[Market Overview]
The computing power leasing market remained flat today, with the index price showing no significant fluctuations. A new signal emerged in the domestically-produced computing power sector: the monthly rental quote for 910B4 in south-west China was 16,000 yuan per unit per month, and the procurement quote for a domestically-produced eight-card complete machine was approximately 760,000 yuan per unit, providing a new reference for the domestically-produced computing power pricing system.
[Domestically-produced Computing Power: Rental and Procurement Prices Simultaneously Emerge]
SMM learned that the rental quote for 910B4 in south-west China was 16,000 yuan per unit per month. Meanwhile, the procurement quote for a domestically-produced Huawei Ascend 910B4 eight-card complete machine (equipped with eight Huawei Ascend 910B4 domestically-produced high-end accelerator cards) was approximately 760,000 yuan per unit, including three years of hardware maintenance and installation services, with a procurement cost of approximately 95,000 yuan per card.
This configuration was aimed at large model inference scenarios, supported by enterprise-level memory, NVMe SSDs, and a 200G network interface. The simultaneous emergence of rental and procurement quotes indicated that domestically-produced computing power was transitioning from a stage of "signals without prices" to a tradable and priceable phase.
[SMM View]
SMM believed there were two points to focus on. First, the correspondence between the monthly rental of 16,000 yuan for the 910B4 and the complete machine procurement price of approximately 760,000 yuan provided an anchor for calculating the domestically-produced computing power leasing yield.
Second, under the sideways market, the increase in quotes for domestically-produced models indicated that the supply side was expanding from a sole reliance on NVIDIA to a diversified computing power structure. The inclusion of three years of maintenance and installation services in the domestically-produced complete machines also improved the standardization of delivery. Overall, domestically-produced computing power was transitioning from a price discovery phase to a price anchoring phase, jointly building a multi-dimensional price gradient with NVIDIA's high-end models.
![[SMM Cooperation] SMM Computing Power Price Index Increases Guangzhou Lingjing Technology as a Price Submitter](https://imgqn.smm.cn/usercenter/obeMy20251217171735.jpg)
![[SMM Computing Power Express] Domestic 910B4 inference machine quoted at 763,000 yuan per unit](https://imgqn.smm.cn/usercenter/wDeWQ20251217171734.jpeg)
![[SMM Computing Power Daily] H200 Spot Cargo: 256 Units Quoted at 5.35 Million, High-End Cards Shipped in Concentrated Batches](https://imgqn.smm.cn/usercenter/nQsOk20251217171736.jpg)
