It's funny, this was actually one scenario that I thought about mentioning but I had to get on a plane and was running out of time.
It is true you CAN do this, but very few do, for a few reasons:
One is, it's bad for margins - when you build a pricing model, you inevitably end up creating a system where some customers subsidize other customers. You assume each user or unit of usage is going to cost you X/unit and you charge X+Y. There is inevitably going to be a distribution of users and their usage patterns and costs, and the 90% percentile is probably going to be 5X, and the 10% is probably going to be .2X. There's not any malice there, it's just that different users have different usage scenarios and they use the product differently.
Another reason relates to the issues with usage-based billing. Even in that scenario, whatever usage dimension you measure on will have users that don't fit the profile and they still end up being subsidized (from a margins perspective) by customers that DO map to the profile. A really naive example - you're a database company, you want to be cheap for people to get started, you go with usage based billing and charge based on storage. For most customers, that works - assuming your product value is apparent and differentiated, I think most people would understand that "I have to pay more because I'm storing more data, and accessing that data can be more expensive, queries more complex, and the utiltiy that I get from the database scales as the quantity of storage increases". Great, usage based billing, let's do it.
But - then you have users who store very small amounts of data but with incredibly high query volumes. Your options are to either just eat the cost of those users (which might be fine for some amount of time) or now start to add additional dimensions on which you meter usage. So now you charge for storage AND cpu time AND maybe concurrent connections if that's a problem AND bandwidth. Congratulations, you have now created the perfect usage-based billing model, which perfectly assigns customer charges to handle the multitude of usage patterns that customers experience.
BUT, it's really complicated to explain to people, and it's really complex to predict costs. That has two implications, one of which is that your value proposition has to be increasingly compelling as complexity increases. To use the database example, at some point someone at a customer will say "honestly, wouldn't it be more predictable if we just spun up a couple of VMs and ran a database instance ourselves?". Complex usage-based pricing works if you've got incredible technology that would be difficult to impossible for a customer to deploy themselves, but if your value prop is convenience and/or abstraction, you're diminishing that value as you make the pricing model increasingly less convenient and less abstract.
The other factor is that someone has to build and manage the metering of all of these things. Even a single dimension like storage is complicated - how do I bill for additional storage? Do I look at the total storage at the end of the month and multiply by X? That hurts users who, say, run end of month batch jobs - but for you, users that use huge amounts of temporary space and then free them before the end of the month, that hits your bottom line (depending on your own architecture). So maybe you want to charge on a daily basis, but now every problem gets more complicated.
Then, if you extend that across multiple billing dimensions, it's just gotten harder and more complicated. Now it's rock and a hard place time - you can stick to one abstract usage measure that is easy to reason about, but you're inevitably going to have some users that underpay based on that usage measure and some that overpay. Or you can add more dimensions and make things more "fair", but everybody's lives are harder, both for the customer and for you and your team.
When you give customers automatic optimization, you get the worst of both worlds - you make less money on the bottom 10% (usage-wise) of users/customers because they end up falling into the usage based billing, and you make less money on the top 10% because there is capped upside for you as the provider. For customers, sure, it saves them money, but what you're really giving them is a price cap (not to exceed X).
I would say for the sales teams, it's also not great, because they have all of the challenges of explaining two different models. For enterprises, it's a mess because 1) they'll probably want to negotiate specific billing terms for their use cases (we don't want to pay X for bandwidth, we want to pay Y) and other structural terms, all of which your billing system needs to support.
At the end of the day, however you charge for anything is an abstraction layer on top of your costs. That's true if you charge per user, or per object, or per gig, or per connection, or whatever else. It's all unit-based pricing even if it's not usage-based procing. You have to decide how much work you want your engineers, customers, salespeople, etc. to do in order to build, explain, and understand how much someone will pay for software.
My general advice is to pick the simplest pricing model that protects your margins and prevents abuse. For infrastructure-y products, things like storage, compute, network, are all reasonable meters. For SaaS products for business users, per-user pricing is well-understood, and there are things you can do if you really want to apply a usage-based element there (bill based on MAU, or have a MAU component separate from seats purchased). But there's really only two scenarios - you pick a small number of meters and understand that some customers will subsidize other customers, or you meter across a bunch of dimensions that align to your costs and create a lot of complexity for your customers. Blending the two gives you worse margins and the complexity of both options combined.