@Jon:
I think you're not too far off... except you didn't include yields in your calculation, and that makes a HUGE difference: for such a big chip, it has to be the main cost-driving factor
a 12" wafer on a high-end CMOS process costs $3K to $5K; maybe this one is not so high-end, but still: the problem here is how many working chips you get out of a wafer
if your numbers are correct and they pay $5K per wafer, that means only 40% of the chips actually work (if they pay $3K, it's 24%); if that's your yield and you try to build a chip that is 40 times larger, you can expect to get one that works out of every 10^16 chips you manufacture
I think they have vastly better yields, but even at 90% (which would be my guess here) that gives you only 1.5% working chips when you build a chip that is 40 times larger (=0.9^40); at 95%, it goes to 13%
bottom line: this thing is massively expensive to produce, unless you can make it with enough redundancies that you get super-high yields