九章云极通过智算系统创新降低大模型算力成本

九章云极作为AI基础设施提供商，通过自主研发的Alaya NeW智算操作系统和DCU标准化计量单位，为客户显著降低了算力支出。1该公司已将某头部大模型厂商的综合算力成本降低60%以上，1并在包月计费与DCU计费的对比测试中实现40%的成本节省。1

DCU标准化计量单位由九章云极定义，其中1度算力（DCU）等于312TFLOPS（FP16）的有效浮点计算持续1小时。1通过这一更精准的计量方式，客户可以根据实际算力消耗按需付费。九章云极已适配智谱GLM-5.2、DeepSeek-V4 Flash、Kimi K3等多个主流大模型，1同时发现不同模型能力的单位Token成本差距高达五六倍，1这意味着合理选择和优化模型配置对降低成本至关重要。

九章云极由曾在微软任职的尚明栋和方磊于2013年3月创立。1尚明栋曾担任微软服务器高可用集群和文件系统开发工程师，参与了Windows 7和Windows 8的研发工作。1公司从2022年底开始向智算云服务商转型。1

Jiuzhang Cloud Extreme, an AI infrastructure provider, has developed proprietary solutions to optimize computing efficiency and lower costs for large language model developers. The company's self-developed Alaya NeW intelligent computing operating system and DCU standardized measurement unit have enabled it to reduce the comprehensive computing costs of a leading large model manufacturer by over 60 percent 1.

The DCU metric is defined as 312 TFLOPS of effective floating-point calculation at FP16 precision sustained over one hour 1. When compared with monthly billing models, the company's DCU-based pricing structure demonstrated cost savings of 40 percent in real-world testing 1. Jiuzhang Cloud Extreme has successfully adapted multiple mainstream large language models to its platform, including Zhipu GLM-5.2, DeepSeek-V4 Flash, and Kimi K3 1. The company notes significant variations in per-token computing costs across different model capabilities, with differences reaching five to sixfold 1.

Founded in March 2013 by Shang Mingdong and Fang Lei, Jiuzhang Cloud Extreme began its transition to intelligent computing cloud services in late 2022 1. Shang Mingdong previously served as a development engineer at Microsoft, specializing in server high-availability clusters and file systems, and contributed to the development of Windows 7 and Windows 8 1.