I think you're being premature with your assessment. NVidia knows exactly what's going on, and have promised further updates to remedy the situation for both BFV and Metro Exodus:
NVidia DLSS, your questions answered.
This is a good explanation of the shortcomings, reasoning, and potential of the tech.
However, I'm not sure it exactly justifies it from a transistor budget.
If we accept that an RTX die is :
~50% 'legacy' features
~25% Tensor cores
~25% RT cores
Then, with the same transistor count and die size, one could hypothetically have a 50% faster GPU with the 25% Tensor removed and replaced with additional standard cores/HW. Eg; 100+50+50 = 200, so 150+50 instead.
If this holds up, then you end up with a GPU that instead of having to rely on DLSS and all the mixed results, simply having 150% performance across the board. Like, DLSS is supposed to make 4k more feasible for slower GPU, but wouldn't an extra 50% of raw grunt make the leap to higher native resolutions be more reasonable by comparison? Certainly it would be hugely more effective across ALL titles rather than depending on specific DLSS support with all of the compromises it comes with :
Per game support
Per resolution support
Only useful at certain framerates, and seemingly useless for high Gysnc/Freesync ranges
Legacy games, thousands of them, almost certainly never supported
Etc.
If we read between the lines a bit, I think Tensor does have great potential with deep learning, professional and academic applications, all kinds of cool stuff. I just remain unconvinced that it is something that will ever be 1:1 worth the trade-off for gaming.
RT on the other hand, despite my reservations about these early days (I feel like it should have probably cooked on Quadro/Tesla type cards for a couple of gens at least), actually will have an important upgrade to bring to gaming, undeniably.
It's an interesting situation. I definitely wouldn't characterize Nvidia as being greedy here, these die sizes are gargantuan, it just feels like the pro/con/cost/performance metrics are a bit borked here. I have to imagine that RTX2 will be a fairly massive improvement starting from this as a baseline. 5nm or 7nm, and no longer comparing directly to a previous gen where so much was sacrificed by ratio just to fit this stuff on.