So you want to build your own data center

For your first Re:

That would have been great, but that is just more inconsistency. Some resources exist in resource groups, but some don't and you cannot nest them. IAM has the same problem, you always have to create elements on two sides since Entra is not really an Azure resource, it's parallel to your tenant. Policies for Azure don't exist in Entra, but in MGs and Subscriptions and RGs they do. Those don't affect Entra of course, so now you have two different non-interacting policy trees, except you can reference Entra principals. But not if you want to target STS instead. But you can't always target STS, because that would mean you wouldn't have to buy a premium version of IAM (be it P1 or P2 or PAM). Technically RGs would have never needed to exist if they had their tagging and policy system available from day one.

For your second Re:

Instead of having 1 class of groups or containers, there are many non-interoperable versions. You know who doesn't do that? Everyone else. Same for say, IAM. Principals are principals. Tokens are tokens. Want to authorise something? One universal policy language that can target principals, tokens or a combination. Want to use metadata? That's available too, including tags. Applies on all resources the same way as well. Sure, you'll still not find perfect consistency (looking at you, S3, with a 3rd extra policy option), but there is no artificial distinction or segmentation. There is no 'conditional access' product since we would just call that a policy. There is no 'PAM' product since again, that's just a policy. There is no 'premium' because all features are always available, to everyone. And you know the best part? It's not a parallel tenant construction, it's just part of the same namespace of all other resources. Even Google's weird identity setup treats it all as the same organisational namespace.

It's not like Microsoft is unaware of all of this, they are digging Azure-flavoured graves (for legacy setups) faster than Google can expand their own graveyard, and some features that were really late to the party (like MGs, RBAC, PIM, tagging scope with policies as well) are not surprising to see. But repairing a large fractured product like Azure is iffy at best. Time will tell.

For the BTW: yeah, everyone can in the end run virtual machines, but a cloud just to run some VMs is a real good way to burn money. The value proposition of a cloud is elasticity and consistent API-driven resources (which includes IAM, policy language and tagging). A web UI that starts and stops a hidden VM is essentially just a VPS and plesk straight out of 2005.

From the way persistence is implemented on Azure, you can pretty much tell it's all just personal templated VMs underneath, which is exactly what I don't want. I don't want a "storage account" that configures a bunch of inflexible subresources. Say I want to store some blobs, I'd want to make a bucket for that and on that bucket I'll do my tagging, policies and parameters (availability, durability etc). And then I want to do it again, but with slightly different parameters. And then I want to do it 100 times again with various parameters. So now I need 100+ storage accounts too? Who thought it would be a good idea to add a storage account as an intermediary? Probably nobody. But the technology wasn't ready, so instead of going witha good idea, they went with "this will fit on the spreadsheet of the sales department" and released it. Below the surface somewhere hidden from the public API, this reserves some SAN for you, as if we're playing datacenter-for-hire in 2005...

You might wonder: why does it matter? It matters when you do a lot of changes every day, not just deployments or cookie cutter rollouts, but many different applications, services and changes to existing resources. Entire environments are created and destroyed with 100's of resources many times per day per team, and we can't sit around waiting because Azure wants to stop and cleanup an instance that they run under the hood, and we definitely don't want to pay (6 to 7 figures) for such a construction. We want to make use of fast public services that provision and scale in seconds and have APIs that will actually do the job instead of time out and return internal errors. If a cloud isn't really a cloud, but behaves like a datacenter with windows PCs in it, it doesn't do enough for us.

I'll admit, after migrating the last users off of Azure, the only remaining ones are not doing anything cloud-native anyway, it's all just staff-type SaaS (think: Intune, M365 and some Dynamics), so the amount of new Azure knowledge and experience for me over the past 6 months is a lot less than it used to be. The period around 2017 was when most stuff in Azure became a bit more usable with RBAC and AZ Policies, but that was like 6 years too late and to this day is a split world with Entra, yet completely dependant on Entra. Even external identities cannot use STS directly and will have to use static SP credentials. A cursory look at the current docs shows it's still a (premium in secure uses) case. I get it, that's how Microsoft can make more money, but it is technically a bunch of nonsense as other clouds have shown.