Fermi estimate of YouTube's video storage footprint in exabytes

You are interviewing for a product manager role at Google. Estimate how much storage YouTube needs to hold every video on the platform. Explain your assumptions and arithmetic, and say how confident you are in the final number. Some interviewers ask a narrower version: estimate only the new storage YouTube must add over the next 12 months, rather than the total for every video it already hosts. ```hint Break the total into factors Think separately about how many hours of video YouTube hosts, how many bytes one hour occupies once it has been processed for playback, and how many copies are kept. Watch the difference between bits and bytes. ``` ### Clarifying Questions - Should the total cover only the files kept for each video (the upload and its encoded versions), or also redundant copies and the caches that sit close to viewers? - Is the question about the bytes actually stored, or the capacity provisioned including free headroom? - Which content is in scope: Shorts, livestream recordings, ads, and private or unlisted videos? Should non-video files such as thumbnails, captions and metadata count as well? - Do you want today's total only, or also how fast it grows each year? - Is the estimate worldwide, or for one region only? - Should I assume every video is kept forever and nothing is deleted, or subtract videos that have been removed? ### What a Strong Answer Covers - A clear equation that splits total storage into hours hosted, bytes stored per hour, and redundancy overhead. - A cumulative-hours estimate that reflects how the upload rate grew over YouTube's history, with every assumption labeled. - A per-hour size that accounts for several resolutions and codecs, with correct bit, byte and unit conversions. - Explicit treatment of replication, caches and headroom, followed by an independent sanity check and a range rather than a single figure. ### Follow-up Questions - Which assumption moves your answer the most, and what data would you use to narrow it? - Most stored hours are rarely watched. What storage policies could cut cost without hurting viewers, and how would you measure the impact? - How much new storage does YouTube need each year, and what does that imply for its hardware budget? - Suppose the interviewer fixes redundancy at five full copies of every file. What do your total and your yearly figure become, before and after headroom?