Apples and oranges. GTA V has a small, entirely hand-built world. It's just 80 square kilometers and was meant to fit onto two DVDs / one Blu Ray Disk. Real-world Los Angeles, which this is based on, is 1,210 square kilometers.

This Flight Simulator on the other hand covers the entire planet. If we are just going by land area, that's 510.1 million square kilometers. It's using a combination of satellite and aerial photography, radar maps, photogrammetry (reconstructing 3D objects - buildings and terrain in this case - from photos), Open Street Map and Bing Maps data, as well as hand-built and procedurally generated detail. There's also information on the climate, live weather data, animal habitats (to spawn the right creatures in each part of the world), etc. pp. We are about two petabytes of data, which is an unfathomable amount outside of a data center.

You can not optimize your way out of this. The developers have the ambition to create the most detailed 1:1 virtual facsimile of this planet. There is no other way of achieving this goal. You can not store two petabytes of data on a consumer PC at the moment, you can not compress two petabytes of data to the point that they are being reduced to a couple hundred gigabytes and if your goal is accuracy, you cannot just reuse textures and objects from one city for another. That's what every prior version of this flight simulator did, but if you remember those, the results were extremely disappointing, even for the time.

By the way, if you don't have an active Internet connection, Flight Simulator 2020 (and 2024, if I'm not mistaken) will still work. They'll just do what you're suggesting, spawn generic procedurally generated buildings and other detail instead (in between a handful of high detail "hero" buildings in major cities), based on low-res satellite photography and OSM data, which is relatively small in size even for the whole planet and tells the program where a building and what its rough outline and height might be - but not what it actually looks like. Here's a video from an earlier version of FS 2020 that shows the drastic difference: https://youtu.be/Z0T-7ggr8Tw

It is worth stressing that you will see this kind of relatively low detail geometry even with an Internet connection any time you're flying in places where the kind of high quality aerial photography required for photogrammetry isn't available of yet. FS 2020 has seen continuous content updates however, with entire regions being updated with higher quality photogrammetry and manually created detail every couple of months - and FS 2024 will receive the same treatment. I am generally not a fan of live-service games, but this is an exception. It makes the most sense here.

The one major downside is that eventually, the servers will be shut down. However, since you can choose to - in theory - cache all of the map data locally, if you have the amount of storage required, it is actually possible to preserve this data. It's far out of reach for most people (we are talking low six figures in terms of cost), but in a few decades, ordinary consumer hardware is likely going to be able to store this amount of data locally. The moment Microsoft announces the shutdown of this service, people with the means will rush to preserve the data. Imagine what kind of amazing treasure this could be for future generations: A snapshot of our planet, of our civilization, with hundreds of cities captured with enough detail to identify individual buildings.