AI's price war could be a win for hyperscalers, Citadel says

What happened

Citadel Securities says falling AI token prices and the rise of open-weight models are reshaping AI economics. It reports token prices fell about 50% in six months, frontier-model prices dropped 62%, and token volumes rose roughly fivefold, with hyperscalers seen as durable beneficiaries.

Confirmed

Global impact / market context

Cheaper tokens mean each unit of AI work costs less, which can lift usage without lifting spending. If open-weight models are good enough for more tasks, frontier developers may earn less profit per sale, while companies selling scarce computing infrastructure capture more of the value.

Analyst inference

Ramp data cited by Citadel shows little growth in overall enterprise AI spending in August and September despite rising usage, suggesting lower prices offset higher volumes. Demand for computing remains strong, with rental rates for Nvidia's B200 graphics processors noted as robust.

Analyst inference

What to watch

1. Watch whether token volumes keep rising roughly fivefold and whether consumption-weighted token prices continue falling about 50% over six months, as Citadel reported. Confirmed
2. Citadel's view implies investors should watch hyperscalers' infrastructure revenue as inference demand grows, since those suppliers may benefit most from cheaper AI usage. Proposed
3. Watch whether lower token prices translate into higher enterprise AI spending; flat spending alongside rising usage would signal cheaper prices are offsetting volume gains. Analyst inference

Evidence

• Original source link unavailable