So do a lot of computers.
Plenty of "real" computers don't have any ability to display *any* data to the operator. They just receive data from computers and send the results to computers. Many supercomputer applications simply take arrays of numbers as input and return arrays of numbers as output, with interfacing to humans delegated to lesser systems. But I don't recall anyone referring to Crays et al as "calculators".
Personally, I would make the distinction according to whether it can execute unbounded loops. A conventional calculator either lacks loops altogether or only uses bounded loops (i.e. iterating over the digits). This makes them not Turing-complete (e.g. there's no halting problem, as a bounded loop cannot fail to terminate).