Time-series databases use a really simple algorithm to compress the data and reduce storage footprint by ~10x.
Delta Compression (encoding) is one of the simplest algorithms for compressing a series of numerical data. Most time-series databases use this (or some variant like delta-delta) to reduce their storage footprints without losing much query capabilities.
The core idea is to store the difference between the current and previous points instead of the actual data point. Given a low variance in most time-series data (ex: CPU utilization, API response times, etc), the numbers can be represented in far fewer bits than a standard 32/64 bit representation.
I spent some time coding this Delta Compression to understand it better and benchmark the reduction in storage. On a random set of monotonically increasing integers, I saw an 8x reduction in storage space which is pretty impressive for something that can be implemented in a mere 15 lines of code.
You can find the implementation and benchmark results in my Database Fundamentals repository linked in the comment.
You will always understand more by coding than by reading. So, build quick prototypes of any and every system or concept you come across.