AI Token Price Collapse Threatens Hyperscaler Revenue
Goldman Sachs has sounded the alarm on the rapidly declining prices of AI model usage tokens. In September 2026, Goldman's Delta One trading desk issued a warning about the structural risks facing hyperscalers due to the collapse in token prices. The Silicon Data LLM Token Expenditure Index dropped by a record 29% in August 2026, reaching $0.97 per million tokens - more than 50% below its peak of $2.05 in May 2026.
The cost of running AI has plummeted due to the rapid improvement of proprietary models and the increasing adoption of open-source alternatives that undercut commercial offerings on price. Goldman's Rich Privorotsky pointed out that while token prices are falling by as much as 30%, usage is growing at only 10%. This gap would directly eat into revenue from AI infrastructure investments.
The companies most affected by this squeeze include Meta, Microsoft, Amazon, and Alphabet, which have all committed to unprecedented capital expenditure programs for AI infrastructure. The return on invested capital for hyperscalers has already dropped to an all-time low, according to Goldman's analysis.