The problem with this distinction is that scientists who refer to their observations as data are almost never interested in a single datum. They consider any single observation to be decontextualized and potentially anomalous, and look for confirmation through statistically-validated trends across large bodies of data.
Therefore, all the data together can be seen as a unitary piece of information, a single way to understand the phenomenon being studied. It's similar to the way that "story" is singular, even though one story can contain many subplots and characters that are stories in their own right. I think this makes sense as the basis for a singular form of the word "data."

Also note that, at least in the social sciences, "datum" and "data" have largely been replaced by "data point" and "data set," respectively. Perhaps that the proper argument is that "data" is plural in Latin and singular in English.