The AI demand bubble: separating real spend from engineered hype

A sharp critique of the AI demand narrative argues the usage numbers are driven by free credits, enterprise pilots, and circular investment - not organic, paying demand.

In plain terms: A widely-read essay argues that AI "demand" is largely manufactured: free-tier usage, corporate mandates to look AI-forward, and infrastructure spending that creates its own demand. The question is whether any of it converts to durable revenue.

The story: The core argument: the AI industry has confused capital deployment with demand. Billions in infrastructure spending generates compute capacity, which generates usage (often free or subsidized), which generates usage metrics, which justifies more investment. The loop is real; the revenue endpoint is not.

The evidence cited includes: ChatGPT's DAU figures relying heavily on free-tier users, enterprise AI pilots that don't survive budget reviews, and the absence of publicly reported AI-driven revenue growth from the non-hyperscaler companies that were first to adopt.

This is not a fringe view. It echoes Palantir's counter-signal: AIP showed real Q2 revenue (+93% growth), but AIP works because Palantir integrates AI into proprietary data workflows - a fundamentally different proposition than selling API access.

Under the hood: The distinction that matters is not "AI vs. non-AI" revenue, but "AI as infrastructure for a differentiated product" vs. "AI as the product." The former has demonstrated unit economics; the latter mostly doesn't yet.

So what: For anyone betting on AI revenue timelines: watch for enterprises reporting AI-attributed cost savings in earnings calls. That's the leading indicator of real demand - not API call volumes or DAU announcements.