3. 英伟达、AMD等公司：据媒体援引知情人士透露，美国官员已起草法规草案，拟限制全球范围内未 经美国批准的人工智能芯片发货，法案将赋予华盛顿广泛的权力，以决定其他国家是否能够以及以何种 条件建设用于训练和运行人工智能模型的设施。拟议法规将要求企业向美国申请许可，才能出口几乎所 有英伟达和AMD等公司生产的人工智能加速器。这将使目前覆盖约40个国家的管制措施扩展至全球范 围。 4. 京东集团：2025年全年收入1.31万亿元，同比增加13%；第四季度收入3523亿元，同比增长1.5%，归 属于公司普通股股东的净损失27亿元，非美国通用会计准则下每股美国存托股摊薄收益为0.57元，电商 业务持续稳健发展。 5. 阿里巴巴：针对近日网络流传"千问模型核心团队集体离职""开源策略调整"等不实信息，阿里集团表 示，目前千问模型团队稳定，没有出现"集体离职"的情况，所有产品与服务运行正常，千问会坚持开源 策略。阿里旗下首家线下潮玩店"好运连得 LUCKY LOOP"已正式落地北京。 // 大中华地区公司要闻 // 1. 哔哩哔哩：2025年全年净利润扭亏为盈达11.9亿元，全年净营业额303.5亿元，同比增长13%；第 ...

Core Viewpoint - Micron has officially launched the shipment of its sixth-generation High Bandwidth Memory (HBM4), breaking the expectation that Samsung and SK Hynix would dominate the supply chain for NVIDIA's upcoming AI accelerator, "Vela Rubin" [1]. Group 1: Micron's HBM4 Production and Market Position - Micron's CFO, Mark Murphy, announced that the company has entered the mass production phase of HBM4 and has begun shipping to customers [1]. - Murphy addressed recent rumors regarding HBM4's exit from the market, stating that the shipment volume is expanding smoothly and is ahead of the previously mentioned timeline by one quarter [1]. - The production capacity for HBM is steadily increasing as planned, with Micron's HBM capacity for 2026 already sold out, and the yield for HBM4 meeting expectations, providing speeds over 11Gbps [1]. Group 2: Competitive Landscape - Prior opinions suggested that Micron's HBM4 could not meet NVIDIA's requirement of over 11Gbps, placing it at a disadvantage against Samsung and SK Hynix in the supply competition [1]. - SemiAnalysis projected that SK Hynix and Samsung would capture 70% and 30% of the supply chain for NVIDIA's Vela Rubin, respectively [2].

Core Viewpoint - TSMC plans to build four advanced packaging (AP) factories to address capacity shortages and maintain its competitive edge in the semiconductor industry, particularly in advanced packaging technologies like CoWoS [1][4][11] Group 1: TSMC's Expansion Plans - TSMC will announce the expansion of four advanced packaging factories in Tainan, including locations in Chiayi Science Park and Southern Science Park [1] - The company aims to start mass production at its AP factory 1 in the Ziyi Technology Park in the first half of this year [1] - TSMC's expansion is a response to concerns about its potential transformation into "American TSMC" due to recent factory expansions in the U.S. [1] Group 2: Industry Trends and Challenges - The global tech industry is facing intense competition for advanced packaging technology, particularly TSMC's CoWoS, which is critical for connecting high-performance chips with ultra-fast memory [4][9] - By 2026, the bottleneck in AI GPU supply will shift from chip shortages to the complex assembly processes required for advanced packaging [4][9] - The transition from wafer-level packaging (WLP) to fan-out panel-level packaging (FOPLP) is expected to increase processing capacity significantly [11] Group 3: Strategic Implications - NVIDIA has secured nearly 60% of TSMC's CoWoS capacity for 2026, influencing competitors like AMD and Broadcom to vie for the remaining capacity [7] - The advanced packaging secondary market is rapidly maturing, with companies like Intel positioning their packaging technologies as alternatives to TSMC [8] - The industry's reliance on TSMC for advanced packaging creates vulnerabilities, as geopolitical stability in the Taiwan Strait remains a critical factor for the global AI economy [8][12] Group 4: Future Outlook - The industry's focus is shifting towards the physical realities of AI hardware, with advanced packaging becoming a crucial factor in the growth of AI capabilities [9][10] - Upcoming challenges include the transition to glass substrates for improved interconnect density and thermal management, which could disrupt TSMC's current dominance [11] - The success of HBM4 chip yields and the ramp-up of TSMC's AP7 capacity will be closely monitored, as delays could impact the release of next-generation AI models [12]

Core Viewpoint - The U.S. storage sector stocks experienced a significant surge, driven by SK Hynix's successful negotiations with NVIDIA regarding the pricing of the sixth-generation high bandwidth memory (HBM4), which increased by over 50% to approximately $560 per unit [1] Group 1: Stock Performance - Major storage stocks such as Seagate Technology (STX.US) and SanDisk (SNDK.US) rose by over 9%, while Micron Technology (MU.US) increased by nearly 8%, and Western Digital (WDC.US) saw a rise of over 5% [1] Group 2: SK Hynix's Market Position - SK Hynix has strengthened its dominant position in the high-end storage chip market by successfully negotiating a price increase for HBM4, which is expected to lead to a record operating profit of over 70 trillion Korean Won next year [1] Group 3: Pricing Dynamics - Initially, NVIDIA resisted the significant price hike due to potential competition from Samsung and Micron, but ultimately agreed to the price set by SK Hynix at approximately $560 per unit [1] - A SK Hynix executive indicated that advancements in technology and increased input costs justified the substantial price increase for HBM4 [1]

Core Viewpoint - SK Hynix has confirmed its position as the leading supplier in the HBM market, with a price increase of over 50% for its sixth-generation high bandwidth memory (HBM4) compared to the previous generation (HBM3E) [1] Group 1: Company Developments - SK Hynix successfully completed negotiations with Nvidia for HBM4 supply for next year, paving the way for record-breaking performance [1] - The company delivered the world's first 12-layer stacked HBM4 samples to Nvidia in March, and began initial shipments in June after receiving positive evaluations [1] - Ongoing price negotiations for HBM4 required for Nvidia's next-generation AI chip "Rubin," set to be released in the second half of next year [1]

Core Insights - Samsung Electronics is preparing for final quality testing of NVIDIA's sixth-generation High Bandwidth Memory (HBM4), with its foundry division ramping up production of the logic chips that serve as the "brain" of HBM [1][2] - The logic chips for HBM4 will utilize advanced process technology, with Samsung's foundry achieving over 90% yield for 4nm process logic chips, indicating readiness for mass production [2][3] - The transition to HBM4 marks a significant shift in performance requirements, with the introduction of customized HBM expected to begin with HBM4E, necessitating tailored circuit designs based on customer needs [3][4] Summary by Sections HBM4 Production and Testing - Samsung's foundry division is fully supporting the production of logic chips for HBM4, aligning with the storage division's increased HBM4 output [1] - The foundry is increasing the wafer input for logic chips and aims to maximize yield to establish a stable production system [1] Logic Chip Technology and Yield - The logic chips for HBM4 are critical components that connect stacked DRAM with GPUs in AI semiconductors, providing power supply and data signal control [2] - The yield for 4nm process logic chips has surpassed 90%, meaning that more than 9 out of 10 chips produced are qualified, demonstrating a stable technology for large-scale production [2][3] Industry Trends and Future Outlook - The semiconductor industry is witnessing a shift towards advanced process technologies, with Samsung's 4nm process entering a mature stage and achieving an overall yield exceeding 80% [3] - The upcoming HBM4E is expected to usher in a "customized HBM" era, where HBM development will be optimized based on specific AI semiconductor applications [3][4] - The importance of foundry process technology is anticipated to increase in the next generation of HBM products, as Samsung aims to address previous product shortcomings [4]