The Silicon Data LLM Token Expenditure Index, which tracks what users are paying for AI tokens, has dropped almost 20% from its peak in May 2026. This index nearly doubled since its inception in December. This decline is significant because total capital spending on AI infrastructure, tracked by this index, already exceeds an estimated $700 billion and is projected to reach $1 trillion by 2027. Some analysts consider token spending a critical indicator of whether end-users are willing to pay for the extensive AI infrastructure being built.
This drop raises questions about the pricing power of AI companies, especially as customers become more cost-conscious. While token prices have actually collapsed more than 90% since 2023, aggregate total spending has continued to rise, suggesting that cheaper tokens have broadened the market. However, a sustained weakness in the index could signal that the expected AI bonanza might not be as profitable as anticipated, and could impact the hardware, memory, and data center sectors. Allianz Research notes a significant gap between AI investment and sales growth, currently at around 46%, which is higher than the 32% divergence seen during the 2001 telecom bust.
The decline in the index can be interpreted in several ways. It could mean that list prices are falling, or that customers are shifting towards cheaper AI models. Another possibility is that buyers are re-evaluating their willingness to pay for extensive AI usage once monthly bills arrive. Some optimistic views suggest the dip is merely a period of digestion after a strong run, and that cheaper models will expand the market, making AI a broader commercial utility rather than an expensive corporate experiment. However, veteran investor Louis Navellier points out increasing reports of users limiting AI use due to high costs, interpreting OpenAI's reported delay of its IPO as a sign of current profitability issues.
The bullish view argues that while token costs have fallen sharply, total usage is still climbing, and the pause in the index could be a result of a demand mix shifting from high-end training GPUs to more inference-optimized parts. This could lead to a change in the winners within the AI hardware market rather than an overall market collapse. However, if customer willingness to pay is peaking just as regulatory headwinds encourage a shift to cheaper models, the massive, expensive investments in AI infrastructure may become vulnerable, as the market transitions from being a "silicon story" to a "pricing-power story."
