> In 1892, the New York Herald Tribune compiled a list of all the millionaires in America. They found 4047 of them. How many had inherited their wealth then? Only about 20%... So it's not 2020 that's the anomaly here, but 1982.
This is an oversimplification. Go back 100 more years to 1782 and I bet you'd find the dominant paradigm is inherited wealth. Go back another 100 years and it's still inherited wealth. I'd assume that below a certain tribal size you stop seeing much inherited wealth, and so the starting point for groups of humans is more like a meritocracy, where the strongest tribe member gets to keep the most stuff, but "inherited wealth dominates" is probably the default mode for most of what we'd recognize as "civilization".
More generally it's extremely dubious to attempt to extrapolate a historical trend from the datapoints 1892,1992, 2020, especially given that your first two datapoints are different; if you saw a consistent trend going backwards in time you'd be more justified in extrapolating from that.
A more nuanced model would be to think in terms of cycles; this is a very well-established concept in both history and economics. Technological paradigm shifts allow innovators to overturn the status quo. Over time power accumulates, monopolies form, big companies figure out how to exploit the new technology, the winners of the cycle entrench, and opportunities for upwards mobility decreases again. The industrial age is one such paradigm shift. The "information age" is another. Inside the latter, you can make the case for smaller cycles of innovation that still have large economic consequences, like mainframe computing, desktop computing, and cloud computing. Each of these paradigm shifts minted new titans of industry.
When a new paradigm arises, the cycle repeats again. We see this with the railroad barons and other industrial innovators, the progenitors of a generation of inherited wealth. We'll probably see this again with the Zuckerberg lineage. It seems entirely possible to me that in 20 years' time, there are essentially no new tech multi-billionaires, and instead the big-5 tech companies just buy any startup that grows big enough to be a potential competitor. This would mint plenty of tech millionaires, but leave the "100 richest" unchanged. Of course, anti-trust laws exist, and so this outcome is far from a certainty. But it would at least be consistent with how previous paradigm shifts have played out.
A potential counter-argument to this line of reasoning would be to look at the cycle period and argue that the troughs of ossification/consolidation between cycles are shortening, and we should expect the existing dominant companies to get unseated by whoever comes up with the next paradigm shift. This is a very interesting subject for discussion; I'd just note that it is a much more nuanced argument than Graham's point, which is essentially claiming historical homogeneity prior to 1892 and using that to imply that things will stay the same as they are now going forwards.