Nvidia has now ported their Maxwell architecture to a 20 nm chip for the Tegra X1, as found in the Google Pixel C. That would be an integrated GPU with two Maxwell SMMs, alongside a boatload of ARM cores and various other SoC stuff.
TSMC's 20 nm process node, like all other 20 nm process nodes available, is built for low power parts. But a tablet needs a low power part, and Maxwell is the most energy-efficient architecture among high performance ones, so it's a reasonable fit.
But this also means that Nvidia has put in the work to port Maxwell to TSMC's 20 nm process node. That raises the possibility of launching discrete GPUs of 20 nm Maxwell parts. Will they?
You can make a high performance chip on a process node optimized for low power, but it's not guaranteed to work very well. Would a 200 W Maxwell card on 20 nm be better than one on a 28 nm process node designed for higher performance? It could easily be worse on 20 nm rather than better, and even if it is better, that's no guarantee that it's better by enough to justify the cost.
What I think is the more interesting possibility is a lower power 20 nm Maxwell card for laptops. If performance that used to take 70 W can now be had for 50 W, that's pretty nifty. Will 20 nm make that possible? Even if it does, will Nvidia do it? It's not like they need something new to catch up to AMD; Maxwell is already ahead of AMD on energy efficiency, which is a major reason why so few laptops have a discrete AMD GPU.
I'm not predicting that Nvidia will make such a card. But it's an interesting possibility to consider.
For what it's worth, AMD will not do that. AMD has publicly announced that chips that were planned for 20 nm have all been moved to a FinFET process node, presumably 14 or 16 nm, depending on the foundry. The considerations that led AMD to shy away from 20 nm for discrete GPUs could lead Nvidia to do the same. But that includes AMD's tablet chips, and Nvidia just delivered a 20 nm tablet chip.