The very first gRPC call between a client and server pod involves establishing connection which takes time, its basically setup over a network (TCP, handshake, name resolution, etc.). Subsequent calls reuse this established connection, making them much faster. gRPC supports persistent connections. So, you might want to configure keep-alives correctly to prevent premature termination for subsequent RPCs.