People invest in services like Twitter knowing that there is a chance that they might not be monetizable, but believing that if they turn out to be, then the payoff will be enormous. So the expectation value of the return is still high.
Let's say I give you a box and tell you that inside is a social media company. You know that the achievable revenue per user could be anywhere from, what, a penny a month to a dollar a month -- two orders of magnitude of variability. You know that the achievable audience could be anywhere from, say, 10 thousand users to 1 billion users -- five orders of magnitude.
So if you want to determine how much you would pay for the contents of the box, you should try to learn about the potential audience rather than the potential revenue per user, because the former completely dominates the latter in the expected value computation.
Put another way -- how much would you pay to flip a coin where heads you get a billion dollars, and tails you get nothing? If you're allowed to purchase more than one coin flip, and the coin flips aren't too correlated, and you have access to efficient capital markets, probably close to half a billion dollars, am I right? :)