Computation/AI capability has increased 20 quadrillion-fold over 80 years.

Judge as of March 2024. The claim is a rough figure: an increase of about 2×10^16 over about 80 years (roughly 1939–1944 to 2023–2024) in the measure Kurzweil is citing, most likely the price-performance of computation (computations per second per constant dollar). True: credible datasets (Kurzweil's own updated chart, Nordhaus, Moravec or equivalent compilations) show an increase in that measure over that period of roughly 5×10^15 to 8×10^16. Mostly true: the increase is within about one order of magnitude of 2×10^16 (10^15 to 10^17 outside the true band), or the span is somewhat different from 80 years. Mixed: the figure fits only a different measure than the one the claim implies, or credible datasets disagree enough that some support it and others are 1–2 orders of magnitude away. Mostly false: the figure differs by about 2 orders of magnitude, or a price-performance figure is presented as raw 'AI capability' in a way that distorts it. False: credible data put the increase off by 3 or more orders of magnitude.

Set before the research began.

1. E1 AI escape velocity: A conversation with Ray Kurzweil Bessemer Venture Partners · 11 March 2024 Own record

   “So this graph represents a 20 quadrillion fold increase for the amount of computation you can get for the same amount of money.”

2. E2 AI escape velocity: A conversation with Ray Kurzweil Bessemer Venture Partners · 11 March 2024 Own record

   “We took the leading computer of that year—the latest one was actually a Google computer—and it's 130 billion calculations per dollar. So that's a 20 quadrillion fold increase.”

3. E3 The Progress of Computing Cowles Foundation for Research in Economics, Yale University · 15 September 2001

   “Performance in constant dollars or in terms of labor units has improved since 1900 by a factor in the order of 1 trillion to 5 trillion, which represent compound growth rates of over 30 percent per year for a century.”

4. E4 The Progress of Computing Cowles Foundation for Research in Economics, Yale University · 15 September 2001

   “Around World War II, however, there was a substantial acceleration in productivity, and the growth in computer power from 1940 to 2001 has averaged 55 percent per year.”

5. E5 AI escape velocity: A conversation with Ray Kurzweil Bessemer Venture Partners · 11 March 2024 Own record

   “So this graph represents a 20 quadrillion fold increase for the amount of computation you can get for the same amount of money.”

6. E6 AI escape velocity: A conversation with Ray Kurzweil Bessemer Venture Partners · 11 March 2024 Own record

   “It did a small fraction of a calculation. 0.000007 calculations per second per constant dollar.”

7. E7 AI escape velocity: A conversation with Ray Kurzweil Bessemer Venture Partners · 11 March 2024 Own record

   “Regardless of what we were increasing—relay speeds or vacuum tube speeds or integrated circuits—it increased by the same amount every year for 80 years.”

8. E8 Cost per GFLOPS exponentialtracker.com

   “the price of a billion floating-point operations fell from over a trillion dollars to about a cent.”

9. E9 Cost per GFLOPS exponentialtracker.com

   “the price of a gigaflop halved about every 20 months from ENIAC through 2011, but has taken about 45 months per halving over the past decade.”

10. E10 Cost per GFLOPS exponentialtracker.com

    “| 1945 | $1,265,000,000,000/GFLOPS | ENIAC |”

11. E11 Cost per GFLOPS exponentialtracker.com

    “| 2025 | $0.0123/GFLOPS | RX 9070 XT |”

Kurzweil's own endpoints (E2, E6), about 0.000007 and 130 billion calculations per second per constant dollar, give roughly 1.9×10^16, so the figure matches his own chart. The independent exponentialtracker series (E10, E11) runs from $1.265 trillion per GFLOPS for ENIAC in 1945 to $0.0123 per GFLOPS in 2025, an improvement of about 10^14 over 80 years. That is roughly two orders of magnitude below the claim. The Cowles Foundation paper (E4) reports about 55% a year growth from 1940 to 2001, about 4×10^11 over that span, and E9 reports a slowdown since 2011, so neither supports the faster growth needed to reach 2×10^16. Because the independent data put the increase about two orders of magnitude lower, this is resolved mostly false, though a different starting machine could narrow the gap somewhat.