智通财经APP获悉，广发证券发布研报称，随着生成式AI迈向具备长期推理能力的代理型系统，AI Agent在执行任务时需频繁存取庞大的矢量数据库以进行检索增强生成，相关数据量庞大且具高度随机 存取特性，将显著推升对高IOPS企业级SSD的需求。供给端，海外原厂投资不再用于大规模新建晶圆 厂以增加晶圆投入，而是高度集中于HBM和先进制程迁移，NAND投入较为审慎；价格端，预计 NAND Flash合约价有望迎来全面且显著上行，2026Q1合约价涨幅或达55%-60%，并有望延续至2026年 底。 供给层面，产能面临着多重约束：根据TrendForce数据，海外原厂投资不再用于大规模新建晶圆厂以增 加晶圆投入，而是高度集中于HBM和先进制程迁移，NAND投入较为审慎。NAND Flash资本支出预计 从2025年的211亿美元小幅增长至2026年的222亿美元，年增仅约5%。 三星优先满足HBM4生产需求，SK海力士预计于2025Q4投产的M15X晶圆厂专注于生产HBM和先进 DRAM，两家公司均将缩减或限制NAND Flash资本支出，优先将投资转向HBM和DRAM领域；美光资 本开支更专注于先进DRAM，NAN ...

Group 1 - The core viewpoint is that the passenger car industry will continue to exhibit a "stable volume and slow price" logic in 2025, with ASP (Average Selling Price) showing a slight decline in the coming years [1][2] - In December 2025, the domestic passenger car compulsory insurance sales were 2.278 million units, a year-on-year decrease of 16.4%, indicating weaker demand compared to seasonal trends due to consumer hesitation influenced by the suspension of trade-in subsidies [1][2] - The dynamic inventory-to-sales ratio for the passenger car industry as of December 2025 was 2.48, with short-term inventory risks considered low due to expected demand release from continued trade-in policies [3] Group 2 - The passenger car ASP showed a year-on-year increase of 13.7% in December, indicating a positive trend after October, while the overall automotive retail sales decreased by 5.0% [2] - The elasticity of terminal sales for passenger cars in 2026 is expected to be influenced by the extension of scrapping and replacement subsidies, with estimated elasticities of 2.1% and 4.3% respectively [2] - The total inventory of passenger cars was 4.708 million units as of December 2025, with a decrease of 37,000 units in December, reflecting a cautious market environment [3]

Core Insights - The report from GF Securities highlights the significance of the Scale-Up architecture in addressing the challenges posed by training and inference of trillion-parameter models, particularly focusing on the advancements in AI computing power [1][2]. Group 1: Current Trends in AI Computing - The core characteristics of model inference, such as long context, high concurrency, and real-time interaction, are driving the upgrade of Scale-Up systems [2]. - The need for high-performance networks during the inference phase is emphasized, as the architecture allows all accelerators' memory to be presented as a single shared pool, addressing bandwidth and latency requirements [2]. Group 2: Domestic Developments in Scale-Up - Alibaba Cloud introduced the new generation of Panjiu AI Infra2.0 AL128 supernode server, which aims to optimize computing power and communication synergy, achieving a 50% improvement in inference performance compared to traditional architectures [3]. - Tencent's ETH-X supernode project is being developed in two phases, focusing on optimizing GPU and memory communication and exploring full optical interconnection for Scale-Up [3]. - Huawei announced a three-year computing power plan, with the Ascend 950 supernode expected to launch in Q4 2026, indicating a strong commitment to advancing Scale-Up technology [3]. Group 3: Investment Opportunities in Optical Interconnection - Current Scale-Up networks primarily use copper cabling, which has limitations in system design and transmission distance, making it less suitable for future expansions [4]. - The advantages of optical interconnection are highlighted, as it can cover much greater distances compared to copper, which is limited to around 7 meters for certain applications [4].

Group 1 - The core viewpoint of the news is that the stock prices of Huahong Semiconductor have reached historical highs, driven by rising DRAM chip spot prices and expectations of a larger price surge in the storage market [1] - The Hang Seng Technology Index has shown a mixed performance, with semiconductor and AI application sectors performing relatively well, while OTA and smart driving sectors have lagged [1] - The sentiment in the Hong Kong stock market has already reflected negative factors such as US-China trade tensions and the peak of stock unlocks at the end of last year, indicating potential for upward beta recovery in early 2023 if liquidity pressures ease [1] Group 2 - Huaxia Fund suggests positioning for a new round of spring market rally in Hong Kong stocks, recommending low-cost ETF investment tools such as the Hong Kong Stock Connect Technology ETF and the Hang Seng Internet ETF [2] - The recommended ETFs are listed on mainland stock exchanges, allowing for flexible trading without the need for a Hong Kong stock account, and support T+0 trading [3]

智通财经APP获悉，广发证券发布研报称，根据数字水泥网和国家统计局数据统计，预计2024/2025水 泥熟料产能利用率53%/50%，产能严重过剩。经过过去两年时间的政策酝酿和宣导，2026-2027年将迎 来以"超产管控+碳市场"为主导的供给端政策管控，有望带来水泥超产产能和落后产能陆续退出，水泥 供给优化将带来盈利中枢提升。中长期来看，水泥行业供给有望持续优化，建议关注底部水泥资产。 广发证券主要观点如下： 超产管控：目前已有实质性产能出清 2025年9月24日，工信部联合六部门发布《建材行业稳增长工作方案(2025—2026年)》，明确水泥企业 要在2025年底前对超出项目备案的产能制定产能置换方案，促进实际产能与备案产能统一。 根据数字水泥网统计，截至2025年底，全国水泥行业通过置换退出熟料产能合计为1.6亿吨(考虑超产后 估算实际退出约2亿吨)；鉴于2025年版补产能规则公示公告时间延迟至2026/3/31，预计到26Q1末，熟 料产能退出总量预计将达到2亿吨(考虑超产后估算实际退出约2.5亿吨)。即到26Q1末，实际产能有望从 21亿吨降至18亿吨，将带来2026年产能利用率提升。 碳市场：20 ...

Group 1 - The AI + programming sector has significant potential, with rapid commercialization progress observed [1] - DeepSeek is expected to launch its powerful V4 model around mid-February, which will further stimulate market expectations [1] - AI-assisted programming tools like Cursor and GitHub Copilot have shown good commercialization results, indicating a positive trend in the industry [1] Group 2 - Domestic AI-assisted programming tools have a higher proportion of utilizing local AI large models and offer better cost-performance ratios compared to overseas products [2] - Although domestic tools may lag in functionality compared to Cursor, their lower pricing makes them attractive [2] - Improvements in DeepSeek's V4 model capabilities and reductions in computing costs could accelerate the commercialization of AI large model applications, including AI-assisted programming tools [2] Group 3 - Zhuoyi Information has extensive technical accumulation and product refinement experience in the IDE field, with its SnapDevelop tool enhancing development efficiency by 2-3 times compared to traditional methods [3] - The tool aids enterprises in efficiently building modern applications through low-code and intelligent features [3]

责任编辑：栎树 1月27日，金山云(3896.HK)再度拉升涨超4%，高见7.3港元创逾3个月新高价，并且录得5连升行情，总市值324亿港元。 消息上，广发证券发布研报称，金山云拥有高可见度的强劲需求，小米金山加大AI投入，基础大模型、智能汽车、WPS均 需要大量AI训练和推理算力，非小米客户AI需求亦旺盛。参考可比公司估值，分别给予AI云/非AI云业务6/2倍PS，对应合 理价值10.49港元/股。维持"买入"评级。 另外，在上周举办的金山云年度Tech Talk上，金山云高级副总裁刘涛宣布，智算平台金山云星流已完成从资源管理平台向 一站式AI训推全流程平台的战略升级。 港股频道更多独家策划、专家专栏，免费查阅>> ...

金山云(3896.HK)再度拉升涨超4%，高见7.3港元创逾3个月新高价，并且录得5连升行情，总市值324亿港元。 消息上，广发证券发布研报称，金山云拥有高可见度的强劲需求，小米金山加大AI投入，基础大模型、智能汽车、WPS均需要大量AI训练和推理算力，非 小米客户AI需求亦旺盛。参考可比公司估值，分别给予AI云/非AI云业务6/2倍PS，对应合理价值10.49港元/股。维持"买入"评级。 另外，在上周举办的金山云年度Tech Talk上，金山云高级副总裁刘涛宣布，智算平台金山云星流已完成从资源管理平台向一站式AI训推全流程平台的战略升 级。(格隆汇) ...

广发证券主要观点如下： 智通财经APP获悉，广发证券发布研报称，根据英伟达及SemiAnalysis数据，随着GPU架构迭代，单 GPU可配套的HBM与CPU DRAM容量均呈现显著上升趋势。需求端，在不同GPU代际与出货假设下测 算，若VR200出货量达1400万颗，对应DRAM容量需求约14.78EB，AI服务器正成为DRAM需求增长的 最大下游应用。供给端，据TrendForce数据，DRAM行业资本开支预计由2025年的537亿美元增至2026 年的613亿美元，但新增产能集中在2027年及以后释放。 需求端，AI服务器用DRAM为下游最大增量应用环节。根据TrendForce数据，受益于AI服务器部署节奏 加快，预计2026年DRAM需求增速为26%。在不同GPU代际与出货假设下测算，若VR200出货量达1400 万颗，对应DRAM容量需求约14.78EB，AI服务器正成为DRAM需求增长的最大下游应用。 供给端，根据TrendForce数据，DRAM行业资本开支预计由2025年的537亿美元增至2026年的613亿美 元，同比增长约14%，但新增产能集中在2027年及以后释放。具体来看，三星为优先 ...

Group 1 - The food and beverage sector's heavy stockholding ratio decreased to 6.1% in Q4 2025, down 0.3 percentage points from Q2 2025, with an excess allocation of +2.4 percentage points compared to Wind All A [1] - The proportion of active equity funds holding food and beverage stocks fell to 32.3% in Q4 2025, while passive funds increased to 63.3% [1] - The liquor sector has experienced a four-year adjustment period and is expected to see a "valuation + performance" double bottom [1] Group 2 - In the liquor segment, the fund holding ratio slightly decreased, with a total and active equity holding ratio of 5.1% and 2.9% respectively, down 0.4 and 0.3 percentage points [2] - The excess allocation for liquor funds is at +2.7 percentage points for total and +0.5 percentage points for active equity [2] - The holding ratio for consumer goods increased slightly to 0.97% in Q4 2025, although it remains in a low allocation state [2] Group 3 - The holding ratio for individual liquor stocks generally declined, with Kweichow Moutai remaining the only food and beverage company in the top 20 heavy stocks, dropping from second to fourth place [3] - The holding ratio for Shanxi Fenjiu rose to second place among liquor stocks [3] - Some leading consumer goods companies, such as Yili and Anjiexin Foods, saw a noticeable increase in fund holding ratios and the number of funds holding them in Q4 2025 [3]