"Even on a cost avoidance basis, the AWS3 bids could reflect economic values," a reliable source emailed me. More spectrum makes it cheaper to add wireless capacity, especially now that LTE-A aggregation is widely available. I ran some numbers and to my surprise even $45B is plausible - if not exactly cheap!
AT&T certainly didn't intend to pay $15-20B for about 40% of the 50 MHz available. John Stankey put $10B on the table for the auctions. That seemed like enough to scare off smaller bidders. Someone wasn't scared, probably Charlie Ergen, and pushed the bidding up.
Telcos need to keep up with traffic increasing at 30-70% per year. They can increase capacity with more cell sites, more antennas, and network efficiencies instead of spectrum. They calculate how much less they could spend on capex if they had additional spectrum available. They calculate cost avoidance numbers from that and develop a maximum bid range. Then they hire nobel-calibre game theorists and auction economists to ensure they spend much less.
The savings in capex just might justify the high prices. AT&T has just cut capital spending by $4B, the largest decline in the last decade anywhere in the world. Just a quarter of that, $1B, would support over $30B in spending on spectrum, at 2.8% effective interest rate.