The Economic Underpinnings and Future of Moore's Law

We often talk about Moore's Law as a law of physics, but this paper from Romain Duval at the IMF is a stark reminder that it has always been, at its heart, an economic law. The exponential growth in computing power wasn't just about shrinking transistors; it was about the economics of scale where each successive node was cheaper to produce per transistor. Now, as capital expenditures for new semiconductor fabs skyrocket into the tens of billions of dollars, that economic cycle is breaking down. The massive upfront costs mean that only a handful of mega-corporations can afford to play the game, leading to consolidation and a slower pace of actual scaling.

This economic shift has huge implications for startups and global productivity. Historically, we relied on hardware-driven productivity gains to lift all boats, allowing even inefficient software to perform well over time. If slower scaling drags down global productivity growth, we can no longer expect a cheap, automatic performance upgrade every two years. We must treat compute as a scarce, expensive asset rather than an infinite utility. In our startup, this means changing how we model our margin and unit economics, because raw compute costs are going to become a much larger slice of our long-term operating budget.

What stuck with me

• Capital expenditure explosion: The soaring costs of building advanced fabrication plants have transformed semiconductor scaling into an exclusive game for a few multi-billion-dollar entities.
• Productivity growth drag: Slower physical transistor scaling directly correlates with a deceleration in global economic productivity, removing the historical tailwinds software developers enjoyed.
• Compute scarcity economics: Startups can no longer assume that hardware costs will continuously trend toward zero, making resource efficiency a core competitive differentiator.