Let's rerun some maths with current numbers:
According to Garr's message here:https://bitcointalk.org/index.php?topic=67547.msg3226890#msg3226890
We are expecting 14x2TH/s boxes, bringing our total to nearly 29TH/s. Both Garr and CoinTerra have confirmed that they are on schedule for January recently.
The last difficulty increase was 41%, which is higher than anticipated. It is unlikely that the diff will continue to increase 41% every 2 weeks between now and January, but I'll use it as an example.
Echoing my previous comments, time to find a block on average T = diff * (2**32) / rate. Expected blocks per unit time B = 1/T, expected BTC per unit time P = B * 25BTC/block
P = 25.0 * 1.0 / ( diff * (2**32) / rate ) => BTC/second (multiply by 60*60*24*7 for weekly profits)
It is the middle of october, so say there is 2 weeks in october, 4 in november, 4 in december, and 4 in january. This means 7 difficulty adjustments. Call it 8 to fudge for growth reducing the cycle time.
Difficulty at the end of January (projected) => 268mil * ( 1.41**8 ) => 4187mil
P = 24.38 BTC/week
At 26420 shares, that is 0.000923 BTC per week, or an annualized return of 0.048 BTC per share, or 20% @ 0.25BTC per share. (of course, the annualized return is instantaneously measured as of january, the actual return over the year will depend greatly on the difficulty leveling off, cognitive continuing to grow exponentially, or both).
Long story short, the maths look ok, even with this "back of the envelope" calculation, but we certainly hope the difficulty will not increase by 41% every 2 weeks between now and January. If the difficulty increases by less than 41%, we receive our mining hardware before the very last day of January, or other beneficial opportunities arise, we stand to make even more profit. The equations are all up there, so feel free to plug in your own estimates.
Let's also remember that mining in general is a poor investment at this time, but if you are going to invest in mining, the way to do it is via a company like cognitive where you benefit from the economy of scale (and Garr's bulk pricing deals). Buying a single share of cognitive today is, as long as cointerra comes through, like buying 1.08GH/s to be delivered in January, meaning at 0.25BTC/share you are paying 0.23BTC/GH in January. Evaluated on the BTC/GH alone this doesn't look like a particularly good deal, except not a lot of people can drop 6k on buying hardware direct from cointerra, but pretty much anyone can afford a few shares of COG (also, it beats the USB block eruptors event at their most recent pricecut, and has way better power/scaling perf). If you look at the future projections of 1 petahash by Q2 2014 it is the long-term value of cognitive which justifies this price IMO (we all hope! =D ).
EDIT: it is worth noting that my previous maths used assumptions about difficulty growth being around 65% per month. This one assumes 41% growth per adjustment period, which is roughly the network DOUBLING every month. Whether this estimate or the other is more accurate remains to be determined (and is probably up to KNC and cointerra's delivery schedule).