[SMM Computing Power News] 100 units of 5090 eight-card complete machines quoted at 455,000 yuan (excluding memory)

On September 15, a new spot quote for complete H100 systems emerged at 3.4 million yuan per unit. During the same period, the highest rental quote for this model in the Yangtze River Delta rose to 85,000 yuan per unit per month. A Shenzhen channel quoted 100 units of 5090 eight-GPU complete systems at 455,000 yuan per unit, explicitly excluding memory. Compared with the 650,000 yuan per unit quote on the same platform on September 11, which included 512G of memory, the price difference was approximately 195,000 yuan. When comparing complete systems, the memory specifications must first be standardized. The 4090 computing power cloud platform quoted 1.92 yuan per GPU-hour, while during the same period, the internal price at a northwestern university was as low as 1.2 yuan per GPU-hour.

All indices consolidated today. On the price collection side: spot H100 full systems were quoted at 3.4 million yuan per unit; spot H200 SXM supply and demand quickly aligned at 134,000 yuan per month, with PCIe explicitly excluded, potentially becoming a tradable reference level; 4090 platform cards were priced at 1.92 yuan per hour, while the internal price at a northwestern university was 1.2 yuan (cost structure and subsidy differences make direct comparison inappropriate). Prices of existing Nvidia cards surged across the board, and the follow-up construction of power infrastructure became a key constraint, requiring multi-party participation to build consensus on computing-power coordination.

SMM learned that a GPU rental cloud platform quoted 4090 computing power at 1.92 yuan per card per hour, continuing the price level for the same model previously collected by SMM from a south-west China platform. During the same period, the computing power price for the same server model used by faculty and students at a university in north-west China was 1.2 yuan per card per hour, about 38% lower than the market quote. SMM believes that the stability of third-party market-based pricing can serve as a price reference, while internal university prices reflect differences in cost-side factors and subsidy mechanisms, and the two types of metrics should not be directly compared.