The Question
How much does a large video platform spend on raw storage in a year? I'll build this bottom-up from the firehose of uploads: how many hours of video arrive every minute, how big a stored hour is once you account for all the resolutions, how long it's kept, and what a terabyte-year of cloud-grade storage costs. This is a storage-cost question, so I'll deliberately ignore bandwidth and compute and just price the bytes at rest.
Assumptions
- Upload rate. About 500 hours of video uploaded every minute. This is the classic figure for a top-tier platform and a fine round starting point.
- Stored size per hour. One source hour, transcoded into many resolutions (4K down to 360p) plus the original, totals roughly 3 GB per delivered hour at modern codec efficiency. I'll use 3 GB/hour as the all-in stored footprint.
- Retention. Video is kept essentially forever, but for an annual cost I care about the stock of bytes that exists, which grows. I'll compute the new bytes added this year and price the year-average storage of that cohort.
- Storage cost. At hyperscale, blended storage (much of it cold/archival) costs about $10 per TB per year. Retail cloud is pricier, but a platform this size runs near-cost.
- Replication. Everything is stored with redundancy across regions, roughly a 2x multiplier on raw bytes.
The Calculation
First, hours uploaded per year. 500 hours/minute x 60 minutes x 24 hours x 365 days. That's 500 x 60 = 30,000 hours/hour, x 24 = 720,000 hours/day, x 365 = about 260 million hours of video per year.
Convert to storage. At 3 GB per stored hour: 260M x 3 = 780 million GB, which is 780,000 TB, or about 0.78 exabytes of new content a year.
Apply 2x replication: 780,000 x 2 = about 1.56 million TB of physical storage for this year's cohort.
Price it. At $10 per TB-year: 1.56M x $10 = about $15.6 million to store one year's worth of new uploads for a year.
But the catalog is cumulative. After several years of uploads at similar rates, the platform is paying to keep all prior years online too. If the back-catalog is roughly 5 to 6 years of comparable volume sitting in storage, the standing annual storage bill is closer to 6 x $15.6M = about $90 to $100 million a year.
Sanity Check
Does $100M feel right for a platform with annual revenue in the tens of billions? Storage at well under 1% of revenue is plausible: storage is famously the cheap part of video. The expensive parts are bandwidth (egress to viewers) and compute (transcoding every upload into a dozen formats), which can each dwarf storage by an order of magnitude. So a storage line that lands in the low hundreds of millions, dominated by delivery costs elsewhere, is internally consistent.
A second angle on the bytes: 0.78 exabytes of new content a year, accumulating into single-digit exabytes of catalog, matches the publicly discussed scale of the largest video archives. If I'd computed petabytes, I'd have been three orders of magnitude too low; if I'd computed zettabytes, far too high. Exabytes is the right shelf.
What Would Change the Answer
The dominant lever is the per-hour storage footprint. My 3 GB/hour quietly bundles a big assumption about how many resolution copies are kept and how aggressively old, unwatched videos are pushed to cheap cold storage. If the platform keeps fewer rendition copies and archives the long tail at $2/TB-year, the bill could be a third of this. The replication factor and the size of the back-catalog are the next biggest swings. Upload rate, oddly, is the assumption I trust most. The honest take: this is a $50M-to-$200M answer, and which end depends almost entirely on storage tiering policy.