No idea what 'power BI' is in particular, but R has over 10,000 packages, so I could imagine 'power BI' covers a small component of what R has to offer. Also if 'power BI' is a commercial software, it likely does not get updated as fast as R packages, since they don't have to wait to be industry vetted and released via an update or maintenance package.
For example, I use SAS for many core analytics, but it does not have LASSO logistic regression for the binomial distribution, so I perform that along with correction for possible false discovery related to the model building process in R. I can think of countless procedures that are available in R, but not say SAS or STATA, etc. without using tedious macros.
PowerBI was used at my old job. If I recall correctly histograms aren't even supported out of the box and you need to download a custom module for that. So doing something like a faceted histogram - pretty much out of the question. We wanted to do that though so I wrote some R to bring that fairly basic functionality into PowerBI. In another case we wanted to fit some logistic regressions and visualize those. I had to write some R code to do that in PowerBI. PowerBI is really nice for some things but sorely lacking in other areas. R can help fill in those gaps if you know what you're doing.