China Semiconductors
ChatGPT推动AI芯应用，算力提升终端多点开花
AI chips: ChatGPT opens up more applications as rising computing power boosts wide-spectrum terminal growth
投资要点/Investment Thesis
投资要点/Investment Thesis
ChatGPT有望带动数据快速增长，AI运算贯穿云-边-端。根据IDC预计，全球数据总量预期2026年将超过221,000 exabyte，2021-2026年年复合增长率达到21.2%，其中非结构化数据占每年创建数据超过90%；透过云、边、终端架构，以数据为中心将算力资源前置，在更靠近数据源的地方为用户提供低时延服务。
ChatGPT and AI computing: data spike drives cloud-edge-terminal markets
The world’s total data volume could exceed 221,000 exabytes by 2026E, estimates IDC, expanding at a 21.2% CAGR over 2021-26E. And unstructured data makes up over 90% of the data created each year. Through cloud, edge and terminal architecture, data-centric computing power resources are front-loaded to provide users with low-latency services closer to the data source.
云端芯片：AI模型训练核心，计算参数指数级增长重点受益。人工智能的模型自2012年AlexNet问世以来，模型的深度和广度一直在逐级扩升；大规模预训练模型成了一个新的技术发展趋势。根据TrendForce数据，截至2022年统计AI服务器年出货量占整体服务器比重近1%，随着大型云端业者开始大量投入AI相关的设备建设，AI服务器2022~2026年复合成长率有望达到10.8%。
Cloud chips: these are at the core of AI model training and they are the key beneficiary of computing parameters’ exponential growth. Since AlexNet, the AI model, was launched in 2012, model depth and breadth have expanded gradually and large-scale pre-training models have set a new trend in tech development. AI servers made up nearly 1% of total server shipments in 2022, according to TrendForce data. With large cloud operators starting to invest heavily in the AI equipment buildout, the AI server market could see a CAGR of 10.8% over 2022-26E.
边端芯片：云端与终端的中继站，满足AI轻量化部署需求。根据IDC预测，2025年64%的数据将在传统数据中心之外创建，意味着更智能的处理将在设备上完成，边缘服务器具有体积较小、环境适应性更优、支持多种安装方式、快速前维护和统一管理接口等技术特点。全球边缘AI处理器市场规模近30亿美元，快速响应需求增加有望带动市场规模快速增长。
Edge chips: relay stations between the cloud and computer terminals meet the need for lightweight AI deployment. IDC predicts that 64% of total data in 2025E will be created outside traditional data centers, which means more smart processing done on devices. Tech characteristics of edge servers include a smaller size, better environmental adaptability, support for multiple installation methods, fast front-end maintenance and a unified management interface. The global AI processor edge market amounts to nearly USD3bn and the need for fast responsiveness to demand increases could quickly expand the market.
终端芯片：AIGC应用多点开花，终端AI SoC迎来升级变革。AIGC生成的内容种类越来越丰富，而且内容质量也在显着提升，随着AIGC模型通用化水平和工业化能力提升，有望降低内容生产和交互的门槛和成本。SoC涵盖声音、影像、AI处理，为智能化场景提供完整解决方案，随着终端朝向AI应用发展，SoC成为为智能终端的算力主控。
Terminal chips: with increasing AIGC applications, terminal AI SoCs are undergoing upgrades and changes. The types of content generated by AIGC have gained more variety and content quality has improved quickly. With improving AIGC model generalization and industrialization capabilities, the thresholds and costs of content production and interaction would fall. SoC covers audio, videos and AI processing, providing comprehensive solutions for smart market scenarios. As terminals shift toward AI applications, SoC computing power allows master control over smart terminals.
存储芯片：大数据同步算力提升，AI需求推动市场增长。DRAM方面AI运算需足够的内存带宽，智能化有望带动内存需求提升；NAND方面发展垂直方向堆栈3D NAND层数，新技术注入成长动力。
Memory chips: computing synchronization of big data has improved and demand for AI is driving market growth. In DRAM terms, AI computing needs sufficient memory bandwidth while higher intelligence raises memory demand; in NAND terms, the development of vertically stacked 3D NAND layers spurs growth of new technologies.
投资建议/Investment Ideas
建议关注：
云端AI相关企业：寒武纪、海光信息（天风计算机覆盖）、龙芯中科、紫光国微、复旦微电、安路科技等。
边/终端AI相关企业：瑞芯微、晶晨股份、恒玄科技、全志科技、乐鑫科技、富瀚微、中科蓝讯、炬芯科技、兆易创新、中颖电子、芯海科技等。
存储相关企业：江波龙（天风计算机联合覆盖）、澜起科技、聚辰股份、北京君正、普冉股份、东芯股份、佰维存储等。
AI应用端相关企业：工业富联、大华股份、海康威视等。
Stock focus (names in bullet points)
•Cloud AI: Hygon Information Technology (688041 CH, ACCUMULATE) (covered by the computer desk), Loongson Technology (688047 CH, BUY), Unigroup Guoxin Microelectronics (002049 CH, BUY), Shanghai Fudan Microelectronics Group (688385 CH, BUY).
•Edge and terminal: Rockchip Electronics (603893 CH, BUY), Amlogic (Shanghai) (688099 CH, BUY), Allwinner Technology (300458 CH, BUY), GigaDevice Semiconductor (603986 CH, BUY), Sino Wealth Electronic (300327 CH, BUY), Chipsea Technologies (Shenzhen) (688595 CH, BUY).
•Storage: Shenzhen Longsys Electronics (301308 CH, BUY) (jointly covered by the computer desk), Montage Technology (688008 CH, BUY), Ingenic Semiconductor (300223 CH, BUY), Puya Semiconductor (Shanghai) (688766 CH, BUY).
•AI applications: Foxconn Industrial Internet (601138 CH, BUY), Zhejiang Dahua Technology (002236 CH, BUY).
Other market players: not rated
Cambricon Technologies (688256 CH), Shanghai Anlogic Infotech (688107 CH), Bestechnic (Shanghai) (688608 CH), Espressif Systems (Shanghai) (688018 CH), Shanghai Fullhan Microelectronics (300613 CH), Shenzhen Bluetrum Technology (688332 CH), Actions Technology (688049 CH), Giantec Semiconductor (688123 CH), Dosilicon (688110 CH), Biwin Storage Technology (688525 CH), Hangzhou Hikvision Digital Technology (002415 CH).
风险提示：下游需求不如预期、库存去化不如预期、研发与技术升级不如预期、宏观环境变动带来的风险
Risks include: weaker downstream demand than expected; slower inventory depletion than expected; less-than-expected R&D and technology upgrades; and the impact of macro changes.
Email: equity@tfisec.com
TFI research report website:
(pls scan the QR code)
本文件由天风国际资产管理有限公司（证监会中央编号：ASF056）编制，所载资料可能以若干假设为基础，仅供专业投资者作非商业用途及参考之用途，会因经济、市场及其他情况而随时更改而毋须另行通知。任何媒体、网站或个人未经授权不得转载、链接、转贴或以其他方式复制发表本文件及任何内容。已获授权者，在使用本文件或任何内容时必须注明稿件来源于天风国际集团，并承诺遵守相关法例及一切使用的国际惯例，不为任何非法目的或以任何非法方式使用本文件，违者将依法追究相关法律责任。本文件所引用之数据或资料可能得自第三方，天风国际集团将尽可能确认资料来源之可靠性，但天风国际集团并不对第三方所提供数据或资料之准确性负责。且天风国际集团不会就本文件所载任何资料、预测及／或意见的公平性、准确性、时限性、完整性或正确性，以及任何该等预测及／或意见所依据的基准作出任何明文或暗示的保证、陈述、担保或承诺而负责或承担任何法律责任。本文件中如有类似前瞻性陈述之内容，此等内容或陈述不得视为对任何将来表现之保证，且应注意实际情况或发展可能与该等陈述有重大落差。本文件并非及不应被视为邀约、招揽、邀请、建议买卖任何投资产品或投资决策之依据，亦不应被诠释为专业意见。阅览本文件的人士或在作出任何投资决策前，应完全了解其风险以及有关法律、赋税及会计的特点及后果，并根据个人的情况决定投资是否切合个人的投资目标，以及能否承担有关风险，必要时应寻求适当的专业意见。投资涉及风险。敬请投资者注意，证券及投资的价值可升亦可跌，过往的表现不一定可以预示日后的表现。在若干国家，传阅及分派本文件的方式可能受法律或规例所限制。获取本文件的人士须知悉及遵守该等限制。