AI Token Spending Index Falls 20%, Raising Profitability Doubts

Pricing pressure builds as token spending gauge retreats

A closely watched measure of AI usage costs has declined nearly 20% from its May peak, according to data first reported by Bloomberg, raising fresh questions about whether the artificial intelligence industry can sustain its pricing power as competition intensifies and regulatory scrutiny grows.

The Silicon Data LLM Token Expenditure Index, which tracks what users pay for AI tokens, had nearly doubled since its December launch before beginning its recent slide. The index represents one of the clearest signals available about demand trends underlying the more than $700 billion in capital expenditures flowing into AI infrastructure.

Veteran investor Louis Navellier pointed to mounting evidence that customers are curtailing unlimited AI usage due to cost concerns. He noted that OpenAI's reported decision to delay its initial public offering until next year suggests profitability remains elusive even for leading players.

Why it matters

The token spending trend carries direct implications for the investment thesis supporting Nvidia, memory manufacturers, and data center operators. If customers are genuinely pulling back on willingness to pay premium prices for AI services, the economic justification for continued massive infrastructure spending weakens. Allianz Research has identified a 46% gap between AI investment and sales—worse than the 32% divergence seen during the 2001 telecom collapse. How this pricing dynamic resolves will determine whether current AI valuations reflect sustainable business models or speculative excess.

Interpreting the signal

Silicon Data cautions against reading the index as a simple price indicator. The gauge blends both pricing and usage patterns, meaning a decline could reflect falling list prices, customers shifting to cheaper models, or genuine softening in demand.

The bullish interpretation holds that cheaper tokens are expanding the total addressable market. While token prices have fallen more than 90% since 2023, total spending has roughly doubled over the past year. David Miller, senior portfolio manager at Catalyst Funds, argues that AI economics improve significantly during the inference stage compared to the training phase, delivering positive returns over the long term.

The bearish case warns that weakening token spending could undermine the entire AI investment cycle. If customers' willingness to pay has peaked, the pricing power story funding the march toward $1 trillion in annual capital expenditures by 2027 begins to fracture.

Regulatory headwinds compound uncertainty

Recent government actions add another layer of complexity. U.S. regulators this week removed foreign access restrictions on Anthropic's Fable 5 model and requested OpenAI stagger an upcoming release. The European Union's AI Act imposes mandatory evaluations and transparency requirements on frontier models.

These regulations don't directly cap prices but create compliance burdens that may push corporate buyers toward cheaper, less-regulated alternatives. DWS strategists led by chief investment officer Vincenzo Vedda cite "unbridled" market enthusiasm, intensifying Chinese competition, and price sensitivity as reasons for caution on stretched valuations.

Top-end graphics processing units and high-bandwidth memory remain sold out through 2026, with limited relief expected until 2028. However, a demand shift from training-focused GPUs toward inference-optimized chips could reshape which companies benefit most from continued AI spending.

The token spending index flattened in late June, leaving both interpretations viable. Whether this represents healthy market digestion or the beginning of pricing power erosion will determine the trajectory of AI investments in the months ahead.

These details were first reported by Bloomberg writers Barnert and Msika.