Google says they are deploying to a fiberhood, for every 250 to 1000 households, and various wikis say that 2.59 people per household in North America, and 560 people per sq. km. in KC... so 1000 households is likely (1000*2.59)/560 which is around 4 sq. km, which, if we assume a CO in the middle means each homeowner needs about 2 km. of fibre.
Lets see, at retail (newegg), I can buy two SFPs for 40$ each. and lets say about $0.80 a meter for optical cable (based on retail price for a 50m patch cable from infinite cable), so thats $1600. Except that you dont need your own fibre, all you need is a wavelength, so google could be just splicing cables together for say 20 households at a time, and you need only about 50m of your own fibre. so that works out to $40 of cable for you uplink to the pole, where there is an optical coupler (say 200$), plus perhaps 5% of the 1600$, so $80 or so... they can adjust using more cable to use fewer couplers... etc...
Then they need to cost out the uplink from the CO. 2x 10 GSFPs say $500 ea. + 20km. of fibre $16000... and one the uplink they need an aggregation switch, say 24 ports / 1000$, which would mean... 42$ for the switch... aww heck lets double it for the uplink of the aggregation switch say 100$ in the CO... so the total is say $17100
say 17K$ for the uplink / 500 households... 34$ per household for the uplink. OK so perhaps that is a little weak as an uplink, but use multiple wavelengths over the same fibre, and you would still need a lot to get to even 100$ per uplink.
So using retail prices, and Googles deployment plans and publically available retail pricing and demographics, the price per link is about 80$ for the SFPs, 40$ for the patch cable to the coupler, 200$ for an optical coupler, $80 for the fibre to the CO, 100$ for the uplink... we are at 500$ for most of the parts of the uplink... now sure.. you can add in the on premises equipment, and get to maybe 1000$ that way... OK, so they charge 300$ for installation, so there is 700$ to recoup... they are charging a 70$/month for the service... so 10 months. pay back.
So I left out labour costs... they might double the payback period, but the business case still looks damn easy. I think your numbers are either phone company motivated, or a decade or two old. either way, they are complete b.s
sources for the pricing: