We could, but the point is that we want to be constantly polling, no matter if there's a listener or not. The main thing that the article doesn't say is that in their original example, there is only one Observable (like a Subject - which would also stay in memory as long as the app runs) whereas the solution with an injector creates a new Observable/Signal for each consumer/component. As a result, there's no real caching at play and the point of using a Signal, in that case, is lower (since there will be one signal for each subscriber - instead of one shared resource for all of them).

Also, since there's only one request, I don't get the "memory leak" argument, since there won't be creation of multiple Observables.