That's not my experience. In ~20 projects in this company, with pretty complicated GIS-related web apps, there have been about 4 cases where I need to create a raw SQL query because Django's ORM didn't do what I needed. Almost all the time, it just works.
Because for pretty obvious reasons in the first case you'd want computation on the back-end. Sure you can do a lot of spatial operations with python after passing through an ORM... but is this really more adapted than raw SQL & PostGIS? Genuinely curious.
What users want to compute can be defined by users, e.g., "take all houses in the Netherlands as polygons, compare with a DEM of the Netherlands to find the lowest point of the house's contour, compare that point to a water depth raster to see if it would be inundated" is a computation a user could define in our GUI and the actual computation would then run in (Python) background tasks.
I work in GIS field and I tend to put a lot of business logic into PostgreSQL because I can re-use all functions/view in a variety of tool from Desktop GIS, BI and even (shamefully) Excel for some our oldest App.
Whenever we'll move to a new webmap or I we'd want to serve an API it's always one connector away. All business logic put into the DB is heavily reusable because a lot of tool "speak" SQL.
I guess in the end it really depend on wether you approach is data centric vs functionality centric.
This doesn't even make sense. It kills the SQL duds, stars are even more important to understand the convoluted SQL when the ORM isn't working as expected, and cases where the ORM can't handle at all (there are always some).