I've seen that regularly with spot instances and new instance types. I assume spot instance availability is correlated to on demand capacity (why would AWS discount instances at spot prices of they were able to sell at the on demand rate).

Azure has been having capacity problems all year and some new instance types aren't available to customers or you need special usage requests approved

Data centers are finite and I assume Amazon has some metrics on when to stop allowing new instances (even if they have capacity) so they can continue to serve existing customers.

Depends on the region, availability zone, instance type, and time of day. There are always resources if you are flexible. I’ve only ever seen that message when dealing with spot instances or cutting edge hardware that was in the process of rolling out.

I’ve noticed a number of changes over the past couple of years to make it easier to manage mixed pools of compute resources - both spot and on-demand - a good way for AWS to spread load across different instance classes.

I've heard of it happening to colleagues twice in the last three years (though it has doubtless happened more frequently than that) - I think that last time was because they were requesting some ancient instance type that AWS is probably phasing out.