How to quickly understand an industry

How to quickly understand an industry

This report systematically analyzes an industry from eight dimensions: life cycle, feasibility, scalability, defensiveness, profitability, valuation, external factors, and economic climate, and provides a complete industry research report.

Installed by
18
CategoryLearn
FromYouMind
钻石行业·行研八维报告

视角:投资视角 | 市场:中国为主 + 全球联动 | 时间:2026年6月


⚡ 30秒速判

钻石行业正经历百年来最深刻的结构性裂变——不是“钻石还能不能买”的问题,而是“你买的是哪个钻石”的问题。 天然钻石处于成熟期→衰退期过渡(0.5克拉以下价格两年跌超30%,De Beers 连续4年降价,全球产量主动削减至20年低点);培育钻石处于成长期中期(中国零售渗透率已达23.6%,全球珠宝渗透率突破41%,但上游产能过剩刚完成三年出清)。核心赚钱逻辑已从“资源稀缺溢价”切换为“技术+成本+供应链控制力”——中国掌握全球63%的培育钻石产量和95%的工业金刚石产能,2025年10月出口管制更是赋予了战略资源属性。护城河判断:天然钻石护城河正在被侵蚀(变窄),培育钻石上游的护城河在价格战中被冲垮(窄且脆弱),真正的护城河正从“矿”转向“工业应用专利+半导体散热 Know-how”。 竞争格局对天然钻巨头不友好(De Beers 被迫卖掉自己),对培育钻石饰品端极度内卷(A 股三家龙头两家巨亏),唯一确定性高的结构性机会在工业金刚石半导体散热赛道。一句话:投资视角下,钻石行业的珠宝叙事已死,半导体叙事方生;追饰品链性价比低,卡工业金刚石上游才是正解。


一、研究对象与边界界定

横向边界:覆盖天然钻石与培育钻石两大平行品类,两者物理化学性质完全相同(碳元素、立方晶体结构、莫氏硬度10),但生产方式和定价逻辑截然不同——前者依赖矿产开采(地球45亿年地质产物),后者通过 HPHT(高温高压法)或 CVD(化学气相沉积法)在实验室数周内“种”出。

纵向边界:覆盖全产业链三环节 —— ① 上游:毛坯钻石生产(天然开采 / 实验室合成 + 核心设备压机);② 中游:切割加工(全球70%产能集中在印度苏拉特);③ 下游:珠宝零售品牌 + 工业应用(半导体散热、切割工具、光学窗口)。天然钻石和培育钻石共享中游加工和下游零售渠道,但培育钻石额外开辟了工业半导体应用的第二增长曲线。

不纳入本报告:莫桑石、立方氧化锆等钻石仿制品(物理化学性质不同的“假钻石”),彩色宝石(红宝石/蓝宝石/祖母绿)等独立品类。


二、生命周期定位

2.1 核心判断

维度天然钻石培育钻石(珠宝)培育钻石(工业/半导体)
渗透率天然钻石仍是存量主体,但份额持续被侵蚀全球珠宝端渗透率41%+(2025末),中国零售端23.6%(2025)工业金刚石在半导体热管理中渗透率<5%,处于导入期
生命周期阶段成熟期→衰退期过渡成长期中期→向成熟期靠近导入期末→成长期爆发前夜
判断依据产量主动收缩(2025全球约1.05亿克拉,较峰值-40%),价格中枢持续下移,消费场景从婚庆刚需→悦己/投资转型渗透率快速从6.7%→23.6%(4年),CAGR 18%+,但上游价格战已持续3年英伟达/AMD 2026年 AI 服务器首次量产搭载金刚石热沉片,技术验证刚完成

2.2 关键渗透率数据

  • 中国培育钻石零售端渗透率:2021年6.7% → 2025年23.6%,预计2030年突破40%(来源:De Beers《2025年钻石行业洞察报告》)citation

  • 全球培育钻石珠宝渗透率:2019年仅5% → 2025年末41%(来源:中研网培育钻石产业年报)citation

AI Agent行业·行研八维报告(创业视角)

⚡ 30秒速判

AI Agent 行业当前处于导入期末→成长期前夜(企业级渗透率从2024年的32%飙升至2025年的58%,但仍远未饱和)。核心赚钱逻辑已从“卖模型”转向“卖结果”——按任务/按效果付费正在取代传统 SaaS 订阅。最大机会不在通用 Agent(大厂领地),而在垂直行业 Agent——金融合规、医疗诊断辅助、工业质检、跨境电商运营等有真实付费意愿与行业数据壁垒的细分场景。**护城河判断:通用 Agent 环节壁垒窄且正在被巨头侵蚀,垂直行业 Agent 凭借领域 Know-how+客户关系可建立中等壁垒,但整体行业的网络效应尚未形成,属“宽但浅”型。**竞争格局:CR5=45%,集中度仍在分散而非集中(新玩家持续涌入),创业者仍有机会窗口但不超过24个月。最大风险:巨头模型层免费化碾压应用层、89%的 Agent 试点无法规模化(Gartner 2026数据)。创业视角结论:值得做,但必须垂直、必须轻资产、必须12个月内找到可复制的付费场景,切忌做通用“套壳”Agent。


一、研究对象与边界界定

核心定义:AI Agent(智能体)是具备环境感知、自主决策、规划能力、工具调用与行动执行能力的人工智能系统,本质为“大模型 + 规划能力 + 记忆能力 + 行动能力”的融合体 citation。它区别于传统 Copilot(辅助型)的核心在于:Agent 可根据业务目标自动拆解任务、调用外部工具、在动态环境中调整执行路径,实现从“被动应答”到“自主执行”的范式跃迁。

研究对象框定:本报告聚焦 AI Agent 应用层——即基于底层大模型 API 构建、直接面向企业或终端用户的智能体产品与服务。这是我们创业视角最可介入的环节。

纵向边界(产业链)

  • 上游:芯片/算力/云基础设施(NVIDIA/华为昇腾/AWS/阿里云)——创业者不可及

  • 中游:大模型层(OpenAI/DeepSeek/智谱/Kimi/豆包)——创业者极难进入

  • 下游:AI Agent 应用层(Cursor/Lovable/实在智能/垂直行业 Agent)——创业者主战场

横向边界:涵盖通用 Agent 平台、垂直行业 Agent、Agent 开发工具/编排平台(Harness)、Agent 分发与交易市场。不包含上游模型训练及芯片层。

地域/目标:中国市场为主、全球参照;创业视角——关注可介入赛道、启动成本、商业模式可行性、竞争卡位与退出路径。


二、生命周期定位

判断:导入期末→成长期前夜

关键指标数据来源/时点
全球企业 AI Agent 渗透率2024年32% → 2025年58%Gartner, 2025
企业应用内置 Agent 比例2025年<5% → 2026年预计40%Gartner, 2026
智能客服 Agent 渗透率超70%(金融/通信行业超80%)第一新声智库, 2025.9
IDM模式存储芯片行业·行研八维报告

⚡ 30秒速判

全球存储芯片正处于从传统周期品向 AI 结构性成长品切换的成熟期+二次成长拐点。2026年全球存储市场规模预计同比暴涨159%至约5749亿美元(韩华投资证券,2026.02),核心驱动力从消费电子换机彻底切换为 AI 算力基建。IDM 模式下三星、SK 海力士、美光三巨头掌握 DRAM 超90%份额,定价权空前强大——DRAM 价格四年累计暴涨约700%,2026Q1合约价再涨约90%。护城河正在快速变宽:美光 FY2026Q3毛利率已飙至84.6%(史上最高)、SK 海力士营业利润率逼近80%,叠加3-5年长协锁定30%-50%出货量,周期波动被大幅熨平。最大机会在 HBM:SK 海力士独占50-55%份额,2026年 HBM 市场规模预计达546亿美元(YoY+58%),产能售罄至2027年。最大风险:价格失控引发监管反垄断诉讼(三巨头已被集体起诉)+ 2027年新产能释放可能带来阶段性供给过剩 + 中国国产替代加速侵蚀中低端市场。投资视角:头部 IDM 原厂是 AI 时代绕不开的硬卡点,当前5-7倍 forward PE 的韩系标的存在显著重估空间,但短期已积累较大涨幅,需等待回调布局。

一、研究对象与边界界定

研究对象:IDM(Integrated Device Manufacturer,垂直整合制造)模式的存储芯片行业,即集芯片设计、晶圆制造、封装测试于一体的 DRAM 和 NAND Flash 原厂。

横向边界:DRAM(含 HBM 高带宽内存、DDR5/DDR4/LPDDR)和 NAND Flash 两大核心品类,不含 NOR Flash、SRAM 等利基存储。纵向边界:聚焦中游 IDM 原厂环节,兼顾上游设备/材料与下游应用以理解利润分配与议价权格局。满足的需求:为 AI 服务器、数据中心、智能手机、PC、汽车等提供数据暂存(DRAM)与持久存储(NAND)——在 AI 时代,存储已成为制约算力释放的核心瓶颈,“内存墙”问题的解决方案。

默认视角:投资视角 + 全球市场(中国市场为重点增量变量),覆盖三星电子、SK 海力士、美光科技、铠侠、长江存储、长鑫科技等核心 IDM 厂商。

二、生命周期定位

阶段判断:成熟期(向结构性成长期切换的拐点)

渗透率判断依据

DRAM 和 NAND Flash 作为通用数字基础设施,在全球数据设备中的渗透率已接近饱和——智能手机/PC/服务器基本100%配置。按传统周期行业定义,这是一个典型的成熟期行业:CR3高达90%+,产品高度标准化,价格由供需驱动的周期性波动主导。

但2024-2026年发生了一个历史性结构变化:AI 算力需求将存储从“消费电子附属品”重新定义为“AI 基础设施核心组件”。具体数据:

  • 2026年服务器存储营收占比突破56%,首次超越消费电子成为行业基本盘(雪球行业复盘,2026.06)

  • 单台 AI 服务器 DRAM 用量为传统服务器8-10倍,NAND 用量为3-5倍(新浪,2026.05)

  • 2026年全球 AI 服务器出货量预计370万台,YoY+51.3%(行业调研机构)

  • 全球约66% DRAM 产能供给 AI 相关需求,云厂商签订2-3年长协锁产能(雪球,2026.06)

这本质上是在成熟行业内部开出了一个二次成长曲线。因此,估值方法论从纯粹周期股 P/B 框架向成长 P/E 框架切换具备合理性——韩系券商已明确提出“估值从周期 P/B 转向成长 P/E”(SK 证券,2026.05)。

三、八维度逐项分析

1. 生命周期定位总结

维度判断依据

Why we love this skill

This skill is the essence of McKinsey-style industry research. Through "search first, judge later" and an eight-dimensional analysis framework, it ensures that every report is based on the latest data, providing in-depth insights and decision-making recommendations, rather than generalities.

Instructions

The author has set the instructions to private. Below is a brief overview.

Description

📋 User Guide: This Skill is an industry research engine driven by McKinsey methodologies, based on the eight-dimensional framework in Xiao Jing's "How to Quickly Understand an Industry." You tell it an industry name, and it generates a systematic industry research report for you. —————————————————————————————— 🚀 Basic Usage: Simply tell me what industry you want to research, the more specific the better: a. "Please analyze the solid-state battery industry for me" b. "Look at the humanoid robot industry chain from an entrepreneurial perspective" c. "Is the photovoltaic industry still worth investing in?" You can include three pieces of information (not mandatory, I will use the default if missing): 1. Industry Name: The more specific the better, such as "perovskite photovoltaics" rather than "new energy," this is "required"; 2. Goal: Investment/Career Selection/Entrepreneurship/Competitive Analysis/Popular Science, default: Investment; 3. Region: China Market/Global/USA/Southeast Asia..., default: China Market; —————————————————————————————— ⚙️ What will it do automatically? The entire process is divided into four stages: 1. Multiple rounds of online searches to gather evidence of market size/growth rate/penetration rate, industry chain, competitive landscape, moat, policies, and valuation—every conclusion has a source and timeframe, and is never fabricated; 2. Life cycle positioning: using penetration rate to determine whether the industry is in the introduction/growth/maturity/decline stage, with completely different research focuses at each stage; 3. In-depth analysis across eight dimensions: business model → market size → moat (mandatory in-depth analysis of dynamic trends) → competitive landscape (mandatory scoring of each of the five forces) → valuation → PEST analysis → business climate; 4. Anti-consensus testing: distinguishing between market-priced, obvious trends and overlooked real problems, and providing decision-making suggestions; —————————————————————————————— 📊 What will be produced? A professional industry research report in Markdown format is structured as follows: 1. ⚡ 30-Second Quick Assessment — Stage / Core Profitability Logic / Greatest Opportunity / Greatest Risk / One-Sentence Conclusion; 2. Research Subject and Boundary Definition; 3. Life Cycle Positioning (including penetration rate data); 4. Eight-Dimensional Item-by-Item Analysis (Moat includes dynamic trend table and Porter's Five Forces scoring table); 5. Anti-Consensus Test; 6. Conclusions and Decision Recommendations; 7. Key Risk List; 8. Data Sources and Time Explanation; —————————————————————————————— 🔑 Two Core Highlights 1. In-depth Moat Analysis: Not just discussing barrier types, but requiring an answer to "has it widened or narrowed in the past 2-3 years," using market share / gross profit margin / pricing power data; 2. Porter's Five Forces Scoring: All five criteria are assessed, leaving no room for shortcuts, determining who currently has the strongest bargaining power and whether the market structure is favorable or deteriorating for leading companies; —————————————————————————————— ⏱️ This takes approximately 2-4 minutes, as it requires multiple rounds of searching, dimension-by-dimensional analysis, and report writing. —————————————————————————————— 🧩 After the report output is expanded, we can continue with two specialized analyses (you will need to manually confirm these; they won't run automatically): 1. "Dissecting the Chain to Find Bottlenecks": Identifying unavoidable physical bottlenecks in the industry chain during a trend surge, and pinpointing the true beneficiaries; 2. "DCF Valuation": Running a complete discounted cash flow model for specific companies within the industry; —————————————————————————————— ▶️ Want to try? Just tell me an industry name, such as "Please take a look at the AI ​​Agent industry from a startup perspective" or "What's the current situation of the hydrogen energy industry chain?"

Related Skills

View all
Learn

How to quickly understand a company

Enter a company name or stock, and use Charlie Munger's multidimensional thinking model to conduct in-depth investment research and analysis of the stock from the perspective of value investing. SKILL automatically connects to the Internet to mine in-depth information, forces a search before judgment, cross-references two sources, and labels the scope and source. It conducts in-depth analysis according to the seven-dimensional framework of value investing, provides three-scenario valuation and 2×2 quality/price decision, and produces an actionable "In-Depth Analysis of Value Investing" report.

57k
Learn

Three-element analysis: Understanding the paper

Using the three-element decomposition method of identifying key elements, understanding the small loop, and mastering the large loop, this tool automatically or interactively dissects the argumentation framework of any paper. It includes 14 built-in AI Prompt templates, suitable for all academic readers.

7399
Learn

Literature search on the method of peeling onions upside down

Enter a topic phrase, and it automatically breaks down the topic structure using the "onion peeling method" → expands synonyms → generates four-layer search queries (CNKI + WoS/Google Scholar), and provides an execution list and filtering strategy. Applicable to all disciplines, the search queries can be directly pasted and used. 🎯 What pain points does it solve? ❌ Directly entering a whole sentence into the search box → 0 results ❌ Searching for only one keyword → 5000 results with no idea where to start ❌ Not knowing how to break down a topic → No direction in search terms ✅ Enter "Middle-class educational anxiety and shadow education practices" → Output 30+ directly pasteable search queries + a thesaurus + a layered execution list 📐 Core method: The onion peeling method expands from the inside out, from narrow to wide, in three progressively deeper layers—the core layer precisely identifies 4 essential reading articles, the expansion layer finds 48 directly related studies, and the panoramic layer establishes a comprehensive understanding of the field's scope. 🔧 Output includes: • Breakdown of the three elements of topic selection (qualifiers + research unit + research dimension) • 5-8 related concepts for each element + Chinese-English comparison + superordinate/subordinate concepts • First layer: 8 sets of three-element combination search formulas (title → topic, tight → loose) • Second layer: 3 types of two-element pairings (marking core/background uses) • Third layer: Single-element panoramic search formula • Fourth layer: Superordinate concept expansion strategy • 📋 Six-step execution checklist + screening criteria for each layer (highly cited + cutting-edge) 👥 Who is it suitable for? Academic novices, graduate students, young scholars—anyone who needs to systematically conduct literature searches rather than searching keywords based on intuition.

28699

Find your next favorite skill

Explore more curated AI skills for research, creation, and everyday work.

Explore all skills