Newer generation SKUs (which that one is) are often in short supply. I have plenty of bad things to say about Azure, but I've never had trouble getting capacity older than bleeding edge in eastus2 and centralus regions, at least.
You'd have to be spectacularly stupid to bet on these kinds of things without having insider knowledge, because you ought to know good and damn well by now that the people with insider knowledge are DEFINITELY betting on them.
This is the story of Microsoft - five different ways to do the thing, none of which do everything, and all of which are in various states of disrepair ranging from outright deprecation on up through feature-incomplete preview. Which one do you use? Who knows, but by the time you get everything moved over to that one and make allowances for all the stuff the one you chose doesn't support, there will be a new more logical choice for "that one" and you'll have to start over again. Wheee.
Don't forget maintenance costs in the TCO calculation too. Transmissions, fuel pumps, timing belts, radiators (mostly), fuel injectors, emissions systems, etc are all out of the picture in an EV. Servicing those things may be infrequent but is often extremely expensive.
We ran 100% of our workloads on VMWare this time last year. We'll be at 0% this time next year. We were heading that direction over the long term anyway, but the Broadcom shenanigans made us double down on that effort. They may actually be more unpleasant to deal with than Oracle, which is something I would have thought to be impossible.
$35k - $26k = $9k
$9k / $0.12/mile savings ~= 75,000 miles breakeven