It depends on what you're doing with the data. If you're moving it to some long term archive then you only need to count the write time of the disks at the source, packing time, and unpacking time.

If you're planning to actually use the data you need to add in the time need to read the tapes and put them on a usable medium.

Once you add in all of the overhead, the network comes out looking much better in comparison.

>Once you add in all of the overhead, the network comes out looking much better in comparison.

That's a broad assumption. There's a reason why AWS is finding demand for Snowball and Snowmobile. Internet connections just aren't that fast.

In total bandwidth the truck is still likely to win, but it won't be a 75x advantage. Closer to 3-4x.

Also note that the Snowball/Snowmobile fit into the first use case. They are loaded up then just plugged in on arrival, and even then the total number of enterprises that have used the services is not very large. The multi-day latency on the first packet means you need to be transferring enough data that it would take more than those few days to transfer on a STM-4 or similar. It works out to a tremendous amount of data.

Of course if you are an arctic researcher who has collected hundreds of petabytes of sensor measurements and only have a satellite backhaul, then this makes tons of sense. But for anybody with a solid backbone connection the calculus is much less favorable.

Also the risk of the truck having an accident and having to "replay" everything.

It is significantly less than the time required to lay the appropriate fiber coast to coast.