It really feels like these cloud platforms are following the Microtransaction model of mobile games.
* e-commerce sites around holidays; ticketing sites right before popular events
* internet fora that get sudden influxes of traffic (e.g. Reddit)
* scientific computing projects that occasionally require huge batch jobs using (hundreds of) thousands of cores [0]
You are correct that for the majority of businesses, whose workload requirements minimally and predictably fluctuate, cloud providers are a rip off. But for the few use cases actually requiring massive transient scaling, there is no viable alternative.
[0] https://cloud.google.com/blog/products/gcp/220000-cores-and-...
You get them for the myriad of other managed services they provide, which let you build stuff that would otherwise take months, in days. The catalogue of AWS services is probably over 200 already, EC2 is just one of them.
Building on a cloud like that is much more "gluing together legos using Terraform and some code" than the alternative.
Configuring and maintaining open source solutions is a lot of work. Building everything yourself is too. I prefer to pay for AWS instead.
The math is different for different companies, workloads, etc.
For some random raw EC2 server? Zilch. Zero. EC2 isn't really worth it by itself.
But EC2 isn't the only thing AWS has. Devs are pretty expensive. If you just need to run some random code in the cloud, futzing with a server somewhere for a couple hours making sure it's compliant with all of the enterprise standards costs many hundreds of dollars in dev pay. Lambda just works, and has a free tier. Setting up a proper queuing system is annoying and takes a few hours to do right. That's another couple hundred dollars. SQS is reasonably good and you can have that dev go do other things instead.
These things start to add up fast, and when you do everything yourself, you start to end up with thousands of dollars per month in maintenance costs. You can absolutely do that with AWS as well, but the goal is usually to just let AWS take care of that and have your employees do something that's actually valuable to the business (usually implementing proprietary business logic, which neither AWS nor Hetzner will do for you).
I recently discovered our logging was costing $3,000/mo at GCP. For freaking logs.
And was there some sort of processing thing running on those logs? You'd have to pay for that compute.
But it's also got me thinking I'm going to buy a bunch of GPUs this month and stand up our own ML compute cluster.
We'll still run CRUD in the cloud, but I'm getting sick of paying for cloud GPUs. The GKE / GPU interface leaves tons of spare compute on the table anyway.
Code, you know, version control, diffs, CI/CD pipelines, inherent live "documentation" (up to a point, of course you won't know the "why", but you'll know most of the "how" and "what", git offering the "who").
I encounter 10-10000x higher bills on aws ; most 0s are self inflicted but the last few 0s I don’t think you can erase in the same way with aws vs a good vps provider.