That's like claiming every data collection should have a reasonably limited length. Even if that were true, scrollbars in general break when the data gets too large (a pixel of screen movement jumps pages of data) not to mention browser layout performance, memory requirements, etc.
Pagination doesn't work either, because data can shift and you get repeated elements (if the page size is less than the remaining data when moving back to the start) or else truncated pages, or you start on page 10 and move backwards to page -3 before reaching the start. If you jump to a particular element, it requires you to know the absolute offset of all data to calculate page numbers correctly, which when jumping and filtering a large table might be a significant performance cost. Plus it breaks up content - if you're looking at chat history for example, often related messages are broken up into several smaller messages. Having one message on page 10 and the remainder on page 11 would be a terrible experience.
Don't blame this on designers, infinite scroll has been around (apparently?) almost 20 years now. Browsers still don't have an accessible solution despite the ubiquity of this pattern today - browser devs are to blame.