Git stores every historical version of every file as an object. A file edited 200 times has 200 object entries. Over time — especially in long-lived repositories with many contributors — the cumulative history can dwarf the latest snapshot. Git's pack compression reduces this, but the ratio between history size and working tree size typically grows monotonically unless branches are pruned or shallow clones are used.

A shallow clone (git clone --depth N) downloads only the last N commits of history, leaving the rest on the server. For a repository with 10,000 commits, --depth 1 downloads the tip snapshot only — approximately the working tree size plus a thin object layer. This can reduce clone bandwidth by 90% or more for old repositories, at the cost of losing full history locally. Shallow clones can be deepened later with git fetch --unshallow.