The 100-Million-Fold Problem: Why “Jevons’ Paradox” Won’t Save the AI Chip Trade
Token prices are compounding downward at 80% per cycle. One analyst’s argument: the standard rebuttal doesn’t hold the math.
⬛ The Clip
“But I think this mostly speaks to the innumeracy of people, that they don’t understand what a compounding price decline of 80% means… You have to see around a hundred million fold growth over the next six years in terms of tokens. Is it possible? Absolutely it’s possible. Is it likely? No, it’s not likely, but it could happen… to say, but Jevons’ paradox, but people will use more, is to really dodge the core problem of the geometric decline in the price… because of the convergence of models.”
Clip via @paulkedrosky with @kantrowitz / Big Technology Podcast — AI Chips Are Losing Value Faster Than Almost Anything Else
The Sharp Point
The argument isn’t that AI demand won’t grow. It’s a math problem. An 80% compounding price decline means you need roughly 100 million times more token consumption over six years just to keep revenue flat. That’s not a bear case. That’s arithmetic.
Jevons’ paradox — the idea that cheaper resources get consumed more, offsetting the price drop — is a real economic phenomenon. But Kedrosky’s argument, as expressed in the clip, is that invoking it here is a dodge. The paradox doesn’t negate a geometric curve. It just slows it down at the margin.
The deeper driver, in his framing: model convergence. As frontier models become more similar in capability, the price compression doesn’t stop — it accelerates. Differentiation erodes. Commoditization follows.
“To say ‘but Jevons’ paradox, but people will use more’ is to really dodge the core problem of the geometric decline in the price.”
— @paulkedrosky, as expressed on the Big Technology Podcast
📐 The Structural Read · FourWeekMBA Analysis
On FourWeekMBA’s Map of AI, chips and inference infrastructure sit at the enablement layer — the substrate everything else runs on. That layer is historically where commoditization hits hardest and fastest, precisely because it’s upstream from differentiation.
The Harness Theory implication is underappreciated: if the chip and inference layer is deflating geometrically, the value doesn’t disappear — it migrates. It moves to whoever is building on top of cheap inference, not whoever is selling it.
This is what Kedrosky’s argument points toward, even if he doesn’t frame it that way. The stack doesn’t shrink. The margin pool relocates.
Why This Framing Matters
Most AI infrastructure bulls are running a demand story. Kedrosky’s cut is a supply-side math story. Those are different arguments requiring different evidence to refute. Conflating them — which the Jevons rebuttal tends to do — is where the analytical slippage happens.
The Bottom Line
Kedrosky’s argument — as clipped — isn’t a prediction that AI fails. It’s a precision instrument aimed at a specific analytical error: mistaking a rebound narrative for a mathematical rebuttal. A hundred-million-fold demand growth over six years is possible. Whether your thesis requires it to be likely is the question worth sitting with.
This post is FourWeekMBA’s editorial analysis of a published podcast clip. Views expressed in the quote are those of the speaker as stated on that episode. This is not investment advice.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
