Interviewers ask for estimates to see whether you can reason about scale before drawing boxes. The skill is a few multiplications done carefully, with units written down and assumptions said out loud: requests per second at peak, bytes per request, data per day, servers at a target utilization, and how a cache changes the load behind it.
Plan for the peak, not the average, and keep headroom: servers run at a target utilization such as 60% so a node can fail or traffic can spike without a cascade. Storage grows every day, so state the retention period. Bandwidth is requests per second times response size. Round sensibly and say which number would change the design: a 10x error in storage might not matter, a 10x error in peak writes might change the database.
Estimate servers, storage, bandwidth and cache effects at peak, with headroom.