Estimation question

Estimate the number of servers required for YouTube.

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Advanced technical estimation combining traffic scale, bandwidth, and server capacity reasoning.

Clarify scope: servers for video delivery specifically, likely meaning CDN edge capacity rather than just core data center compute.
Estimate concurrent peak viewers globally, a fraction of YouTube's roughly 2 to 3 billion monthly users, say 50 to 100 million concurrent at peak (assumption).
Estimate average bitrate per stream, roughly 3 to 5 Mbps for a typical HD stream (assumption).
Calculate total peak bandwidth: 75 million concurrent viewers times roughly 4 Mbps equals roughly 300 terabits per second at peak.
Estimate throughput per server, a modern edge server might handle roughly 10 to 40 Gbps of video delivery (assumption).
Divide total bandwidth by per-server throughput to estimate server count, landing in the range of tens of thousands of edge servers globally, before accounting for redundancy and geographic distribution.

Breaks the problem into concurrent viewers, bitrate, and per-server throughput as three distinct assumptions, each stated explicitly (assumption).
Shows the full calculation chain from total bandwidth demand to a server count estimate.
Notes that redundancy and geographic distribution would push the real number meaningfully higher than the raw calculated figure.

Skipping the bandwidth-to-server-count conversion and giving an arbitrary number.
No concurrent viewer or bitrate assumption stated before calculating.
Ignoring redundancy and global distribution needs, undercounting the real requirement.

Chapters of the AI PM course that teach what this question tests.