2026-09-18
Dear Users,
Greetings!
As China's computing power rental market enters a phase of large-scale expansion, multiple GPU and accelerator card products — including NVIDIA H100, A100 and H20, Huawei Ascend 910B and RTX 4090 — now circulate in parallel. Quotations differ materially across card models, hardware configurations and contract terms, leaving the market without a price benchmark capable of uniformly measuring changes in unit computing power cost. To fill this gap, and to help market participants track the overall movement of computing rental costs while improving cost estimation and negotiation efficiency, SMM — after extensive market research, methodology validation and technical development — will officially launch the SMM Computing Power Rental Composite Price Index with effect from 21 September 2026. The index will be published on a daily basis, updated before 11:30 a.m. Beijing Time on each trading day.
Details are set out below:
I. Data Point Overview
The newly launched data point is the SMM Computing Power Rental Composite Price Index, which measures — through conversion into a unified Standard Computing Unit (SCU) equivalent — the composite movement of unit computing power cost in China's computing power rental market.
Full index name (with complete specification): SMM Computing Power Rental Composite Price Index, China, Bare Metal, VAT included, yuan/GPU·hr
II. SMM Computing Power Rental Composite Price Index
2.1 Product Positioning
The SMM Computing Power Rental Composite Price Index is an index-type data point. It converts heterogeneous GPU/accelerator products including NVIDIA H100, A100, H20, Huawei Ascend 910B and RTX 4090 into a unified Standard Computing Unit (SCU) equivalent, establishing a unit computing power price benchmark that is comparable across card models and across pricing specifications. Market participants may use it for direct reference and comparison, and to track the composite trend of computing power rental costs.
2.2 Coverage
Region: China (nationwide);
Hardware and delivery form: bare metal, whole-server configurations;
Products covered: mainstream GPU/accelerator cards including NVIDIA H100, A100, H20, Huawei Ascend 910B and RTX 4090, published after conversion into SCU equivalents; coverage will be updated dynamically as data sources expand;
Unit of quotation: yuan/Card·hr, VAT included.
2.3 Standard Specification (Pricing Basis)
The index uses a uniform set of standardized delivery terms as its pricing basis, as follows:
-
Hardware basis: the benchmark equivalent is a Standard Computing Unit (SCU) equivalent to an H100, delivered as a whole server containing 8× NVIDIA H100-80G GPUs;
-
Electricity: GPU-level power consumption included; shared whole-machine overhead is noted with PUE;
-
Public bandwidth: local egress of up to 1 Gbps included;
-
Storage: in-chassis NVMe per specification included;
-
O&M: hosting, O&M, imaging and large language model platform services excluded;
-
Lease commencement: minimum 1-month lease term, with the minimum lease unit being one full 8-GPU dedicated server;
-
Delivery lead time: within 2 weeks;
-
Payment terms: cash, and other payment methods normalized to cash.
2.4 Price Notes
The index reflects the composite price level of computing power rental in China under the standardized specification defined above, quoted in yuan/Card·hr (VAT included). After eliminating differences arising from non-standard terms, the index is compiled on a composite basis to reflect the direction of change in unit computing power usage cost. The index is released daily, before 11:30 a.m. Beijing Time on each trading day.
III. Usage Guidelines and Notes
3.1 Use Cases
Tracking computing cost trends and internal cost management; floating pricing benchmarks and settlement references for computing rental contracts; annual enterprise computing budget estimation; cost benchmarking between self-built and rented computing capacity; and capacity selection across card models and regions.
3.2 Important Notes
This index is a composite index of unit computing power rental prices defined under a standardized specification. It is not equivalent to the actual transaction price of any individual card model, and does not cover customized, hosted or non-publicly traded computing capacity prices. Actual transaction prices are affected by supply and demand, negotiating power, contract terms, and value-added configurations such as bandwidth and storage, and may therefore differ from the index. SMM will continue to optimize source coverage and calculation methodology, and will progressively extend coverage to additional card models and regions. For data corrections, price verification or new data source recommendations, please contact the SMM Computing Power Research Team.
IV. Contact
Should you have any questions or suggestions, please reach us at:
SMM Computing Power Research Team
E-mail: fangwenda@smm.cn ; liuxiaolei@smm.cn
Tel: +86-21-51666821
SMM Information & Technology Co., Ltd. — Computing Power Research Team