Realtime applications. Anything having to do with video & audio mixing or interaction (WebRTC, VoIP, AR/XR, NLU, Gaming, etc...) Having those as close to the end consumers as possible can dramatically improve the experience and make them far more engaging/enjoyable.

In fact, it's critical enough in many applications that clients may want to access the services in the same cloud they use /as well as/ in the same region (or potentially in their own data center.) So that means designing the systems to deploy on any cloud provider.

While this does limit the number of niceties you can use from a single provider, it allows for some pretty amazing flexibility.

A lot of businesses like to have their cloud ops located in and subject to the laws of the country said businesses are based in.

governments as well.

Yes many hosting companies in NZ rely on that. It's going to be hard for them to keep their niche.

Depends on your architecture.

You're still restricted by physics with edge computing. If you need to make a call back into us-east-1 to get data out of a data store, it doesn't matter that there's an edge CDN and API layer.

Outside of that, there's compliance reasons too. Partitioning your data by region because storing customer information in say the EU is different than the US. More likely CN is much different than the rest of the world if you do business there.

It does in fact matter that your tcp packets are roundtripping over lower rtt connection even if you do have to fetch from the origin

A large part of the actual answer is that they are doing a build in for their first party services like Amazon Prime Video.

Those services don't want to be sent intercontinental because it's just an enormous amount of data, and they will incur back-haul charges. Same is true for customer multimedia services.

The other reason is data governance/sovereignty use cases. With a bit of ultra high latency on the side.

Redundancy, but I don't know how easy AWS makes it to run say, a container, in two different regions. I think their load balancers can balance between regions but I'm not sure how well other services can do it.

What comes to mind is DNS based load balancing. You'd have a full stack running in each region, and users get routed to the one nearest them. I could see this working for something like Dropbox or an email service where users can be easily segmented.

You can do that or you can use their Global Accelerator thingy that gives you an anycast IP but technically they dont have an lb that spans regions (nor would you probably want one)

Yeah I think I was thinking of the Route 53 based region load balancing, not the load balancer products themselves.

There are many use cases for compute in every (or at least, 3+) regions. But it's very difficult to manage. So difficult that most things that would benefit can't afford the complexity.

I don't think it is so much compute in every region as it is data storage there.

My company stores user data and deploys to most AWS regions. It's a big win for some companies to be able to have their data stay in a given area (esp the EU, but I've also heard Canada mentioned).

Anything where you want minimal latency, worldwide.