The Limit Does Not Exist

AI is entering a new scaling era in which repeated improvements in computing infrastructure could drive a fourth industrial revolution rather than a conventional S-curve slowdown.

• Goldman Sachs Research found that 68% of S&P 500 companies mentioned artificial intelligence on their calls in the second quarter of 2026, while AI capital expenditure exceeded US$1 trillion annually.
• The authors argue that AI development can improve by 10x repeatedly over several years, distinguishing it from technologies whose performance quickly reaches a plateau.
• Moore’s Law delivered roughly 39 doublings of compute power over 60 years, but the authors say semiconductor and system-level advances can continue even as transistor scaling slows.
• The chessboard analogy shows why compounding improvement becomes economically explosive: the final 32 squares contain 99.9999999% of the rice, or 18.4 quintillion grains in total.
• Historical examples such as the light bulb, jet engine and iPhone suggest that technological improvement usually follows an S-curve, but the authors identify semiconductors as an exceptional technology that avoided asymptoting for more than six decades.

Takeaway: Investors should focus not only on current AI adoption but also on the infrastructure and successive scaling curves that could sustain unusually rapid improvement, while recognizing that the duration of this acceleration remains uncertain.