Demand for AI infrastructure remains generally healthy, but subtle changes are occurring within its internal structure: 1. Token spending on large language models (LLMs) has slowed after strong growth in June. Open-source models are eating into market share—for example, Kimi K3 has entered the top five in spending, breaking the previous dominance of closed-source models. 2. The rental price of the new generation GPU B200 has accelerated its growth, rising 7.3% month-over-month to $5.72 per GPU-hour. On the other hand, the rental price of the older H100 has decreased by 1.1% month-over-month, marking its first decline in seven months. This suggests demand is shifting toward the latest generation of chips (such as the Blackwell architecture). This generational product shift benefits NVIDIA (as customers are willing to pay higher rents for the new architecture, supporting B200 scaling and extending the product cycle). 3. Price increases in the storage market have narrowed significantly. Contract prices for DRAM and NAND are expected to grow only 16% and 13%, respectively, in Q3 2026 (compared to growth rates as high as 56%/58% in Q2). This is mainly due to high base effects, weak consumer demand, long-term agreements by cloud service providers (CSPs), and high inventory levels caused by pre-stocking in the first half of the year. (Source: JPMorgan)