That's mostly a function of blocks stored on a DataNode. For example, a rule of thumb is one GB heap size for DN for every one million blocks stored on that DN.
Can you provide more information on reporting load (for low-latency operations) issue when we have datanode with 100T+ storage? We need archive node for HDFS storage only purpose. No Yarn/spark running on it. It will only storage data based on storage migration policy. Node's network/storage IO bandwidth is considered be able to handle the larger storage size.