I read the entire article and that wasn't my takeaway. After reading, I assumed that AWS was (somehow) billing for loopback bandwidth, it wasn't apparent (to me) from the article that CPU costs were the sticking point
From the article intro before they dive into what exactly is using the CPU.