Batching substantially improved my computation time.
Now I should maybe address the recomputation of already computed values. I do fetch results from the DB when fetching entities. But I do not fetch the results when generating new ones which happen to have been previously generated.
Will it be better if I fetch the DB or worse?
I want a software engineer to setup some good performance metrics!
Follow
We have good news though. One million rows should take "just" about 5GB from these initial results.
That is not too bad. With 5 tera I can store a billion. Should be good enough. I probably do not have enough computational power to get to a billion in any reasonable amount of time.
"reasonable amount of time" = before your retirement age 
Developing a large project in python.
All very cool, we have type annotations that solve a lot of problems.
But interfaces, I'd really like to have interfaces. No, abstract classes don't do the job. I want compile time reminding me I fucked up.