> The signal servers don't even have that bit of metadata. See [1], they only store the last time that a user connected to the server.
Note that it's what they claim at least. It's not verifiable client side, and to be honest, it's hard to come up with a scalable protocol where this is the case, but you should still not repeat their claim as a matter of fact while in reality we only have their word that the code actually matches what's deployed. And even if they don't store anything, AWS could still provide interested entities access to the infrastructure to capture what Signal doesn't want to capture. Yes, features like sealed sender are awesome and are an important step, but the service still gets ip addresses, which do provide hints about the sender. Again, likely Signal doesn't store ip addresses but people with access to their infrastructure could.
Furthermore, Signal's encryption doesn't help against people storing all of Signal's traffic and waiting until attacks on crypto algorithms become practical (quantum computers, theoretical progress on attacks). Some secrets become irrelevant with time, others increase in value. The best defense is never having the message leave your country's network in the first place.
And there's the DOS problem. What happens if the american president decides that the EU should be cut off from all US network connections? The EU parliament members can't even organize a good response to this because they use an american service...