Yeah for an OLTP system until you hit certain scale normalized is fine and actually really needed.

For OLAP reporting and analysis denormalized is the way to go for sure. Reducing the number of joins needed (preferably to 0) makes things go very fast.