制造业 · 一件产品的诞生 × AI · 2026-07Manufacturing · The Birth of a Product × AI · Jul 2026

机器换人,换到哪一道工序了?——先吃确定性工序,再吃调度决策,最后才啃非标柔性;而车间转得再快,也救不了被账期拖死的厂 How far down the line has the machine replaced the hand? — certainty-heavy steps first, scheduling decisions second, non-standard dexterity last; and however fast the floor spins, it cannot save a plant strangled by payment terms

因果要摆正:不是机器人先进到要换人,是人先贵到、缺到、难管到要换机器人。制造业年均工资十年翻倍(4.6万→9.7万,C)、人才缺口近 3000 万(B)、「宁送外卖不进厂」——访谈研究把离厂原因挖到制度层:显性是收入与自由,隐性是最低工资模糊、社保压低难转移、休息权不明与年龄性别歧视(B)。剪刀差越过之处,标准工序不上机器人反而是亏钱。Set the causality straight: robots did not get good enough to replace people — people got too expensive, too scarce and too hard to keep first. Manufacturing wages doubled in a decade (¥46k→97k, C), the talent gap nears 30 million (B), and «better delivery than the factory» — interview research digs the exit down to institutions: pay and freedom on the surface; vague minimum wages, suppressed and non-portable social insurance, uncertain rest rights and age-and-gender bias beneath (B). Past the scissors’ crossing, NOT automating a standard step is what loses money.

金融律:制造业一半是金融业——订单质量与账期＞生产效率(四源全数命中,本图第一铁律)。DSO 45 天+、新订单垫资、大客户压账期:资产负债表比损益表更致命;这条暗轨贯穿主脊全部十节点。The finance law: manufacturing is half a finance business — order quality and payment terms ＞ production efficiency (all four sources concur; this map’s first iron rule). DSO past 45 days, new orders demanding upfront cash, key accounts stretching terms: the balance sheet kills before the P&L; this dark rail runs beneath all ten nodes.

主脊是一件产品的十节点制造生命周期(订单→设计→打样→采购→排产→加工装配→质检→仓储物流→维护→售后闭环)+一条金融暗轨。第一筛选轴是制造形态四分法(流水线/小批量/代工/白牌快反)——同一 AI 在四种形态下经济性天差地别。本图以离散制造为主,流程工业(化工/钢铁)另行处理。姊妹分工:画得好看→design,写代码→code,就业大盘→work,原材料上游→farm,需求进厂→shop。 The spine: a product’s ten-node manufacturing lifecycle (order → design → sampling → sourcing → scheduling → machining & assembly → inspection → warehousing → maintenance → after-sales loop) plus a finance dark rail. The first filter is the four manufacturing forms (mass line / small-batch / OEM / white-label fast-reaction) — one AI, four economics. Discrete manufacturing is the subject; process industries are left declared. Siblings: looking good → design, writing code → code, the labour market → work, raw materials → farm, demand’s doorway → shop.

传统节点Traditional

AI / 机器人轨AI / robot track

工业软件 · 数据Industrial software

金融暗轨 · 账期The finance rail

硬骨头 · 死法Bones & deaths

567台/万人

中国制造业机器人密度(IFR WR2025,2024 年数,A)——已超德国与日本;全球平均 162(2023,7 年翻倍)。⚠️与「470 台、全球第 3」是不同年份口径,勿混排;第三种口径更冷:按全部 1.05 亿法人实体的全口径分母折算,密度仅约 167 台/万、低于美国(独立研究 C)——样本决定叙事China’s manufacturing robot density (IFR WR2025, 2024 data, A) — past Germany and Japan; the global mean is 162 (2023, doubled in seven years). ⚠️Don’t mix with the «470, world #3» 2023 basis; a third, colder basis: against the full 105M-legal-entity denominator, density runs ~167/10k, below the US (independent research, C) — the sample writes the story

54.2万台

2024 全球工业机器人年新增安装量(连续第 4 年超 50 万,IFR A);中国占 29.5 万台=全球 54%,在役超 200 万台——机器换人的主战场只有一个;供给侧同样倾斜:中国机器人产业 2024 营收约 334 亿美元、2025 增速 28%,各级补贴超 200 亿美元、另设 20 年期 1,370 亿美元引导基金(智库 B)Global industrial-robot installations in 2024 (a fourth straight year above 500k; IFR, A); China took 295k — 54% of the world — with 2M+ in service: the replacement war has one main theatre; supply tilts the same way — China’s robot industry booked ~$33.4B in 2024, +28% in 2025, on $20B+ in subsidies plus a 20-year $137B guidance fund (think-tank B)

57%

2024 国产机器人中国市场份额——首次超过外资(IFR,A;MIR 2025 口径 54%+);2025 中国首次成为工业机器人净出口国(出口 +48.7%,C)——四大家族的十年霸权翻页Domestic robots’ China market share in 2024 — past foreign brands for the first time (IFR, A; MIR’s 2025 basis: 54%+); in 2025 China turned net exporter (+48.7%, C) — the Big Four’s decade closes a page

≈3000万

2025 年中国制造业十大重点领域人才缺口预测(《制造业人才发展规划指南》,B)——「宁送外卖不进厂」是自动化的第一推动力:缺人,才是换机器人的原因The forecast talent gap across China’s ten key manufacturing fields in 2025 (the national talent plan, B) — «better delivery than the factory» is automation’s prime mover: the shortage, not the robot, drives the swap

口径警告:本页综合四份深度研究交叉整理(一份带可验 URL 为主力;一份全程无联网、其独有数字一律降 C/D;一份引文为重定向链接;一份原文末截断、引文表缺失——已于 2026-07 重跑修订、完整收束且引文恢复,存量分级维持、新增按新源分级)。灯塔工厂 ≠ 制造业平均水平:201 家灯塔(WEF 2025.9)对数百万工厂、占比远低于 0.1%——用灯塔代表中国制造,等于用陆家嘴代表中国收入;灯塔效益%(富士康 −97% 缺陷、+73% 效率等)一律厂商自述 D 级。密度 567=2024 年数,勿与「470/全球第 3」(2023)混排;国产份额 IFR 57% 与 MIR 54%+ 机构口径略异;「义乌 210 万」两文档含义相左,本页禁用;富士康 WEF 认证 8 座 ≠ 自评 80 座。渗透深度%为编辑综合估算。覆盖声明:本图以离散制造为主,流程工业留白;机器安全标准/欧盟 AI Act 工业条款为文档盲区,不作虚构。 Basis warning: cross-compiled from four deep-research reports (one URL-verified as backbone; one fully offline with its unique figures capped at C/D; one citing via redirect links; one once truncated with its reference table lost — re-run in Jul 2026, fully closed with citations restored; legacy grades stand, new figures graded on the new source). Lighthouse factories ≠ the manufacturing average: 201 lighthouses (WEF, Sep 2025) against millions of plants — far below 0.1%; representing Chinese manufacturing with lighthouses is representing Chinese income with Lujiazui. All lighthouse gains (Foxconn’s −97% defects, +73% efficiency) are vendor-reported, grade D. Density 567 is 2024 data — never set beside the «470 / world #3» 2023 basis; IFR’s 57% and MIR’s 54%+ differ by house; the «Yiwu 2.1M» figure means different things in two sources and is banned here; Foxconn’s 8 WEF lighthouses ≠ its 80 self-assessed ones. Penetration percentages are editorial estimates. Scope: discrete manufacturing only, process industries declared out; machine-safety and EU AI Act industrial clauses are source blind spots, not invented here.

◆ 诚实层 · 灯塔与人海The honesty layer · lighthouses & the human sea

橱窗与车间,两者都真The showcase and the shop floor — both are real

读图前先接受一个双峰分布:一头是 201 家灯塔工厂的数量级改善,另一头是数百万人力密集中小厂(规上数控化率约 58%,中小微无权威统计)。富士康同时活在两峰:对外 8 座 WEF 灯塔,对内郑州厂常态约 20 万人、旺季靠数千元返费补员(C)。Accept the twin peaks before reading: at one end, 201 lighthouses with order-of-magnitude gains; at the other, millions of labour-dense small plants (above-scale CNC-isation ~58%; below scale, no authoritative count). Foxconn lives on both peaks at once: eight WEF lighthouses abroad, and a Zhengzhou plant of ~200k that staffs peak season with thousand-yuan signing bonuses (C).

▸ 灯塔峰 · 展示极限The lighthouse peak · the demonstrated limit

201 家,+40% 生产率201 plants, +40% productivity

WEF 2025 批灯塔:劳动生产率 +40%、交付周期 −48%(B);富士康 Ingrasys AI 服务器灯塔自述效率 +73%、缺陷 −97%(D)。灯塔证明的是技术可能性——对供应链形成「不数字化就丢订单」的反向压力;但麦肯锡既当评委又当申报顾问的身份重叠,请一并记入口径(C)。The WEF 2025 cohort: +40% labour productivity, −48% delivery time (B); Foxconn’s Ingrasys AI-server lighthouse self-reports +73% efficiency and −97% defects (D). What lighthouses prove is technical possibility — and a «digitise or lose the order» pressure down the chain; that McKinsey both judges and consults on applications belongs in the same ledger (C).

◆ 人海峰 · 车间现实The human-sea peak · the shop floor

数百万厂,数据不落盘Millions of plants, data never lands

规上关键工序数控化率约 58%(约 2022,C)且只覆盖规上;90%+ 法人是中小企业。车间的核心生产资料仍是经验+关系+现金流:计划员的 Excel、老师傅的耳朵、采购员的酒桌、老板的插单。「流程不改,数据不真,AI 只是贵一点的看板」。Above-scale key-process CNC-isation runs ~58% (~2022, C) and covers only the large; 90%+ of firms are SMEs. The floor’s real means of production remain experience + relationships + cash flow: the planner’s Excel, the master’s ear, the buyer’s banquet, the boss’s rush order. «Unchanged processes, untrue data — AI is just a dearer dashboard.»

▚ 剪刀差 · 全图的经济学引擎The scissors · the map’s economic engine

一条线是人的综合成本(工资十年翻倍+社保+管理+流失),一条线是机器的综合成本(本体降价+集成简化+国产替代:协作机器人从数十万降到数万元级 C)。两线相交之处=经济拐点:焊接/喷涂/搬运/码垛/上下料已越过(回本 ≤18–36 个月即值得认真考虑,G 阈值);柔性装配仍在拐点之外——这就是「同一车间里,已赢与珠峰并存」的原因。判断口诀:先看这道工序在剪刀差哪一侧,再决定上不上机器人——标准侧不上是亏钱,柔性侧硬上也是亏钱。One line is the human’s all-in cost (wages doubled in a decade, plus insurance, management and churn); the other is the machine’s (falling hardware, simpler integration, domestic substitution: cobots from hundreds of thousands to tens of thousands of yuan, C). Where they cross sits the economic tipping point: welding, spraying, handling, palletising and machine-tending have crossed (payback ≤18–36 months merits serious study — the G threshold); flexible assembly still lies beyond it — hence «the already-won and the Everest share one workshop». The rule: check which side of the scissors the step sits on before buying — not automating the standard side loses money; forcing the flexible side loses it too.

立场声明:本页是批判性、祛魅的行业结构分析,用 A–D 角标区分 IFR/WEF/官方披露与厂商自述(灯塔效益%一律 D),口径冲突处标 ⚠️ 并裁定;渗透%为编辑估算。产品指南是市场地图,不构成采购、经营或投资建议。核心判断一句话:AI 进厂的顺序是先确定性、再决策、最后柔性——而物理世界的长尾与人情世界的信用,是车间最后的护城河。 Stance: a critical, demystifying structural analysis; A–D badges separate IFR/WEF/official disclosures from vendor claims (all lighthouse gains graded D), clashes adjudicated under ⚠️; penetration percentages editorial. The product guide is a market map — not procurement, business or investment advice. The core judgment in one line: AI enters the plant in order — certainty first, decisions second, dexterity last — while the physical world’s long tail and the human world’s credit remain the floor’s final moat.

从这张图带走的五条规律Five patterns to take away