I have used __databricks_internal catalog to find the underlying location of the materialized views. It turns out they are the ones which weighing so much. I was suspecting it can be them, but thank you for tipping me @Isi.

I did some more digging and the silver_raw table which is a streaming dlt table weights 30GB, silver publish materialized view, which is based on that silver_raw table, weights almost 1TB. I guess Databricks is smart enough not to store the same data twice, or I am missing something ?