Switching Costs in Software Development

Amazon has a great service, but when I experiment with things on AWS the cost always seems to end up at $30-$100 to run the experiment on AWS where when I use Google they typically send me a bill for $1.48 or more recently I got a $0.24 bill. It always seems more expensive to even kick the tires on something with AWS versus GCP.

Having said that the company I work for now was spending so much on AWS they had to do something to reverse the trend. We changed how we provision and deploy services on AWS and it has brought our costs down approximately $75,000 / mo. I don't know what we were paying before versus now though so I'm not sure how much we saved as a percentage.

If you have an MSDN license through your employer, Azure offers $50-150/mo credits depending on license types for dev and test environments. I've never run into issues getting feedback on prototypes in test from real customers, nor have I ever broken through that credit on a single app. Occasionally I'd pass $150 if we were running 3-5 apps simultaneously all with some measurable load but often the "usage" was in the $5-50/mo range.

Whether your employer will let you use their credits for personal use is another issue entirely (I've had several who didn't use them at all and several who let you use some).

That was the first that came to my mind too.

I think the "sunk cost fallacy" is related here, although I'm not sure how to draw it out.

In general, it seem obvious if any third-party platform is irreplaceable for your business, you are running a significant risk. I'm not sure "irreplaceable" is always correlated with number of hours you've sunk into it.

It could be irreplaceable despite having very few hours sunk into it.

Also, even if you've sunk a bazillion hours into it, if there's a viable replacement that could be put into place _without_ spending a corresponding bazillion hours on it, it might not be irreplaceable at all. (This part might be the 'sunk cost fallacy' -- just cause you've spent a lot of hours on it, doesn't neccesarily mean it would be a bad idea to throw all that away and start over -- if you can afford to.)

Anyway, the hard part, of course, is avoiding an irreplaceable dependency on a third-party platform. Especially because using such third-party platforms significantly decreases your total cost to launch or operate. The trick is figuring out how to take advantage without making them irreplaceable. I'm not sure there are any obvious or easy answers. Using the option with the smallest market share, as the OP suggests, might help encourage a better marketplace, but probably _increases_ your actual direct local risk.