AI Doesn't Have ROI

TEXT START: If you liked this piece, you should subscribe to my premium newsletter. It's $70 a year, or $7 a month, and in return you get a weekly newsletter that's usually anywhere from 5,000 to 18,000 words, including vast, detailed analyses of NVIDIA, Anthropic and OpenAI's finances, and the AI bubble writ large.

THE DISSECTION

This is a long-form financial-skeptic polemic targeting the AI investment thesis. The author builds an evidentiary case that: (1) AI costs are deliberately obfuscated through subsidized subscriptions and vague token billing, (2) true inference costs are rising rather than falling despite industry claims, (3) the "dot-com redemption arc" analogy fails because AI infrastructure is domain-specific and useless for other purposes, and (4) the entire AI investment thesis rests on deception about economics. It's well-sourced, angry, and structurally coherent within its own frame.

THE CORE FALLACY

The author commits the investor-centric fallacy. He treats "no ROI" and "bubble collapse" as the terminal threat. The DT framework locates the existential threat elsewhere entirely: AI doesn't need ROI to cause terminal economic damage. The author argues AI will deflate because it can't justify capital allocation. The DT thesis says: AI can destroy the wage-consumption circuit regardless of whether it generates investor returns. A trillion-dollar capital write-off and mass structural unemployment can happen simultaneously. The author mistakes the financial bubble for the disease; the disease is productive displacement running independent of balance sheets.

HIDDEN ASSUMPTIONS

1. Market price signals still govern deployment. The analysis assumes bubble pop = correction = recovery. But AI adoption isn't purely a market decision. State actors, sovereign wealth funds, and strategic national interests will absorb AI infrastructure regardless of commercial ROI.
2. ROI is the binding constraint on AI deployment. The author frames this as "AI fails because it has no ROI." DT says: AI deployment is constrained by physical infrastructure and compute access, not by whether CFOs can justify it on spreadsheets.
3. The dot-com analogy fails in the direction of understatement. The author argues AI is worse than dot-com because infrastructure won't be repurposable. But he's measuring the wrong variable. The dot-com bubble didn't leave behind useful infrastructure for humans to use—it left infrastructure that made human labor more productive. AI infrastructure left behind will accelerate AI labor replacement. This isn't a consolation prize.
4. Bubble pop = harm contained to capital allocators. The author focuses entirely on venture capital, corporate AI budgets, and startup ecosystems. He never addresses what happens to the 60-70% of workers whose employment depends on the consumption circuit when it breaks.

SOCIAL FUNCTION

This is financial-skeptic copium for investors and analysts who want to believe the AI story will end cleanly. It's the Reagan-era realist telling tech executives the party is over while assuming the party is the only thing at stake. It performs rigorous skepticism about capital allocation while leaving the structural unemployment mechanism entirely unexamined. The author correctly identifies that AI economics are broken—but treats this as a story about bad investments rather than a story about capital destroying the labor market that generates the demand for everything else.

THE VERDICT

Mechanically accurate on cost structures and financial dynamics, but analytically blind to the actual threat vector. The author is documenting a $1 trillion capital misallocation while the real misallocation—automating human cognitive labor before any social infrastructure exists to absorb the displaced—proceeds uninterrupted. His bubble will pop. The displacement won't stop when it does. This analysis is the financial version of noting that the house is on fire by counting the mortgage payments being missed. Technically correct. Structurally irrelevant to the burning.