Storage capacity for datacenters and other data-hungry business operations (think YouTube scale).
Current alternatives don’t reach the same capacity per dollar without caveats (magnetic tape has an incredibly slow seek time and SSDs are too expensive for non-cache usage).
Of course, AI data harvesting is essentially creating artificial (sorry) demand for even more data capacity, and it doesn’t make rational sense for them to use other forms of hardware.
Fun fact - the scale of data involved is so great that Google famously used “sneakernets” (give an employee a backpack of hard drives and tell them to go from A to B) over traditional internet or intranet connections between their larger facilities in the 2010s, because it was faster.
The data used to create that image of the black hole had a transfer speed of 14GB/s because ~700TB of it was captured in Antarctica and spent 14 hours traveling by plane
Storage capacity for datacenters and other data-hungry business operations (think YouTube scale).
Current alternatives don’t reach the same capacity per dollar without caveats (magnetic tape has an incredibly slow seek time and SSDs are too expensive for non-cache usage).
Of course, AI data harvesting is essentially creating artificial (sorry) demand for even more data capacity, and it doesn’t make rational sense for them to use other forms of hardware.
Fun fact - the scale of data involved is so great that Google famously used “sneakernets” (give an employee a backpack of hard drives and tell them to go from A to B) over traditional internet or intranet connections between their larger facilities in the 2010s, because it was faster.
The data used to create that image of the black hole had a transfer speed of 14GB/s because ~700TB of it was captured in Antarctica and spent 14 hours traveling by plane