To do this you would need to check in with a central billing service every time you want to charge, and that central billing service must keep a consistent view per customer to ensure it can't spend over the cap.
This is not too hard if the billable event is, say, creating a VM. Creating the VM takes a few seconds, it's easy to add a quick API call to check billing. But what about the next hour, when you charge for the VM again? You now have the number of VMs checking in every hour (worst case, at the same time), and you need to handle all those hourly checkins consistently per customer.
That's still probably easy enough, but what if it's not a VM hour, but an API gateway request? Now on every single request you have to check in with the billing service. API gateways need to be very low latency, but you've just added a request to that process that possibly needs to go across the world to check in with the billing service running on another continent.
What if the billable resource is "database query seconds", and now you need to know how many seconds a query is going to take before you start it? Oh, and add the check in time to every database query. What if the billable resource is streaming video, do you check in on every packet you send out? What if it's CDN downloads, do you have every one of thousands of points of presence around the world all check in, even though the point of the product is to be faster than having a single far away delivery node?
There are bad workarounds for each of these, but they all mean either the cloud provider losing money (which, assuming a certain scale of customer, is too expensive), the customer over-spending (which assuming a certain scale, could still be waaay over their budget), or slowing down most services to the point that they functionally don't work anymore.