第一产业 · 长出来的,挖出来的 × AI · 2026-07The Primary Sector · Grown & Mined × AI · Jul 2026

最土的行业,最先无人化——无人矿卡比城市 Robotaxi 早了十几年;原因不浪漫:封闭、高危、招不到人,而会种地的在变老,愿种地的没出生 The earthiest industries went unmanned first — the driverless mine truck beat the city robotaxi by a decade and more; the reason is unromantic: enclosed, hazardous, unhireable — and those who can farm are ageing while those who would were never born

人口沙漏:全职务农劳动力平均超 50 岁、55 岁+占约 1/3、35 岁以下仅 19.2%(B/官方口径);一亩地的算术:主粮现金收益乐观时几百元/亩·季,外出务工月入 4000–6000 元——打工一个月 ≈ 种十亩地一整季(C)。AI 填的不是「优化空间」,是绝对的劳动力真空。The demographic hourglass: full-time farmers average over 50, a third are 55+, and only 19.2% are under 35 (B/official); the one-mu arithmetic: staple-grain cash returns run to a few hundred yuan per mu per season at best, against ¥4,000–6,000 a month in town — one month of wage work ≈ ten mu farmed for a season (C). AI fills not an optimisation margin but an absolute labour vacuum.

总裁判:AI 赢在执行(手),弱在预测(天)——渗透最深的环节全是执行(打药/运矿/分选),预测环节(天气/疫病/地质/地缘)渗透最浅。AI 管得了钻速(+40%),管不了霍尔木兹海峡:第一产业的天花板不在算法,在自然与地缘。The referee’s line: AI wins at execution (the hand) and stays weak at prediction (the sky) — the deepest-penetrated links are all executional (spraying, hauling, sorting) while prediction (weather, disease, geology, geopolitics) stays shallowest. AI can lift the drilling rate (+40%); it cannot manage the Strait of Hormuz: the primary sector’s ceiling is set by nature and geography, not algorithms.

对称双主脊:🌾 长出来的(一季作物七节点:育种→耕播→植保→水肥→收获→养殖→卖粮,时间以「月」计)与 ⛏️ 挖出来的(一座矿七节点:勘探→穿爆→铲装→运输→选矿→冶炼→点价,时间以「年/十年」计);中部以共通规律桥相连,底部合流于「大宗定价漏斗」。姊妹分工:餐桌端→dine,柔性采摘姊妹难题→fashion,冶炼下游→factory。 Twin symmetric spines: 🌾 the grown (one crop season in seven nodes: breeding → tillage → protection → irrigation → harvest → husbandry → selling, clocked in months) and ⛏️ the mined (one mine in seven nodes: exploration → drilling → loading → hauling → beneficiation → smelting → pricing, clocked in years and decades); a bridge of shared laws joins them mid-page, and both drain into the commodity-pricing funnel. Siblings: the table end → dine, the soft-picking twin problem → fashion, smelting’s downstream → factory.

传统节点Traditional

🌾 农业轨The farm spine

⛏️ 矿业轨The mine spine

种子与源头Seeds & sources

定价与金融Pricing & finance

2080→3832台

全球自动驾驶矿卡车队规模,2024 年中→2025 年中(+84%,B)——采矿是自动驾驶增速最快的落地场景;⚠️「中国 2500 台」为另一宽口径,两数不可同框;对照全球在役大型矿卡数万台,拐点已过、存量渗透仍低The global autonomous mine-truck fleet, mid-2024 → mid-2025 (+84%, B) — mining is autonomy’s fastest-growing venue; ⚠️«China’s 2,500» is a wider separate basis, never in one frame; against tens of thousands of large trucks in service, the tipping point has passed while stock penetration stays low

>80%

大疆+极飞合计占中国植保无人机份额(C)——中国大田渗透最深的农业 AI 单环节;「机+服」按亩收费绕过了小农买不起的死结:设备没下乡,服务下乡了DJI plus XAG’s combined share of China’s crop-protection drones (C) — the deepest single farm-AI link in the field; per-mu service pricing routed around the smallholder’s capital wall: the machine never went to the village — the service did

≈45%

良种对增产的贡献率(官方口径,C 易核)——「种子=农业芯片」的数据支撑:价值链最上游、乘数最大、政策最敏感;AI 育种(基因组选择/GASEL/iMAP)是杠杆最高的赛道The share of yield gains attributed to improved seed (official basis, C, easily verified) — the maths behind «seed as agriculture’s silicon»: uppermost in the chain, largest multiplier, most politically sensitive; AI breeding (genomic selection, GASEL, iMAP) is the highest-leverage lane

5.32亿亩

2020 年家庭承包地流转面积(≈35%,两源换算一致 A-)——规模化的地基铺到 1/3;制度进度是农业无人化的真节拍器:无人农机开不进被田埂切碎的「巴掌田」Household contract land under transfer by 2020 (~35%; two sources agree, A-) — the base of scale reaches one-third; institutional pace is farm autonomy’s true metronome: no driverless tractor turns inside a palm-sized, ridge-cut plot

口径警告:本页综合四份深度研究交叉整理(第四份拒答后补交,增量标【补】×6——主体仍以三份为底)。一份离线撰写、全部数字自标「需核实」(其独有数字一律 C:小松累计运量/极飞 IPO/先正达资本事件);一份带 49 条 grounding 引用(B);一份原全球矿卡数最硬但尾部截断——已于 2026-07 重跑为完整收束的重写版(约 20 条引文;⚠️重写版转向定性综述,原版独有的矿卡计数等存量数字维持原分级、不再受其背书);补交的第四份带 ~19 条可点链接,官方政策数字最全(B),公司台数多官网自述(D)——北斗「百万台级」已按其白皮书拆分修正,少数口径(老龄化/亩利润锚)已获第四源印证。⚠️已裁定冲突:全球矿卡趋势采 2080→3832(B),「中国 2500 台」单列并注「口径较宽」;务农平均年龄采「>50 岁」。两个准确率勿混:采摘识别 85%+(采摘环节)≠ 分级视觉 >99%(加工环节)。渗透%为编辑按定性分级折算的口径估计,建议当色阶读。「已商业化」≠高渗透:矿卡拐点已过、存量渗透仅个位数~10%。留白声明:垂直农场/植物工厂、海外精准农业老牌(Trimble/AGCO/CNH)、勘探初创(KoBold 类)、力拓/必和必拓甲方视角——三份报告覆盖不足,本页不作虚构。 Basis warning: cross-compiled from four deep-research reports (the fourth refused at launch and filed late; its six additions carry [added] — the body still rests on three). One wrote offline with every figure self-flagged (its unique numbers capped at C: Komatsu’s cumulative tonnage, XAG’s IPO, Syngenta’s capital events); one carries 49 grounded citations (B); one once held the hardest global truck counts but truncated at the tail — re-run in Jul 2026 as a complete rewrite (~20 citations; ⚠️the rewrite turns qualitative, so its legacy truck counts keep their old grades without fresh backing); the late fourth brings ~19 clickable citations — strong on official policy figures (B), self-reported on company counts (D); the «million-level BeiDou» is now split per its white paper, and a few bases (ageing, the per-mu arithmetic) gained fourth-source confirmation. ⚠️Adjudicated: the global trend runs 2,080→3,832 (B), with «China’s 2,500» listed apart under a wider basis; farming age taken as «over 50». Two accuracies must not mix: picking recognition at 85%+ (harvest) ≠ grading vision at >99% (processing). Penetration percentages are editorial conversions of qualitative tiers — read as shading. «Commercialised» ≠ penetrated: the truck’s tipping point has passed at single-digit stock share. Declared blanks: vertical farms, legacy precision-ag names (Trimble/AGCO/CNH), exploration startups (KoBold-class), and the Rio/BHP owner’s view — undercovered by all three sources, not invented here.

◆ 中心装置 · 为什么最土的最先无人化The central device · why the earthiest went first

2008 年,皮尔巴拉;2020 年代,城市 Robotaxi2008, the Pilbara; the 2020s, the city robotaxi

小松 FrontRunner 无人矿卡 2008 年起在智利与澳洲皮尔巴拉规模运营(C→B-)——比城市自动驾驶出租车早了十余年。同一逻辑正在农田重演:无人农场、智能鱼塘都跑在城市前面。不是技术更强,是场景更宽容。Komatsu’s FrontRunner trucks have run at scale in Chile and the Pilbara since 2008 (C→B-) — a decade-plus ahead of the urban robotaxi. The same logic replays in the fields: unmanned farms and smart fishponds outrun the city. Not stronger technology — a more forgiving scene.

▸ 被迫自动化 · 三重驱动Forced automation · the triple driver

封闭 + 高危 + 招不到人Enclosed + hazardous + unhireable

矿区是封闭场景(无行人、路线固定)、高危环境(粉尘/爆破/塌方)、极度缺人(偏远高薪也招不到)——三者叠加处,自动化不是选择题,是生存底线。农田同构:老龄化+苦脏累+一年一季的孤独。无人化路径全行业一致:辅助驾驶 → 远程遥控 → 去安全员——「1 个人从开 1 台机器,变成在控制室监看 N 台」。The mine is an enclosed scene (no pedestrians, fixed routes), a hazardous one (dust, blasting, collapse) and unhireable (remote, and high pay still fails) — where the three stack, automation is not a choice but the survival floor. The field mirrors it: ageing, drudgery, one lonely season a year. The path is identical everywhere: assisted driving → remote control → safety-officer removal — «one person goes from driving one machine to watching N from a control room».

◆ 一亩地的算术 · 谁还在种The one-mu arithmetic · who still farms

打工一月 ≈ 种十亩一季A month in town ≈ ten mu a season

三大主粮亩均净利润长期低位甚至为负、现金收益乐观时几百元/亩·季(C);外出务工月入 4000–6000 元。「种粮不如打工」是诚实的算术——它同时解释了抛荒、规模化与补贴的存在(种粮不赚钱由财政兜底,粮食安全是公共品)。人口沙漏由此而来:平均 >50 岁、55+ 占 1/3、小学及以下 44.32%(B)。Staple-grain net profit per mu runs low to negative, cash returns a few hundred yuan a season at best (C); town wages pay ¥4,000–6,000 a month. «Farming loses to wage work» is honest arithmetic — and it explains abandonment, consolidation and subsidy in one stroke (the treasury underwrites unprofitable grain because food security is a public good). Hence the hourglass: average age over 50, a third 55+, 44.32% with primary schooling or less (B).

▚ 扩散排序梯子 · 什么时候轮到你这块地/这座矿The diffusion ladder · when your plot or pit comes up

露天矿运输 · 平原大田Open-pit hauling · flat fields 已到arrived→ 工厂化养殖 · 渔业Factory husbandry · fisheries 进行中under way→ 丘陵 · 果园温室 · 井下矿Hills · orchards · underground 5–10 年+yrs+

你的时间表由连片程度、封闭程度、缺人程度三个变量查表可得,不由 AI 进步速度决定——场景三要素＞算法先进度。地基刻度【补】:耕种收综合机械化率已过 3/4(75.64%,小麦 98/玉米 92/水稻 89,官方,B)——无人化长在机械化 3/4 的地基上;而农业生产信息化率目标才 32%(2028)→40%+(2035)(B,年份配对回源可核):机器先到,数据后到。Your schedule reads off three variables — contiguity, enclosure, labour scarcity — not off the pace of AI progress: the scene’s three factors ＞ the algorithm’s sophistication. The base line [added]: mechanisation has passed three-quarters (75.64%; wheat 98, corn 92, rice 89 — official, B) — autonomy grows on that base; the informatisation target sits at 32% (2028) → 40%+ (2035) (B, re-verify pairings): machines arrived first, data comes after.

立场声明:本页是批判性、祛魅的行业结构分析,用 A–D 角标区分交叉印证与离线记忆数字(初版三缺一、第四份后补并入,冲突已裁定并标 ⚠️);渗透%为口径估计。产品指南是市场地图,不构成农事、采购或投资建议。核心判断一句话:最土的行业最先无人化——AI 填的是劳动力真空,赢在执行、弱在预测;而天花板从来不在算法,在自然、制度与地缘。 Stance: a critical, demystifying structural analysis; A–D badges separate cross-verified figures from offline recollections (three-of-four at launch, the fourth folded in since; clashes adjudicated under ⚠️); penetration percentages are basis estimates. The product guide is a market map — not farming, procurement or investment advice. The core judgment in one line: the earthiest industries went unmanned first — AI fills a labour vacuum, winning at execution and weak at prediction; and the ceiling was never the algorithm but nature, institutions and geography.

从这张图带走的五条规律Five patterns to take away