行 业 2025 年 7 月 25 日 刘梦麟 SAC 执业证书编号： S0340521070002 电话：0769-22110619 邮箱： liumenglin@dgzq.com.cn 周 报 陈伟光 S0340520060001 电话：0769-22119430 邮箱： chenweiguang@dgzq.com.cn 资料来源：东莞证券研究所，iFind 超配（维持） 半导体行业双周报（2025/07/11-2025/07/24） 长江存储全国产化产线今年试产，关注半导体设备国产替代机遇 投资要点： 本报告的风险等级为中高风险。 本报告的信息均来自已公开信息，关于信息的准确性与完整性，建议投资者谨慎判断，据此入市，风险自担。 请务必阅读末页声明。 半导体行业 半导体行业指数近两周涨跌幅：截至2025年7月24日，半导体行业指数近两 周（2025/07/11-2025/07/25）累计上涨4.33%，跑赢沪深300指数0.86个百 分点；2025年以来申万半导体行业指数累计上涨5.47%，跑赢沪深300指数 0.03个百分点。 SAC 执业证书编号： 半导体行业（申万）指数走势 行业新闻与公司动态：（1 ...

在 2025 年第一季度，全球存储市场传来令人振奋的消息：中国的长鑫存储（CXMT）与长江存储 （YMTC）季度营收均突破 10 亿美元大关。这一里程碑式的成就，标志着国内存储企业在全球舞台上 正逐步崭露头角，打破了长期以来由国际巨头主导的市场格局。 长鑫存储专注于 DRAM 领域，自成立以来，通过持续的技术研发与产能扩张，已成为国内 DRAM 产业 的领军者。其在合肥的生产基地不断提升产能，技术工艺也逐步向国际先进水平靠拢。 长江存储则在 NAND Flash 领域深耕细作，凭借自主研发的 3D NAND 技术，在全球市场中占据了一席 之地，产品广泛应用于消费电子、数据中心等多个领域。 长鑫存储与长江存储的崛起，对国内存储产业乃至整个半导体行业都具有深远的影响。一方面，它们的 成功为国内其他存储企业树立了榜样，带动了整个产业的技术升级与发展。另一方面，国内存储企业的 壮大，也有助于提升我国在全球半导体产业链中的地位，降低对进口存储芯片的依赖，保障国家信息安 全。 加入"中国IC独角兽联盟"，请点击进入 投稿 、 商务合作 请微信 dolphinjetta 是说芯语，欢迎关注分享 近年来，全球存储市场需求持续 ...

Group 1: Trade and Tariffs - The U.S. has announced the initiation of the withdrawal of the "AI diffusion rules" from the Biden administration, strengthening export controls on overseas AI chips, claiming that using Huawei's Ascend AI chips globally may violate U.S. regulations [3] - The U.S. Trade Representative's office plans to impose a 100% tariff on Chinese cranes and 20%-100% tariffs on container and chassis equipment, citing the need to bring shipbuilding back to the U.S. [3] - The U.S. Commerce Department has preliminarily determined that Chinese battery component materials receive "high subsidies," proposing countervailing duties that distort normal market competition into a "government subsidy war" [3] Group 2: Technology and Export Controls - The U.S. has included Huawei's Ascend chips in the category of "violating U.S. export controls," warning global companies against using U.S. AI chips to train Chinese AI models [4] - Reports indicate that the Trump administration is preparing to add multiple Chinese tech companies to the control list, expanding the "entity list" beyond Huawei and ZTE to include chip manufacturers [4] - The U.S. aims to cut off the development of China's tech industry through targeted measures, transforming technology exchange into a "tech cold war" [4] Group 3: U.S. Strategic Misjudgments - The U.S. is driven by political opportunism, with some politicians leveraging anti-China sentiment for political gain, even at the cost of U.S.-China economic relations [5] - The U.S. exhibits "hegemonic anxiety" as China's GDP approaches that of the U.S., attempting to delay China's industrial upgrades through tariffs and technology blockades [5] - The U.S. maintains a zero-sum game mentality, ignoring the deep integration of global supply chains and attempting to preserve its top position in the global value chain through "decoupling" [5] Group 4: Consequences of Unilateral Actions - Historical evidence shows that unilateral sanctions ultimately backfire, as seen in the 2025 trade war where the U.S. imposed a 145% tariff on Chinese goods, leading to significant losses for companies like Tesla and Apple, and increased living costs for American households [6] - The intensified technology blockade has caused companies like NVIDIA and AMD to lose access to the largest AI chip market, prompting China to accelerate the development of its semiconductor industry [7] - The U.S. efforts to form a "semiconductor alliance 2.0" have faced challenges, as countries like South Korea seek exemptions for chip exports to China, and companies like ASML continue to supply China with lithography machines [7]

Core Viewpoint - Huawei's CloudMatrix 384 super node technology signifies a breakthrough in China's computing capabilities, demonstrating that technological blockades cannot hinder the country's advancements in AI and computing power [1][3][8]. Group 1: Technological Advancements - Huawei's history of overcoming challenges is marked by significant technological breakthroughs, including the launch of its self-developed AI chip in 2023, which competes with Nvidia's A100 [3]. - The CloudMatrix 384 super node, consisting of 384 Huawei AI chips, achieves a performance surpassing Nvidia's H100, with a throughput of 1920 Tokens/s compared to H100's 1850 Tokens/s [5]. - The architecture of CloudMatrix 384 employs a fully peer-to-peer interconnect bus technology, achieving an inter-card bandwidth of 2.8 Tbps and training efficiency at 90% of single-card performance [5]. Group 2: Market Impact - In Q1 2025, Huawei's AI chip market share in China reached 38%, while domestic production of AI chips surged, with imports dropping by 60% and local shipments increasing by 180% [6]. - The adoption of Huawei AI chips in China's government and enterprise sectors has surpassed 50%, with 70% of equipment in local intelligent computing centers utilizing Huawei technology [6]. - Southeast Asian countries, including Malaysia and Thailand, have begun signing cooperation agreements with Huawei for computing power, with Penang's packaging plant expected to meet 30% of global AI inference demand by 2026 [6]. Group 3: Energy Efficiency and Cost Reduction - The implementation of liquid cooling technology in data centers has reduced the Power Usage Effectiveness (PUE) to 1.1, cutting energy consumption by 40%, with a total power consumption of only 172.8 kW for a single cluster [5]. - The training costs associated with Huawei's technology have decreased by 75% compared to three years ago, thanks to the integration of the open-source MindSpore framework across over 3000 application scenarios [5]. Group 4: Strategic Positioning - The release of CloudMatrix 384 reflects a shift in the logic of computing competition, moving from single-point breakthroughs to system-level leadership in AI infrastructure [8]. - Huawei's advancements are seen as a response to the U.S. sanctions, effectively breaking the "digital Berlin Wall" and establishing a parallel ecosystem based on self-developed technologies [8].