From 1999 to 2003 I ran a startup where we had a digital signage solution with displays placed around Copenhagen that had a built-in OEM GPRS phone module. They ran a full TCP/IP stack, and were able to update the displays on demand.

If I remember correctly we had it up and running around Q1 2000 which seems to be before this (Not quite sure though, the article doesn't mention specific dates)

IIRC, when we were evaluating GPRS modules for potential use in the hiptop, we would consistently find ourselves testing modules that worked well for a minute or two at a time but would fall over flat if you tried to keep a connection going for any length of time.

If all you needed was a few dozens of seconds of connectivity — such as, I'm guessing, to update a sign — that was probably just dandy. However, that's not how the hiptop worked; it generally aimed to open a connection to the "mothership" soon after boot, and it would hold it open, for hours or even days, network willing.

Well, for certain definitions of "working" it did. I remember not so fondly trying to do emergency work with SSH over GPRS back in the day, with incredible lag and connection freezes every couple of minutes. Perhaps they figured out how to make TCP/IP work _properly_ over such connections.