Flat cells are great if you're trying to build a flat product, where z space is much more critical than x and y. However, flat cells have an unfavorable effect on specific capacity (both weight and volume) since you need two flat sheets to package the display top and bottom, sandwiching a relatively thin active cell, instead of a single cylindrical package wrapped around a larger volume.

In another comment [1] in this thread, owenversteeg says that Tesla batteries are not space-constrained, but rather weight-constrained, and that the cylindrical shape is only suboptimal in its space efficiency, not any other important design parameters.

[1] https://news.ycombinator.com/item?id=13482273

My understanding is that the sheer size of Tesla's battery pack is too large to take a layered approach - at least across the whole size. That forces the designers to go with a segmented approach, in which case the cylindrical is tried and true.

From a safety perspective, it also has many more years of data/research behind it. It would be more tolerant to defects. Trying to create small sheets that are then rolled up is easier than creating big sheets, just like smaller silicon processors have better production rates than ones of equal complexity but larger surface area.

Cheaper to buy standardized cells at scale, due to production capacity?