The 10^3-per-decade growth rate of computational capacity is conservative because the rate of exponential growth is itself growing exponentially.

The claim: in 2002, a growth rate in computing capacity of about 1,000x per decade was conservative, because the exponential growth rate itself had been rising (double exponential growth). Judge it against historical data on price-performance of computing (e.g. calculations per second per constant dollar from about 1900 to 2002, as in Moravec, Nordhaus, or later analyses) as of 2002. True: the data show growth of at least about 1,000x per decade around 2002, and a growth rate that rose measurably over the historical record. Mostly true: the trend is right, but the per-decade figure or the acceleration is only roughly supported (e.g. the growth rate rose in some periods but not steadily). Mixed: the 1,000x-per-decade figure is supported, but evidence of an accelerating growth rate is weak or disputed (or the reverse). Mostly false: growth was exponential, but the rate was broadly constant, or contested analyses do not support acceleration, so the claim overstates what is known. False: the data show growth well below 1,000x per decade, or a slowing growth rate.

Set before the research began.

1. E1 the Kurzweil Library writingsbyraykurzweil.com (Ray Kurzweil essay, "the Law of Accelerating Returns") · January 2001 Own record

   “Indeed, we find not just simple exponential growth, but “double” exponential growth, meaning that the rate of exponential growth is itself growing exponentially.”

2. E2 the Kurzweil Library writingsbyraykurzweil.com (Ray Kurzweil essay, "the Law of Accelerating Returns") · January 2001 Own record

   “We doubled computer power every three years early in the twentieth century, every two years in the middle of the century, and close to every one year during the 1990s.”

3. E3 January 2018 – Soft Machines, by Richard Jones Soft Machines (blog by Richard Jones, soft matter physicist) · 14 January 2018

   “It was around 1986 that the rate of change really took off – between 1986 and 2003 computer power increased at an astonishing 52% a year, a doubling time of just a year and a half.”

4. E4 January 2018 – Soft Machines, by Richard Jones Soft Machines (blog by Richard Jones, soft matter physicist) · 14 January 2018

   “But after 2011, according to Hennessy and Patterson, the growth rate slowed further – down to 3.5% a year since 2015.”

E1 and E2 are Kurzweil's own essay. They state that growth is double exponential and that doubling times shrank across the twentieth century, but they are the subject restating his own claim. The only independent source on the period is E3, a 2018 blog post, which reports 52% annual growth in computer power from 1986 to 2003 [E3]. That works out to roughly 65x per decade, and it measures performance, not the price-performance per constant dollar the criteria require. E3 does suggest the growth rate picked up around 1986. E4 covers the period after 2011, so it is later knowledge and does not bear on the claim as of 2002. Without independent price-performance data, the evidence cannot settle either the 1,000x-per-decade figure or the claim that the growth rate was accelerating.