Step by step explanation:

params Delegate?[] delegates - It is an array of nullable Delegate

params Delegate?[]? delegates - The entire array can be nullable

Since each parameter is of the type Delegate? and you return an index of the Delegate?[]? array, then it makes sense that the return type is Delegate? otherwise the compiler would return an error as if you were returing and int from a method that returns a string.

You could change for instance your code to return a Delegate type like this:

public static Delegate Combine(params Delegate?[]? delegates)
{
    Delegate defaulDelegate = // someDelegate here
    if (delegates == null || delegates.Length == 0)
        return defaulDelegate;

    Delegate d = delegates[0] ?? defaulDelegate;
    for (int i = 1; i < delegates.Length; i++)
        d = Combine(d, delegates[i]);

    return d;
}

Nullable Reference Types are new in C# 8.0, they do not exist before.

It's a matter of documentation, and how warnings at compile-time are produced.

The exception "object not set to an instance of an object" exception is quiet common. But this is a runtime exception, it can partially discovered at compile time already.

For a regulate Delegate d you can always call

 d.Invoke();

meaning, you can code it, at compile time nothing will happen. It may raise exceptions at runtime.

While for a new Delegate? p this Code

 p.Invoke();

will produce a compiler warning. CS8602: Dereference of a possibly null reference unless you write:

 p?.Invoke();

what means, call only if not null.

So you document a variable may contain null or not. It raises warnings earlier and it can avoid multiple tests for null. The same what you have for int and int?. You know for sure, one is not null - and you know how to convert one to the other.

In C# 8 one should explicitly mark reference types as nullable.

By default, those types are not able to contain null, kinda similar to value types. While this does not change how things work under the hood, the type checker will require you to do this manually.

Given code is refactored to work with C# 8, but it does not benefit from this new feature.

public static Delegate? Combine(params Delegate?[]? delegates)
{
    // ...[]? delegates - is not null-safe, so check for null and emptiness
    if (delegates == null || delegates.Length == 0)
        return null;

    // Delegate? d - is not null-safe too
    Delegate? d = delegates[0];
    for (int i = 1; i < delegates.Length; i++)
        d = Combine(d, delegates[i]);

    return d;
}

Here is an example of an updated code (not working, just an idea) leveraging this feature. It saved us from a null-check and simplified this method a bit.

public static Delegate? Combine(params Delegate[] delegates)
{
    // `...[] delegates` - is null-safe, so just check if array is empty
    if (delegates.Length == 0) return null;

    // `d` - is null-safe too, since we know for sure `delegates` is both not null and not empty
    Delegate d = delegates[0];

    for (int i = 1; i < delegates.Length; i++)
        // then here is a problem if `Combine` returns nullable
        // probably, we can add some null-checks here OR mark `d` as nullable
        d = Combine(d, delegates[i]);

    return d;
}