How can I avoid repeating code initializing a hashmap of hashmap?

This is an excellent use-case for Map#computeIfAbsent. Your snippet is essentially equivalent to:

allInvoicesAllClients.computeIfAbsent(id, key -> new HashMap<>()).put(date, invoice);

If id isn't present as a key in allInvoicesAllClients, then it'll create mapping from id to a new HashMap and return the new HashMap. If id is present as a key, then it'll return the existing HashMap.

computeIfAbsent is a great solution for this particular case. In general, I'd like to note the following, since nobody mentioned it yet:

The "outer" hashmap just stores a reference to the "inner" hashmap, so you can just reorder the operations to avoid the code duplication:

HashMap<LocalDateTime, Invoice> allInvoices = allInvoicesAllClients.get(id);

if (allInvoices == null) {           
    allInvoices = new HashMap<>();
    allInvoicesAllClients.put(id, allInvoices);
}

allInvoices.put(date, invoice);      // <--- no longer repeated