Raso: AI bubble talk is an echo chamber, Anthropic math says otherwise

Raso's argument rests on a specific reading of where AI companies' money is going: he contends the unprecedented spending is aimed at building the data centers needed to run AI, and that once construction is finished, outlays should fall sharply. That reframing matters because the sustainability critique usually targets the scale of infrastructure spending itself, and Raso answers it by pointing to Anthropic's per-dollar computing economics.

The bubble skeptics' other common concern — that open-source models will catch up and turn frontier AI into a commodity — gets a fast-food analogy from Raso. He notes that burger and sandwich chains have no lasting moat over rivals yet keep earning profits, and argues ordinary users are unlikely to run open-source models locally or switch models often enough for commoditization to bite.

Raso also draws a line between past bubbles and today's AI firms: in a classic bubble, valuations are visibly overheated, whereas he sees Anthropic and OpenAI at sound levels, with most AI developers still private and therefore less exposed to a funding freeze. Even so, he concedes demand could fall on data-center backlash, regulation, or misuse, and says he wrote the skeptical post because online AI-bubble discussion had become an echo chamber. Commenters on Hacker News pushed back that data centers are not warehouses that sit idle after construction, but demand ongoing large-scale capital spending and operating costs.