Reasonable is relative to what you want to do: Cloud instances are only cheap if your machines are usually idle, and you overprovisioned your datacenter massively. Cloud instances are only cheap if there is no great amount of output/input data to transfer like simulation results or large model input datasets. Cloud instances are only cheap if you don't need assurances about locality or security for compliance. Cloud instances are only cheap if you can live with a slow network between the instances, as soon as you want Infiniband or at least dedicated Ethernet links, you are either out of luck or in "really expensive" territory again.

You left out management cost. Once you’re talking something more than a personal laptop the cost of administration needs to be factored in. This is where cloud environments have some big advantages since you’re not paying skilled people to run cables, debug firmware, swap parts, etc. Yes, you can still beat cloud (especially if you serve a lot of network traffic) but it’s not easy unless you’re giving up security, features, or availability.

> run cables, debug firmware, swap parts, etc.

Data centers already do that. You could build your own rack and send it to them, but the more common model is renting preconfigured servers on a monthly or annual basis with them swapping out broken hardware as needed etc.

Cloud computing can save on administrative costs, but there is a huge world between building everything in house and AWS.

Data centers do that for additional costs. It can be a way to save if your needs are simple and don’t change but it’s also often a source of downtime or operational overhead since you’re looking at managing separate groups of people versus using an API.

You can definitely save money this way but it rarely ends up being as much as touted - double digit percentages, not whole multiples - unless you’re really bandwidth intensive.

Perhaps have a look at Profitbricks. They are an interesting small alternative to AWS for a niche market.

If I can't get a machine for local development I wouldn't care personally. I like the idea of many many threads as I work a lot with data, but i need something like a PC to try things on.

Also, don't really like vendor lock-in.

I'm doing some development on AWS currently on g4dn instance with T4 accelerator. With pings lower than 50ms and remote VS Code it gets really efficient. The biggest issue of course is additional mental burden on being remote + you can't simply print image in ssh console...

I thought it would be horrible, but after a brief adjustment period it turned out to be okay-ish for my current work.

But you can view images in terminal emulator. Not every one, but kitty, alacritty, st, iTerm2 for MacOS, etc. should support it. This works for me (need to install ueberzug, e.g. via pip):

    { declare -p c=([action]=add [identifier]=0 [x]=0 [y]=0 [width]=80 [height]=80 [path]="mypic.png") ; sleep 5;} | ueberzug layer --parser bash

In 2004 I was using telnet/X Windows into the UNIX development server of the company, in 2020 I use ssh/Web/RDP for the cloud environment instead.

To be fair it is using their own hardware (Graviton2, mentioned in the article). Other than Ampere are there any other hardware vendors offering ARM based server CPU's that are commercially available?

Amazon's Graviton2 is in some of the Anandtech benchmark results. The availability & pricing is probably related to how far behind it is in those benchmarks, too...