Estimation question

How much storage space is required to host all the images of Google Street View?

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

Tests estimation combining geographic scale, image data volume, and storage math into one coherent chain of assumptions.

State the funnel: total road miles covered, images captured per mile, image file size, then total storage, plus a multiplier for re-capture/updates over time.
Assume Street View covers roughly 10 million miles of road globally (a commonly cited public figure), captured with images roughly every 10 meters (~0.006 miles).
Estimate images per mile at roughly 160 (1 mile / 0.006 miles per capture point), giving about 1.6 billion capture points total.
Assume each capture point stores a 360-degree panorama at roughly 20MB uncompressed across multiple camera angles, giving roughly 32,000 petabytes, then adjust down significantly for compression, say to around 3,000-5,000 petabytes.
Add a re-capture multiplier, since areas are re-photographed periodically (every few years) and old versions may be retained, roughly doubling or tripling the raw estimate to a final range in the low tens of thousands of petabytes (tens of exabytes).

Builds the estimate through a clear chain (road miles, capture density, file size, compression, re-capture) instead of guessing a number directly.
Explicitly separates raw versus compressed image size, since compression radically changes the answer and shows technical awareness.
Accounts for the fact that Street View re-captures locations over time, which most naive estimates would miss entirely.
States the final answer as a wide order-of-magnitude range (tens of exabytes) rather than false precision, appropriate given the many compounding assumptions.

Forgetting to account for image compression, leading to a wildly inflated raw-image estimate.
Ignoring that Street View data accumulates over time with re-captures, not just a single one-time capture.

Which single assumption in your chain has the widest uncertainty range, and why?
How would cloud storage cost economics change how Google decides what to retain versus delete?