Goldman Sachs reported that the Silicon Data LLM Token Expenditure Index, which measures the weighted average price per million tokens used in the market, fell 29% in August alone to $0.97, breaking below the $1 mark for the first time and accumulating a drop of over 50% from its May peak. Rich Privorotsky, head of Goldman Sachs' One-Delta trading desk, believes this shakes the "more tokens equal more revenue" capital expenditure logic that has underpinned AI sector valuations for the past two years.

Meanwhile, JPMorgan Chase data shows that OpenRouter platform's token usage increased by approximately 47% month-over-month in August, but dollar expenditure only grew by about 7%, revealing a "volume up, price down" divergence. The Goldman Sachs report suggests that intensifying competition and improved local inference capabilities are structurally eroding the "pay-per-token" business model, with credit markets already signaling caution as bonds of related cloud vendors have seen "repricing."

OpenAI stated on September 1 that its next-generation model, Astra, has met critical cybersecurity thresholds, and is seen as the only near-term catalyst that could potentially reverse the trend, but Goldman Sachs remains cautious.