AI Labs Face Three Distinct Economic Problems

Frontier AI labs are not competing on the same playing field. Each has a unique identity that creates its own structural bottleneck.

OpenAI serves over 900 million monthly active users, mostly consumers. This produces a severe usage-mix problem: high volume of low-value prompts drives enormous inference costs with low revenue per token.

Anthropic serves 300K+ businesses with strong enterprise adoption. Excellent revenue quality, but this creates a GPU-supply problem: demand consistently exceeds capacity, forcing constant rate-limiting.

xAI possesses massive GPU supply through Colossus but virtually zero natural demand. The challenge is utilization.

Meta, Google, Microsoft, and Amazon operate as full-stack or cloud giants with entirely different incentives and moats.

*The Core Economics*

Training is a one-time sunk cost.

Inference is the factory running 24/7.

The decisive metrics are:

•⁠ ⁠Tokens per watt-year (supply-side efficiency)

•⁠ ⁠Revenue per token (demand quality and utilization)

•⁠ ⁠Revenue per watt-year (the number that determines whether a lab can self-fund the next cluster)

No lab has crossed the self-funding threshold yet.

An article that expands how founders, AI builders, investors, and systems thinkers understand the real economics and strategic positioning of the major AI players.