Workflow
智算中心情报大览:万卡集群性能弱导致上市公司资金链紧张;有十万卡集群因算力摸底被叫停;某智算中心验收通过后使用率从85%一路下降
雷峰网·2025-05-06 10:56

Core Viewpoint - The performance issues of a certain computing cluster in Northwest China have led to financial strain for Hongxin Electronics, which had partnered with a well-known chip company to build the cluster. The focus on upfront construction profits rather than operational revenue has created challenges for the company [1][3]. Group 1: Financial Strain and Revenue Models - Hongxin Electronics is relying on local procurement of its self-developed software products as a significant revenue source [2]. - The increasing rationality of local support policies has made it more difficult and risky to rely on subsidies for profit [3]. - Company A's significant revenue from "guaranteed sales commitments" has become a hindrance to its IPO process due to incomplete payments from local governments [4]. Group 2: Green Energy and Profitability - In a key computing hub in Northwest China, green energy indicators are viewed as a crucial profit mechanism for some computing center builders [5]. - The construction of computing centers aligns with national policies promoting green energy and new infrastructure, allowing builders to potentially profit from green energy integration [5]. Group 3: Utilization Rates and Market Dynamics - The actual utilization rates of computing centers are becoming less critical, as cloud giants are consolidating their computing resources to comply with local energy consumption indicators [7][8]. - A computing center in Changsha saw its utilization rate drop from 85% to around 50% after passing inspection, leading to dissatisfaction from local authorities [9][10]. - Some leading model startups have pressured computing centers to lower rental prices and provide financing, which has disrupted normal business logic [12][13]. Group 4: Challenges in Construction and Demand - A computing center in Chengdu faced a near financial collapse due to over-purchasing equipment without sufficient demand, but was revived by the success of a model startup [14]. - The demand for certain computing resources has surged following the popularity of specific models, leading to increased utilization rates for previously underused equipment [15]. - A major manufacturer faced issues with a procurement order due to excessive price pressure, resulting in no suppliers willing to fulfill the order [17]. Group 5: Policy and Project Viability - A computing cluster project in Inner Mongolia was halted due to new regulatory guidelines limiting large computing projects to designated national hubs [19]. - The rapid proliferation of computing center projects in Northwest China, without clear policies or planning, may lead to chaotic outcomes [18]. - Corruption issues within a cloud company have hindered the effective deployment of computing resources, further complicating the procurement process [20].