The blog should be updated now, might need a refresh to see:
https://blog.linode.com/2017/02/14/high-memory-instances-and...
I use one of my DO instances as staging for file transfers. 2 days a month, I can barely get more than 30K/s from their sf data center to comcast. Other days, I get 5M/s. Their network is shitty.
And I don't think I'm being rate-limited; it's 1-2G transfer per day.
DigitalOcean might happily be able to push 10g to their next hop, but beyond that things are significantly more outside of their control. This is mistaken as 'throttling' a lot, but is actually just ISPs not investing in peering to handle their peak demand. Similar to the Netflix and Cogent issue that happened a couple of years back.
At times, I can get better speeds by tunneling or proxying using a DigitalOcean Droplet and downloading something from overseas than I can doing it directly. The path taken from my home internet to the DO Droplet uses different peering than my path to Europe, and the speeds are faster overall even though it creates more hops.
So it appears that 75Mbps was the initial speed but after some internal deliberation (possibly influenced by feedback), they raised it to 1000Mbps.
[0] https://blog.linode.com/2017/02/14/high-memory-instances-and...
