making a solid bet on the long term value of bitcoin rising. We should be able to look forward to at least 50-100% price increase in 2013, maybe much, much more. Once the hockey-stick curve comes on BTC value BFL and others want to be all spun up and ready to sell into the consumer level ASIC wars that will result.
Right now we are modeling the supply side of the equations because our demand on ASIC has been starved for so long. We also have to remember to look at the demand side of the equation where more folks will rush to participate if the margins (or perceived margins by those bad at math) get too high.
So right now BFL (and others) are making some money, but when I can buy a terahash for under 200BTC, they get to say:
So, I think it's a bad idea to count out price changes, do your model at a few different price points for EoY with a linear model and see how it effects things.