Nvidia has several things going for it: They can build a card and power section that hash's on efficiency. When they can show they unearth coins perhaps not a quickly but "each coin" they find are cheap(er) to mine... that can prevail. Lower cost of energy used, and if Nvidia can sell two of these for a little over what one "overprice 580" it's win! Miner can hunt around for one big 580 shovel, or get two of these for $350 and perhaps unearth 20% less, but on +30% less electricity cost. It's like this; two little guys with two small shovels that are able to be outfitted and working today, can move dirt near as fast all while feeding and paying them 30% less verses the one big guy!
GP106 is obviously cheap(er) to make and Nvidia probably has more of an opening to more production at TSCM than RTG at GloFo. Also, Nvidia has cash and teams that could be thrown at this project/production, and have such solutions in the market in weeks (as we are seeing). We know the constrains on AMD/RTG.
While lastly if Nvidia get stuck with product when it all goes bust, they can spend the marketing $'s without batting an eye to dump such card for other specialize tasks like; Folding CUDA/OpenCL number crunching etc for cheap... and just move then out for more than the $30 it cost to make.
And they get in on the action even if not a big money make they steer the conversation to shine on them and a solution while thumbing AMD/RTG as can't meet market demand.