The way computer (and especially internet) companies work is that once you have a working product, you can sell it a practically infinite number of times for practically zero marginal cost.
This means that the successful software/internet companies are INCREDIBLY profitable, since they can sell the same product millions of times.
Are startup investors good at predicting which company is going to be successful? Oh god no, they are awful at it. 90% of the companies they invest in fail completely.
It doesn't matter, though, because they only need one success to make all those failures worth it. The cost of investing in a failure is WAY lower than the cost for NOT investing in a success.
This means they don't really care that they can't evaluate which companies will be successful, they just cast a wide enough net to not miss the successes.