Best algorithm for large scale procedural level generation?

I guess this question has been asked somewhere too, but it seems i can't find it.

Anyway. So i got a project, which is going to be developed for Oculus Quest 2, which means its a mobile game. It will be used for training porpuses, so it has to be realistic.

Here is the problem: I managed to create a terrain 500x500 with 1m punctuality (made of multiple tiles). I want to populate it with environmental elements, like trees, grass, rocks and so on. But to look realistic i thought it has to have a resolution of 1m too, so I created a node at every meter (array of positions only but it is also 250 000 point) and i try to populate it. Currently i do it with item ID-s and an object pool, to save memory, and avoid creating hundreds of thousands of gameobjects. But it takes minutes to calculate the 250 000 points and their item ID-s alone even on my top grade PC. My current algorithm places objects with different sizes on the map, and disables overlapping nodes. Refreshes everything, recalculates distances from objects of nearby nodes, and repeat... thousands of times.

Question is: Is there any faster algorithm for this? Like... populate a chess board with different sized tiles?

Thanks in advance

Nova