The long tail hypothesis went as follows. Plot a curve with products in descending order of popularity on the x axis and units sold (or other equivalent metrics

The long tail hypothesis went as follows. Plot a curve with products in descending order of popularity on the x axis and units sold (or other equivalent metrics such as downloads or views) on the y axis. The curve will approximate an exponentially decreasing function. As the cost of keeping inventory decreases, the curve will become more shallow. I.e the fraction of the curve's total area under the first x% will decrease. This hypothesis was utterly wrong and for most products the curve has become peakier (sharper) not shallower. I think people who comment haven't read Anderson's book - in hindsight it is laughable how wrong he was.