Adversaries don't always ask nicely. Sometimes they break in and silently take the data. These services centralize traffic flows and make it so that an adversary only needs to tap one or two circuits to get a full picture for all users of a service.
CIA is not stupid enough to break into a guarded data center in Switzerland or one of the less America friendly EU countries. They tell the NSA to look for security holes and spread narratives that only criminals use VPN hoping that a politician will notice and try to ban them, like what's happening in the UK.
Big tech services are less private than you think but almost every provider who cares about privacy is safer than you expect. Most of the people who work there are committed to their mission, and if they ever get a gag order someone will leak it in no time because they know exactly how to do it without exposing their identity.
Are you sure? Someone broke into a Hetzner data center and a Linode one, physically intercepted the Ethernet cables for jabber.ru, and got certificates signed on their behalf.
dear sockpuppet =) I think the above commenter meant illegal physical access. The jabber.ru MitM situation was likely carried out by a cybercrime unit of the German police forces with a court order... and this is the adversary most people forget about.
Lesson to learn: Let's Encrypt does not protect against MitM by a state-level actor, unless you take precautions.
*let's assume network admins won't MitM your server for personal vengeance reasons
Why would they serve a secret subpoena and gag order, when instead they can just drive to a secluded location 5km away from the super secure datacenter, dig a few meters down, passively tap a strand or two, facility and service operators none the wiser?
They’ve already done both. When they can get a cooperative party (AT&T, for example), they colocate their splitter equipment. When they can’t, they tap undersea and overland cables.
When it’s a hostile environment entirely, they hack and do secret operations and bribe.
The data would/should be encrypted; while the NSA did successfully tap Google's inter-datacenter traffic before the Snowden leaks, since then it is encrypted, too. Hopefully other providers won't fall for that trick anymore, either.
IIRC, Google addressed the incident you're referring to by adding E2E encryption to sensitive inter-DC RPC sessions, rather than by fully encrypting inter-DC traffic at the link level. It would be nice to be able to reasonably expect carrier/ISP backbones to be secure against this threat, but in our actual reality this seems like fantastical thinking.
