Token ' s recent weakness in the use of key indicators of expenditure has led the market to revisit the demand base of this round of AI commercial expansion. By reference to the Sil token n Data data, Bloomberg states that the LLM Token Expendure Index, which it produced, has fallen nearly 20 per cent since the peak of May, after having almost doubled since its launch last December.

This indicator is often used to observe the marginal willingness of enterprises and users to pay for AI services. However, index fall does not amount to an overall fall in AI service prices. In the view of Sil token issuen Data, this indicator is influenced by both price and use, which is closer to the proxy variable of a change in willingness to pay and therefore cannot be simply interpreted as a single price signal.

Market differences are concentrated on the demand side

There are two main interpretations of the market for this downturn.

A more optimistic view is that the price of Token has fallen significantly since 2023, with the subsequent lowering of the use threshold, and that overall spending expansion is likely to continue. In such cases, the index fall more like a restructuring of demand than a clear weakening of overall demand.

A more cautious view was that the marginal willingness of users to pay might be close to a phased ceiling. Allianz Research mentioned that there is a large gap between current AI investment growth and actual sales, which means that the risk of associated valuation pressures will be more rapid once the demand side continues to slow.

The need for power is shifting from training to reasoning.

AI infrastructure investments have not been significantly reversed despite fluctuations in payment demand signals. The report mentions that high-end GPU and high-bandwidth RAM are still under stress and that supply and demand imbalances may persist until 2026, and part of the judgement may even continue until 2028.

However, market concerns are changing. Previously, capital expenditure had been more focused on model training and was now moving towards the reasoning chain. This means that the need for computing power is no longer focused solely on the top training chip, and that the importance of optimised reasoning hardware is increasing and the direction of the benefits of the chain may be adjusted accordingly.

From this perspective, current changes do not necessarily mean that the chip industry is entering a lower cycle, but the sources of growth are changing. The past high-end GPU-driven model is moving towards a more decentralized hardware demand structure.

Regulatory changes increase commercialization costs

In addition to demand and hardware structures, regulatory factors also influence the pricing and deployment of AI products. The report mentions that the United States regulatory hierarchy has recently requested adjustments in the timing and access arrangements for some of the models, while the EU ' s AI Act has increased assessment and transparency requirements for the front-line models.

These changes do not directly lower model prices, but increase deployment and compliance costs. For enterprises, the importance of cost optimization will continue to rise when work loads are distributed among different models, which may further affect the ratio and pricing capacity of high-end models.