With kind assistance by the GeoDatenZentrum, I have managed to set up a prototype of a self-hosted Blazegraph database on a Linux server and populate it with a Wikidata dump file from 6.11.2017. I was basically following the instructions on https://github.com/wikimedia/wikidata-query-rdf/blob/master/docs/getting-started.md . The import took the virtual machine (64GB RAM, 4 cores, TB) about three weeks. It seems reasonable to use a faster machine next time, also because Wikidata grows at high pace.
In addition, the memory swap must be disabled to avoid runtime errors:
sudo swapoff -a sudo sysctl vm.swappiness=0
If properly installed and poulated, a SPARQL endpoint can be accessed via a http request:
http://[ip address of the server]:9999/bigdata/sparql?