"This worst‐case simulation came up with 2.2 million deaths by simply assuming that
81% of the population gets infected –268 million people– and that 0.9% of them die. It did
not assume health systems would have to be overwhelmed to result in so many deaths, though it did make that prediction."
Historically, viruses peter out before they reach 30% of the population.
"The key premise of 81% of the population being infected should have raised more alarms than it did. Even the deadly “
Spanish Flu” (H1N1) pandemic of 1918–19 infected no more than
28% of the U.S. population. The next H1N1 “Swine Flu” pandemic in 2009-10, infected 20-
24% of Americans.
To push the percentage infected up from 20–28% to an unprecedented 81% for COVID-19 required assuming the number of cases and/or deaths keeps doubling every three or four days for months (deaths were predicted to peak July 20). And that means assuming the estimated reproduction number (R0) of 2.4 remains high, and people keep mingling with different groups, until nearly everyone gets infected. Long before 8 out of 10 people became infected, however, a larger and larger percentage of the population would have recovered from the disease and become immune, so a smaller and smaller share would still remain susceptible."
https://www.cato.org/blog/how-one-model-simulated-22-million-us-deaths-covid-19