Agreed. The context was with Prime Video where they had microservices that needed to pass video frames around to one another, and they did this by uploading the frames to S3 from one service, and downloading them from S3 in the next service: https://www.primevideotech.com/video-streaming/scaling-up-th...
The first major "fix" was moving the logic from these separate services into a single service so that the network steps were not necessary. Once they only had a single monolith, it made sense to explore tools like EC2/ECS that run single services well, and often for cheaper than serverless tools when you have predictable scale (or otherwise efficient auto-scaling).
For companies operating at extremely large scales and dealing with expensive, hard-to-cache operations like video monitoring, it's likely financially and computationally worth it to optimize that scaling. For many of us, it many not be worth pre-optimizing that (even Amazon in that story tried to use a step-function based approach to start with). Or it may be worth starting with serverless, expanding your app in production, and changing over once you hit sufficient scale and can predict your usage.
When the tweet that reads this story sensationalizes the story as "jfc the prime video team absolutely kneecapped AWS serverless", and then in the follow up tweet says that AWS says folks "can't" use serverless when the tweet they reference says "should", and then this HackerNews post reduces even more context in its title, this does feel quite misleading.
The point of cloud scaling is that it’s dynamic. I don’t have to worry about an upfront cost for server hardware because I just rent as needed. If traffic drops, I decommission instances and vice versa.
After a certain point, my traffic is always high enough that I can justify the upfront server cost. That’s “sufficient scale” as mentioned in the post.
Well... yeah. Title seems misleading.