Why I'm Short the AI Buildout Even Though I'm Long Computers

The argument separates the usefulness of AI from the financial case for its infrastructure boom: large language models are valuable tools, especially for coding and knowledge retrieval, but the scale of data-center investment has outrun plausible returns. Big tech companies and specialist operators are increasingly borrowing to buy GPUs that lose value within years, while rising interest rates raise financing costs. CoreWeave illustrates the strain: despite sold-out capacity and a large contract backlog, its interest expenses far exceed operating income, and its bonds trade at distressed levels. The author argues that chipmakers may look healthy for now because customers pay for GPUs upfront, but orders will eventually slow if financing and returns no longer pencil out.

The expected demand also depends on AI labs sustaining extraordinary revenue growth. Anthropic’s quarterly sales rose sharply, but its monthly run rate barely advanced in the final two months of Q3. Some AI spending comes from venture-funded startups rather than durable customer savings, while cheaper open-weight models threaten US labs’ pricing power. Lower prices will increase usage, the author says, but not enough to offset steep price declines; added computation also brings diminishing gains. AI productivity so far is concentrated in tech, where coding is central and results can be tested, while other industries show little financial improvement. The buildout reflects both belief in imminent superintelligence and a corporate arms race, even though the technology’s demonstrated value remains narrower than the investment assumes.