Well that's clearly not true.
As long as your secure world is not fully isolated but has any interactions with the physical world at all (e.g a human being somewhere receiving and reading your message with his eyes), then it's only a matter of resources allocated to trace you. You can pile up layers of "hops" through uncooperative jurisdictions -- this certainly helps to raise the bar but doesn't give you a mathematical proof of security.
Consider a building or a server. You can absolutely make them secure. Sure, eventually, everything can be broken/bypassed/hacked/cracked whatever, but if there is no chance of that happening for the duration that the security has to persist, then it is secure.
I'm not sure it's a good example. A server that you build from off-the-shelf components will likely come with the IME, providing direct tcp-to-ram access. Motherboard manufacturers probably add their own backdoors on top. We know about Gigabyte because they were caught red-handed, but how many we don't know about? How many rootkits in the SSD firmware? In hundreds of other firmware blobs installed on your Linux server right now?
I'm not even talking about Open Source backdoors which are hard as they have to be done in the open. Hardware/firmware backdoors are not in the open, they have been around for decades, they have been found and confirmed numerous times and god only know how many were NOT found.
Building a secure server nowadays is an extremely complex task, only solvable at the government level perhaps and only an a few select countries, if solvable at all. You need full control over the whole supply chain that includes tens or hundreds of thousands of corruptible employees.
