I don't think you can pick out any one reason. XMPP is very old and has extensions for a huge variety of features. Many people have experience with older versions which had many major missing features (such as strong multi-device with offline support and server-side history) and a lot of the "hype" has died out long ago.

Matrix is new and made a lot of decisions that really helped its popularity.

1. Having a HTTP-based client-to-server protocol makes web clients very easy to make.
2. It is based on sync and merging rather than messages which moves some difficult problems (like multidevice and server-side history) into the core protocol meaning that it works well out of the box.
3. Having HTTP based protocols make hosting it familiar for many people.
4. The "default" Element clients have lots of features out of the box, features that for a long time were not always present on XMPP servers or clients. This gives a more consistent experience.

We will see what the history holds. Matrix is still very new and maybe the hype will die out and we end up moving back to XMPP. Or maybe something new. Overall I don't think there are major fundamental differences. I think Matrix making graph sync the core primitive to build off of was a good idea, but in practice I don't think it matters much.

You say that XMPP is much lighter. But I think that is mostly due to Synapse not being very efficient. Other implementations are fairly light. Even then my Synapse is using fairly small amounts of resources. You should also check that you are making an apples-to-apples comparison with large rooms, media and message history like you would typically see in a common Matrix server.