K-means clustering multi-temporal Sentinel 1 - how does cluster this when it is multi-band?

Hi Guys,

I received a multi-temporal SAR dataset with 30 bands over a region.
I used SNAP to output a PCA product of it to reduce dimensionality, but I also used unsupervised classification on both the 30 band dataset and it’s PCA output (6 bands). However both of these have more than 2 bands (30 and 6 bands respectively). As this makes the dataset larger than 2 dimensions and all I’ve read indicates that k-means clustering only works on 2D datasets.
How is it clustering the datasets?
Does it merge the 30 band stack into one and then cluster?
Or does it just take one band and cluster that?

I tried to do similar clustering in R_studio and Python with the same dataset but was not able so I don’t understand how SNAP is clustering this data.