They're pointing out that run-rate revenue is based on essentially sampling revenue over some limited time interval, then extrapolating from there assuming revenue always occurs at the same rate (or greater) over all similar intervals in the future. More specifically, they're pointing out that estimates of ARR derived from this kind of sampling are fundamentally prone to error and can be arbitrarily inflated based on how the time interval is sampled.

As far as I understand run rate revenue is just a fancy way of saying that "the last month we had sales, and if that continues for a year we will have a AAR of 30B. meaning it's not 30B yet, but the sales numbers indicates that we get there by continue selling at the current speed. But to have revenue of $100 and get $30B in ARR I guess the period looked at needs to be seconds....

(Run Rate = Revenue in Period / # of Days in Period x 365)

There are about 30 million seconds in a year. If they made $100 over the last hundred milliseconds, then that’s $30B annualized.

(That said, their numbers are much realer than that.)

If you make a hundred dollars in 0.1 seconds, you could say your annualized revenue is $100 / 0.1 * 60 * 60 * 24 * 365 = -$30 billion.

That said, most people would use a monthly or quarterly period to estimate ARR. I'm not sure what Anthropic used. Probably monthly.