I constantly see this argument but let's face it, it's very unlikely that enough people will ever switch to something like Matrix. I like decentralization and the matrix protocol is brilliant, but it brings many problems:

• Many people will have to care enough to host their own servers (which now is not remotely as common as it should be) otherwise everyone will just use the biggest severs, weakening the advantages of decentralization
• It's way harder to implement new features that people care about
• People are not used to Element's UI and there aren't clients good enough to compete with Telegram, Whatsapp or even Signal
• The performance wouldn't be as good exept for the biggest servers, causing centralization again
• If people don't use it, it becomes useless, which is the same problem other alternatives have. This means that people must want to change naturally to it, meaning clients, ease of use and performance would have to be at least on par with what they're using at the moment

On the other hand, Signal:

• Is very similar to the way Whatsapp works, which is what most people are used to
• The Android and iOS clients are getting better with time (the desktop client needs to abandon electron but it's hard with only few developers and it's a lot of work)
• The protocol is robust and audited
• It doesn't leak metadata, as Matrix does
• Even if it's centralized and Signal runs their servers on AWS, the only useful information third parties could gather are timestamps and the recipient of the message, not even the sender
• It's easy to jump on Signal, the network of contacts already exists and you wouldn't need to ask for usernames or email addresses
• Don't foget that the clients and the server are open source, and even if the Signal Foundation decides to stop working on Signal, shutting down the services (VERY unlikely), we could fork the projects and bring them back up

Centralization can be problematic, but if it's done correctly the pros may outweigh the cons, and in my opinion this is the case for Signal, but I'd happy to be proved wrong in future