Estimation question

Estimate the cost of running 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

Structured cost estimation across infrastructure categories, checking whether you can decompose a huge system into its major cost drivers.

Clarify scope: annual operating cost, covering storage, bandwidth, and compute for transcoding and serving, not content licensing or payouts to creators.
Estimate storage cost: assume 500 hours of video uploaded per minute, average 1 GB per hour after compression, roughly 260 million GB uploaded yearly, at cloud storage rates near 2 cents per GB per month.
Estimate bandwidth cost: assume 1 billion hours watched daily at roughly 1 GB per hour average, so 1 exabyte of daily egress, at a blended CDN cost of a fraction of a cent per GB.
Estimate compute cost for transcoding each upload into multiple resolutions, a meaningful multiplier on raw storage cost.
Sum the categories: bandwidth dominates the total given the sheer scale of daily watch time, likely billions of dollars annually.
Sanity check against YouTube's known ad revenue, tens of billions annually, to confirm the cost estimate is a plausible fraction of revenue.

Breaks cost into storage, bandwidth and compute separately instead of guessing one blended number, showing structured thinking.
Identifies bandwidth as the dominant cost driver given daily watch hours, the correct intuition for a video platform.
Sanity checks the final estimate against known revenue figures to confirm plausibility.

Estimating only storage cost and ignoring bandwidth, which is the larger driver for a streaming platform.
Producing a number with no sanity check against any known reference point like revenue or headcount.

How would this estimate change if YouTube shifted more traffic to higher resolutions like 4K?
What's the biggest lever to reduce this cost without hurting quality?