[SMM Computing Power Flash] High-end computing power 128-unit futures were quoted at 13.2 million per unit, and the buyer said actual transaction prices were just over 10 million per unit.

SMM learned that recently 10 Ascend 910B2 eight-card servers were deployed in Zhejiang region, with a monthly rental of 12,000 yuan. The servers are equipped with 4× Kunpeng 920 processors, 8× Ascend 910B-64G accelerators, HCCS inter-card connectivity, 2TB memory, and dual 3.84TB NVMe data disks, while the network side is equipped with 8× 200G RoCE v2 high-speed interconnect. Market sources indicated that the 200G RoCE networking configuration has certain advantages among domestically-produced computing power at the same price point, and is suitable for distributed training and high-throughput inference scenarios.

SMM learned that on July 28, a personnel structure survey was conducted in a computing power industry WeChat group. The group had about 120 members, among whom around 40 were intermediaries (brokers/matchmakers), accounting for roughly 30%. This means that about 3 out of every 10 members focused on deal matchmaking as their core business, highlighting the prominent layering of intermediaries in the computing power leasing market. The high density of intermediaries causes the same batch of supply to pass through multiple middlemen, widening the price spread between the end-user quotation and the actual owner's price. Consequently, the difficulty for buyers to identify first-hand supply sources increases. The 30% intermediary ratio confirms a high degree of information asymmetry in the current computing power leasing market, with transaction costs continuously rising due to the accumulation of intermediate links.

The highest monthly rental price for A800 40G GPUs in the Yangtze River Delta surged to 20,000 yuan (up 21.21%), with the highest hourly rate reaching 3.47 yuan, driven by the clustering of the AI industry combined with rigid energy costs; the lowest monthly rental price for A800 80G GPUs in the Beijing-Tianjin-Hebei region dropped to 31,000 yuan (down 3.13%), as clean energy from western Inner Mongolia and supply growth from the "East Data, West Computing" project create a pricing advantage. Regional price differentials reflect a structural trade-off between "proximity to demand" and "proximity to power.