Softbank aren't really doing this any more though. They are making very large bets on relatively few companies. Their vision fund 1 was something like 100 billion spread between about 100 companies, so 1 billion each. This is rather different than the usual VC. It's hard to get data, but most data suggest smaller rounds have better success [1].
Due to the power law distribution of startup outcomes, the "expected value" of the "average" (mean) startup is infinity (assuming there are infinite startups). This means YC's model to invest a modest amount in as many startups as possible is the rational strategy. Even the tiniest percentage of equity in infinite companies yields infinite returns.
This seems as counter-intuitive as saying "superconductors have zero resistance". It makes people think you meant to say "nearly zero resistance" or "approaches zero resistance".
But both cases are true - the power law distribution really does tell us that the expected return is actually infinite and superconductors really do have zero resistance (up to the critical current limit).
Is a small number of people with huge amounts of money investing in things nobody wants better than lots of people with small amounts of money investing in things they want?
Downside is capped, upside is uncapped, why not bet on as many companies as you can?