ChainThink reports that on July 6, the Silicon Data LLM Token Consumption Index, which tracks users' actual compute spending, fell nearly 20% from its May peak, indicating a significant slowdown in AI compute consumption and intensifying market skepticism toward large model providers' pricing power and the return on hundreds of billions of dollars in AI capital expenditures.
Allianz Research data shows that the growth gap between AI investment and sales has reached 46%, higher than the 32% observed during the dot-com bubble burst in 2001.
However, bullish perspectives argue that although the average token price has dropped by about 90% since 2023, total spending has nearly doubled, and the index's decline more closely reflects structural adjustment following the price drop. Regulatory factors are also increasing the cost of corporate adoption.
The U.S. has tightened its review of frontier model releases and cross-border access, while the EU’s Artificial Intelligence Act imposes stricter compliance requirements on top-tier models, prompting some companies to shift workloads toward lighter, less regulated models. Changes are also occurring on the hardware side.
Although top-tier GPU and HBM orders are booked through 2026, and a noticeable relief in supply and demand may not occur until 2028, the market's purchasing focus has begun shifting from training chips to inference-optimized hardware, reshaping the industry's beneficiary landscape.