智通财经APP获悉，东方证券发布研报称，英伟达(NVDA.US)于CES 2026大会上发布推理上下文内存 存储平台，该平台是一个POD级AI原生存储基础设施，核心目标是在GPU内存和传统存储之间创建一 个专为推理优化的新型内存层，以支撑AI长期运行。AI大模型推理过程中需要通过高频次地访问数 据，来实现高质量内容生成，将使得存储结构发生较大变化，提升对存储芯片的需求。当前存储供不应 求持续，同时海外存储巨头在通用存储方面的扩产进度可能有限，为国内存储厂商扩产、提升份额带来 历史性机遇。 东方证券主要观点如下： 东方证券此前强调，AI大模型推理过程中需要通过高频次地访问数据，来实现高质量内容生成，将使 得存储结构发生较大变化，提升对存储芯片的需求。展望未来，AI有望从"一次性对话的聊天机器人"演 进为理解现实世界、持续推理、调用工具完成任务的智能协作体，而这需要持续扩大上下文容量、加快 跨节点共享，从而带动存储芯片需求高速成长。 存储供不应求持续，重视存储产业链国产化机遇 存储供不应求持续，同时海外存储巨头在通用存储方面的扩产进度可能有限，为国内存储厂商扩产、提 升份额带来历史性机遇。技术方面，在DRAM领域 ...

说起存储芯片这事儿，美国人本来以为自己稳坐江山，谁知道长江存储这家中国企业硬是杀出一条血 路，不但技术上追赶上来，还在美国本土法庭上把美光这家老牌巨头逼得节节败退。 长江存储从2016年成立到现在，才短短几年，就从零起步变成全球NAND闪存市场的一股重要力量，这 中间的弯弯绕绕可不少，尤其在美国层层封锁下，它愣是靠国产化突围，搞得美国那边措手不及。 如果说以前的剧本是西方列强拿着专利大棒在谈判桌上教训后者，那现在的场面足可以让所有也是搞科 技的人瞠目结舌。 就在2025年的几个关键节点上，局势显得格外滑稽：一家号称代表美国顶尖技术的公司，在面对被指控 侵权时，祭出的最大护身符竟然是"不能说"。 当长江存储在美国加州法院要把专利官司打到底，要求美光把核心技术文件拿出来比对时，美光没有选 择硬碰硬地展示技术差异，而是高举"国家安全"的盾牌，死活不愿意通过披露程序。 他们大概是急了，一口气输掉了四场试图保密的裁决，最后愣是把状纸递到了美国最高法院，试图 用"紧急令"来捂住盖子。 这种心虚的表现，让人很难不联想到那个名叫Xtacking的技术架构——这正是让中国企业翻盘的"杀手 锏"。美光在法庭上的躲闪，恰恰印证了技 ...

Core Viewpoint - The global information technology industry is undergoing a fourth revolution centered around artificial intelligence (AI), which poses unprecedented challenges to the underlying hardware infrastructure, particularly in storage technology [1]. Group 1: AI's Impact on Storage Technology - The demand for storage technology driven by AI is primarily focused on three aspects: high performance, large capacity, and low power consumption, requiring storage manufacturers to find a delicate balance among these factors [3]. - Yangtze Memory Technologies Co. (YMTC) has introduced the Xtacking architecture, which separates storage cells from logic circuits, achieving breakthroughs in I/O speed (3.6 Gbps) and single die capacity (2 Tb) [3][4]. Group 2: Xtacking Architecture Advancements - The Xtacking technology has evolved to its fourth generation, with significant improvements in I/O speed from 800 MT/s to 3.6 GT/s, representing over a fourfold increase [4]. - Three NAND products based on Xtacking 4.0 were showcased, including a 512 Gb TLC flash chip with a 50% increase in I/O speed and over 40% improvement in storage density [4][5]. - The 1 Tb TLC flash chip also achieved 3.6 GT/s I/O speed and a 36% increase in storage density, gaining widespread market recognition since its mass production [5]. - The QLC flash chip under Xtacking 4.0 has a single die capacity of 2 Tb, with a 147% increase in data throughput and a 33% improvement in lifespan compared to previous generations [5]. Group 3: Product Line and Market Coverage - YMTC's product solutions encompass three main lines: embedded products, consumer products, and enterprise products, catering to various applications from IoT devices to cloud computing [9]. - The embedded product line includes flagship products like UC420, UC341, and UC260, designed for high-end AI smartphones and offering significant performance improvements and energy efficiency [10]. - In the PC segment, flagship products like PC550 and PC450 utilize PCIe 5.0 and PCIe 4.0 interfaces, respectively, achieving high performance while maintaining low power consumption [11]. - The upcoming enterprise-grade SSD PE511, based on Xtacking 4.0, is expected to offer a 100% performance increase and larger capacity options, enhancing operational efficiency in data centers [12]. Group 4: Future Outlook - YMTC's growth trajectory underscores the importance of innovation as the sole path to breakthroughs, with a focus on making storage faster, smarter, and more ubiquitous in the AI era [15]. - As AI evolves towards embodied intelligence and general artificial intelligence (AGI), storage technology is poised to become a core hub in the intelligent world, with Chinese companies likely to lead the global storage industry into a new cycle of value growth [15].