Bloom, Jones, Van Reenen & Webb (2020) found research productivity is falling sharply — reproducing Moore's Law now takes more than 18× the researchers of the early 1970s

"Are Ideas Getting Harder to Find?" (Nicholas Bloom, Charles I. Jones, John Van Reenen & Michael Webb, American Economic Review 110:4, 2020) decomposes economic growth into research productivity (ideas produced per researcher) times the number of researchers, and reports that across every case they examine — from aggregate US growth to firm-level data to specific technologies — research productivity is "declining sharply." Exponential progress is being sustained only by throwing exponentially more inputs at it.

The paper's most-quoted instance is Moore's Law: "the number of researchers required today to achieve the famous doubling of computer chip density is more than 18 times larger than the number required in the early 1970s." Idea output per researcher in semiconductors has fallen by a factor of roughly 18 in that window; constant progress required a corresponding rise in the research workforce.

This is the brake on the engine described in claim-kremer-1993-technology-growth-proportional-to-population: Kremer's "more people → more ideas" runs forward into diminishing returns, where more minds buy proportionally fewer breakthroughs. Critically, the poster child for the decline — Moore's Law — is the very substrate on which the AI cost collapse of claim-inference-cost-collapsed-280x rests, which is why that 280× figure is the weakest instance of the scale-drives-improvement bridge: it lives in the exact domain where economics documents the idea-engine sputtering. The same sub-linear-returns law appears in biology (claim-desai-fisher-murray-2007-clonal-interference-logarithmic-speed-of-evolution) and, qualitatively, in neural scaling — the pattern is drawn together in observation-population-scaled-improvement-hits-a-sublinear-brake-across-domains.