本周，富士通正式官宣加入软银牵头的下一代AI内存开发项目，将与软银、英特尔及东京大学组成联 盟，打造"SAIMEMORY新型内存" 《日经亚洲》报道称，此次合作聚焦大型语言模型等复杂AI计算需求，旨在突破现有存储技术瓶颈， 该联盟将以三维堆叠DRAM技术打造高带宽内存（HBM）的直接替代品，研发兼具高性能与经济性的 存储器解决方案，为AI基础设施建设提供全新选择。 作为项目核心参与者之一，富士通将自身深耕多年的量子启发技术注入研发体系，包括张量网络模拟器 等关键技术，可有效优化深度电路分析效率，为新型内存的性能提升提供支撑。 该项目的技术路线涵盖多重创新方向，既包括动态存储器硅梳化等压缩技术，也涉及与光子芯片的深度 整合及后量子安全增强方案，其中光子芯片在生成式AI任务中的效能已被证实远超英伟达传统硅芯 片，成为差异化竞争的核心亮点。 在技术协作层面，项目将整合英特尔的垂直堆叠技术与东京大学在热管理、数据传输领域的学术成果， 原型设计与制造环节则携手新光电气（富士通曾为其大股东）、力积电推进，形成"产学研用"一体化的 研发闭环。 可以说，这不仅是富士通重返存储器领域的关键布局，也是日本近期发力半导体产业竞争力 ...

Group 1 - Xiaomi's 17 Ultra Leica version sold out within hours of its pre-sale launch, with the highest resale price reaching 20,000 yuan, significantly above the official starting price of 8,999 yuan [1][1][1] - The 16GB+512GB version is available on Xiaomi's official store, while the 16GB+1TB version is out of stock across various platforms, indicating high demand [1][1] - Over 1 million reservations were recorded on JD platform for the Xiaomi 17 Ultra Leica version [1] Group 2 - Zhou Dafu responded to criticism regarding its "cow and horse" gold pendant, stating it is only sold online and has seen high sales; they are open to customer feedback for future improvements [2][2] - The pendant is priced starting at 3,044.7 yuan, and the company clarified the pendant's meaning as representing resilience and good fortune [2] Group 3 - Xiaopeng Motors officially entered the Mauritius market after establishing a strategic partnership, following its previous entry into Qatar [7][7] - The company also opened a flagship showroom in Abu Dhabi and a large service center in New Cairo, Egypt [7] Group 4 - Douyin launched a long-form text feature allowing users to publish articles up to 8,000 words with up to 30 images and background music [6][6] - The mobile version of this feature is expected to roll out soon [6] Group 5 - LV responded to a customer's complaint regarding a pair of shoes that caused injury, stating that they only offered an exchange and did not support returns for used products [8][8] - The shoes were purchased for 9,250 yuan, and the customer required medical treatment due to the injury [8] Group 6 - The first mandatory standard for electric vehicle energy consumption will be implemented in China starting January 1, 2026, tightening limits by approximately 11% compared to previous recommendations [10][10] - This standard aims to guide the development of energy-saving technologies while accommodating diverse vehicle types [10] Group 7 - The first batch of L3 autonomous vehicles in China has begun large-scale road operation in Chongqing, consisting of 46 vehicles equipped with L3 autonomous driving systems [16][16] - These vehicles are authorized to operate on specific urban roads [16] Group 8 - Fujitsu will join a next-generation memory development project led by SoftBank, aiming for commercialization by the 2027 fiscal year [22][22] - This project will utilize technologies from Intel and the University of Tokyo [22]

Core Viewpoint - NVIDIA has revised its plans for the low-power DRAM (LPDDR) module SOCAMM, abandoning the SOCAM1 project in favor of SOCAM2, which is currently being tested with three memory manufacturers [1][2]. Group 1: SOCAMM Development - NVIDIA has started sample testing for SOCAMM2 with Samsung, SK Hynix, and Micron, shifting focus from SOCAM1 due to technical issues that prevented large-scale orders [1][2]. - SOCAMM aims to provide high-capacity memory at lower costs and power consumption compared to high bandwidth memory (HBM) [1]. Group 2: Technical Specifications - SOCAMM2 features the same 694 I/O ports as SOCAM1 but offers a data transfer speed of 9,600 MT/s, surpassing SOCAM1's 8,533 MT/s [2]. - The power consumption of SOCAMM is reported to be one-third lower than that of standard DRAM-based RDIMM server memory modules [2]. Group 3: Market Implications - The delay in SOCAMM1's launch has created opportunities for Samsung and SK Hynix, as they are now on equal footing with Micron in the development of SOCAMM2 [3]. - SOCAMM2 is expected to begin mass production in early next year, with increasing adoption anticipated as the demand for memory solutions to address data bottlenecks in AI computing rises [3].

Group 1 - The core viewpoint of the article highlights the stagnation of commercialization plans for CXL memory due to insufficient demand, despite technical readiness for mass production [1] - Major manufacturers like Samsung Electronics and SK Hynix are facing ongoing challenges in bringing next-generation memory technologies to market, particularly CXL and PIM [1][2] - The strong demand for High Bandwidth Memory (HBM), driven by NVIDIA's focus on this technology in its accelerator products, has delayed the application of CXL and PIM technologies [2] Group 2 - NVIDIA dominates the AI data center GPU market with an estimated market share of about 92% last year, making it difficult for competitors like AMD and Broadcom to gain market share [2] - Concerns are rising that Chinese companies may achieve commercialization of CXL and PIM technologies first, potentially altering the global competitive landscape [2] - Industry insiders suggest that proactive measures are needed to build an ecosystem conducive to future commercialization opportunities, as the current situation reflects a waiting game among stakeholders [3]

Core Viewpoint - Samsung Electronics is struggling with the mass production strategy for the next-generation NAND V10, facing delays and uncertainties in demand and technology introduction [2][5]. Group 1: NAND V10 Production Challenges - Samsung's V10 NAND features a stacking layer count of 430, surpassing the current V9's 290 layers by 100 layers [2]. - The company initially planned to start mass production investment in the second half of this year, but this has been postponed to the first half of next year due to supply chain finalization delays [2][5]. - The introduction of new low-temperature etching processes has proven difficult, with evaluations indicating that immediate application for mass production is unlikely [3][4]. Group 2: Equipment and Cost Factors - Samsung is collaborating with major equipment manufacturers like Lam Research and TEL to assess low-temperature etching equipment, but the results have led to a reevaluation of the approach [3][4]. - The investment cost for new equipment is a significant factor in delaying the V10 NAND mass production, as diversifying suppliers may reduce the utilization rate of existing equipment [4]. Group 3: HBM3E Supply Agreement with AMD - Samsung has secured a supply agreement with AMD for the fifth-generation 12-layer HBM3E memory, which will be used in the upcoming MI350 AI accelerator [6]. - This agreement marks a breakthrough for Samsung, especially after previous setbacks against competitors like NVIDIA [6]. - The new HBM3E chip offers over 50% improvements in performance and capacity compared to the previous eight-layer version, supporting a bandwidth of up to 1,280GB/s [7]. Group 4: Future HBM4 Developments - HBM4 is seen as a critical battleground for market dominance among Samsung, SK Hynix, and Micron, with plans for mass production by the end of this year [8]. - Samsung aims to regain its competitive edge in the HBM market by adopting advanced sixth-generation process technology, which could provide significant advantages over competitors using fifth-generation technology [8].