How can I efficiently scale a data lake?

I am running into problems with AWS Athena data lakes. They were initially very quick, but as we've scaled, they're performing poorly and are becoming expensive and inefficient. What can we do to solve this without reducing data size?