The disclosed technology relates to systems and methods for automatically scaling down network resources, such as servers or gateway instances, based on predetermined thresholds. A system is configured to detect a reduction in one or more network metrics related to a first server, and instruct the first server to issue a rekey request to a plurality of devices connected to the first server. The system is further configured to instruct a load balancer to route to at least one other server responses from the plurality of devices to the rekey request, and determine a number of connections remaining between the first server and the plurality of devices. The system may be further configured to instruct the load balancer to terminate the first server based on the detected number of connections remaining between the first server and the plurality of devices.