For instance if you use "uninterruptable" instances in GCS it is cheaper than using comparable instances in AWS to train neural nets on a GPU.
On the other hand I hear the "uninterruptable" instances in GCS get interrupted all the time, but I know the interruptable instances make training inexpensive in AWS (eg. $8 to build a simple but strong language model) It is a pain to deal with interruption but any job that could run more than 8 hours needs serious attention to crash recovery no matter what kind of kit you're running it on.
I shut down all of my mysql instances in Amazon because I was not happy with the value, but I do have a "CAN-SPAM compliant e-mail marketing system" which uses Lambda, Dynamo DB, SES, SNS, SQS and all that and it costs under $10 a month to run if you have a list of 10,000 people or so and e-mail them a few times a month. It runs trouble free if I walk away from it for a month, whether or not the number of subscribers explodes.
Could you make something like that with the other clouds? I am sure you can. I'd imagine the cost would be similar.
The reason I don't investigate the other clouds when working on my own dime is that I know how to work AWS on a deep level, I know it won't blow up my Python install the way GCS does, etc. I know or have embodied in code the details to create a staging environment in some region and availability zone, placement group, subnet, etc in AWS. It is the same on GCS and Azure but I can save work by not replicating it.
What bends the needle for me is working for someone who is using some other cloud. I think those people are motivated by what cloud their customer uses ("hybrid cloud" rings about as true as "luxury sedan") The pricing factor I see is that it is easy for academics, startups, VARs, etc. to get free credits to use on GCS or Azure. AWS had to do that long ago, but today they are the standard and GCS and Azure have to pay more to get the customer.
The right price is not the price it cost to make something, but the price the market can pay and you can get away with.
I will say, though, that it looks more like an arms race to me. AWS is rarely the most expensive option for long before Amazon announces a price drop.