Woofun AI data shows that the Silicon Data LLM Token Consumption Index has declined nearly 20% from its May peak, signaling a halt in high-growth trends. This drop coincides with a widening divergence between AI capital expenditure and revenue, which now stands at a 46% gap, surpassing the 32% imbalance recorded during the 2001 telecom bubble burst.

While bearish analysts cite this discrepancy as a warning of diminishing pricing power for large model providers, bullish counterparts argue that total expenditure has nearly doubled despite a 90% drop in average token prices since 2023.

Concurrently, stricter regulatory frameworks in Washington and the EU are driving CFOs to shift workloads toward lighter models to mitigate compliance costs. In the hardware sector, procurement leaders are transitioning from training-focused chips to inference-optimized solutions, despite top-tier GPU and HBM orders remaining fully booked until 2026.