Yes costs are going up and staying high (for now*), but surely basic logic says that the design costs are a fixed cost, so it makes sense make up for that with volume.
Also, doesn't x3 design costs includes the common work? That is, whatever these Navi share with any as yet unannounced Navi parts and next gen consoles. So it's not just the masks and validation which are expensive, but the whole design. Again, a reason to try to sell as much volume of each chip as possible.
So it makes little sense for AMD disappointing all their potential customers with these disappointing prices.Even if they are limited by wafers and would rather use them for Zen2, mindshare is important.
And looking at how little perf/watt they gained from 7nm, it's pretty likely that when Nvidia get to 7nm Navi will be well beaten in perf/watt and AMD will then have no choice but to cut prices hard, or sell very little. Which is again not a good way to spread the very high fixed-costs of masks etc.
* at some stage around 5nm or 3nm, progress is likely to get even slower or stagnate. Then it makes sense to design only for volume. Something like AMD's chiplet design makes the most sense then. And with no new nodes, the same design might get used for 5 years or more, so economies of scale would be the big factor.