Cloud computing is expanding exponentially, doubling in speed every year for the same money.
Mixed Partly true and partly false in roughly equal measure.
Factual: Statements about the present or the past, checked against the record: did they hold up?
What would count as true or false
As of mid-2024, the test is whether the speed per dollar of cloud computing (price-performance, such as FLOPS per dollar for rented compute) was growing exponentially and doubling about every year. True: credible data show cloud or compute price-performance doubling roughly every 9–15 months over recent years. Mostly true: exponential growth with a doubling time of about 15–20 months. Mixed: exponential growth with a doubling time of about 20–30 months, or the data clearly support yearly doubling for one measure (for example, AI accelerator FLOPS per dollar) but not for general cloud compute. Mostly false: improvement is real but much slower (doubling time over about 2.5 years) or not clearly exponential. False: cloud price-performance was flat or not improving in a sustained way.
Set before the research began.
Evidence (4)
-
E1 Trends in GPU price-performance Epoch AI · 27 June 2022
“we find that the amount of floating-point operations/second per $ (hereafter FLOP/s per $) doubles every ~2.5 years.”
-
E2 Trends in GPU price-performance Epoch AI · 27 June 2022
“We find that FLOP/s per dollar for ML GPUs double every 2.07 years (95% CI: 1.54 to 3.13 years) compared to 2.46 years for all GPUs.”
-
E3 Trends in GPU price-performance Epoch AI · 27 June 2022
“For this subset of models, we find a doubling time of 2.95 years (95% CI: 2.54 to 3.52 years), which is statistically significantly longer than the typical doubling time.”
-
E4 Trends in GPU price-performance Epoch AI · 27 June 2022
“We, therefore, think that a doubling time of 2 years is too aggressive and ~2.5 years accurately describes the doubling time of price-performance for GPUs over the past 15 years.”
Reasoning
Epoch AI's 2022 analysis of GPU price-performance, an independent measure of compute price-performance, finds that FLOP/s per dollar doubles about every 2.5 years. The figure is 2.46 years for all GPUs and 2.07 years for ML GPUs (E1, E2). Epoch concludes that even a two-year doubling time is too aggressive (E4), and for one subset of models it finds a longer doubling time of about 2.95 years (E3). The growth is exponential, but the main estimates fall in the 20–30-month band rather than the yearly doubling described, so the claim resolves as mixed. Because the headline figure sits at the boundary with mostly false, this is a close call, and none of the evidence measures cloud rental prices directly or covers the period up to mid-2024.
Cite this claim
Ray Kurzweil, “Cloud computing is expanding exponentially, doubling in speed every year for the same money.” (said 9 July 2024). Verdict: mixed. Reliability API, claim c_b4c1eefc6d45, method 0.1.0+28da144b, dataset release v1.1.2. https://epistacklabs.com/claims/c_b4c1eefc6d45
This claim is in dataset release v1.1.2, with the same verdict.
Reply to this claim
More factual claims by Ray Kurzweil
-
Mixed
“But I got to 10 billion. It started to do things.”
Kurzweil said that at about 10 billion connections neural networks started to do things.
-
Mostly false
“a 20 quadrillion–fold increase over 80 years.”
Computation/AI capability has increased 20 quadrillion-fold over 80 years.
-
Resolved true
“It’s remarkable how long it’s lasted, but it wasn’t the first, but the fifth paradigm to provide exponential growth to computing.”
-
Mostly true
“A computer can also remember billions or even trillions of facts perfectly, while we are hard pressed to remember a handful of phone numbers.”
-
Mostly false
“That was not true a year ago. But that’s true today.”
Verdicts are made automatically; how verdicts are made.