Similar principle applies to server-side constraints. I run a file processing API on Cloudflare Workers with 128MB RAM — no filesystem, no disk. Had to build a fully streaming ZIP64 archiver that never buffers more than a few KB at a time. It's processed 10TB+ of data so far.

Constraints like DuckDB targeting the cheapest MacBook or Workers' 128MB limit force you into better architectures. You end up with streaming pipelines instead of "load everything into memory" approaches, which scale better regardless of hardware.