I agree, without log files the issue could be anything from an IP pool refresh or subnet change on the tower, a buffer overflow (segfault) on the IoT device, AWS server config not liking persistent connections or, even 'critters'. How do you upload? Push with FTP, http post, SCP, or pull with an always open port at the AWS end? Are you using a DynamicDNS service? A quick fix might be a reboot script on the IoT device that runs every 24hrs.
EDIT: You mention nothing about the terrain or the EMC environment the IoT device is operating in. I might suspect the problem is the cell tower is hanging up your connection as it assumes you are a cell phone that has been accidentally connected for too long. We don't know? Is the IoT device still running or has it crashed? The quick solution for network loss is to have the IoT device examine the HTTP response code from the upload attempt. If it is anything but OK/200, hang up the modem and reconnect. Remembering to buffer the data needing to be sent.