I think we can't really separate noise reduction and sharpening with this kind of algorithm, because the new algorithm "creates" details.
We're now increasing SNR not by simply reducing noise, but also by increasing "signal". We're creating synthetic details "over the top" of noise, so while we might not exactly be
sharpening the image, the new details we're creating respond to sharpening (even LR's
default sharpening) far better than the noise grain they replaced.
In the old LR regime, I would apply sharpening and NR simultaneously, attempting to create the sharpest image I could that didn't over-sharpen the noise in the image (for the intended output size). It was generally a balancing act where I tried to use as little masking as possible, and as much "Detail" setting as I could get away with, without creating a very noisy file.
In the new LR regime, if I see noise I apply "AI" noise reduction. Afterwards, I sharpen the image as if it were a fresh photo. And because there's very little noise remaining, I can
sharpen the heck out of it.
The end result of this new noise reduction feature is sharper images. It's just our workflow is different than before.
Here's an example of an ISO6400 shot pushed two stops. (So, effectively ISO25600 in the shadows.) I used the new NR on the left, taking advantage of the less-noisy file to apply much more sharpening than I could on the old file. Noise is reduced, but I also gained detail!