「AI 吞噬世界」全景图系列 · 交通运输篇 · TRANSPORTTHE «AI IS EATING THE WORLD» ATLAS · TRANSPORT
AI 先拿走的不是方向盘,
是派单权AI Took the Dispatch First —
the Steering Wheel Last
一次位移 · One Movement · 组织位移完成时,执行位移进行时One Movement · organisation fallen, execution under way
AI 冲击交通运输的顺序出人意料:最先失守的不是方向盘,而是调度室——派单权、定价权、路径权、结算权、数据权已经全部算法化,千万司机骑手成了被算法直接管理的第一代劳动者。而「谁来开车」按场景封闭度缓慢推进:矿山和港口已经无人,高速干线正在闯关,城市街头的 Robotaxi 刚驶出围栏,就撞上了第一堵墙——不是技术,是几百万司机的饭碗。技术时间表和社会时间表,是两张不会同步的表。AI hit transport in an unexpected order: the first thing to fall was not the wheel but the dispatch room — dispatch, pricing, routing, settlement and data are all algorithmic now, and tens of millions of drivers and riders became the first generation of workers managed directly by models. «Who drives» advances slowly, by enclosure: mines and ports already unmanned, highway trunks at the gate, and the city Robotaxi barely out of its fence before hitting the first wall — not technology, but millions of livelihoods. The technical timetable and the social timetable are two clocks that will not synchronise.
本图相信THIS MAP BELIEVES
调度、派单、定价、分拣的算法化是完成时(A/B);Robotaxi 成本下探与局部规模运营是真的(累计 2200 万次,B);按封闭度落地的顺序是全图最稳的预测骨架(四源);8 小时红线与阳光行动说明监管对象已从牌照转向算法本身(B)。The algorithmisation of dispatch, pricing and sorting is the completed tense (A/B); Robotaxi cost declines and local scale are real (22M cumulative rides, B); the enclosure order is this map’s steadiest forecasting spine (all four); and the 8-hour red line plus the transparency campaign show regulation has moved from licences to the algorithm itself (B).
本图不相信THIS MAP DOES NOT BELIEVE
不信「X 年内司机消失」的倒计时——油门在政府脚下,蓄水池不能抽;不信厂商的每公里成本与效率倍数——口径里藏着折旧、安全员和保险(C/D);不信「进了 N 城」=「N 城打得到」——那是 ODD 围栏,不是城市;不信体量叙事——龙头 Robotaxi 一个季度的全无人单量,约等于滴滴两个小时(※编者算术)。Not the «drivers gone in X years» countdown — the throttle sits under the government’s foot, and the buffer pool can’t be drained; not vendor per-km costs or efficiency multiples — depreciation, safety officers and insurance hide in the basis (C/D); not «live in N cities» meaning you can hail one there — that’s an ODD fence, not a city; and not the volume story — the leading Robotaxi’s quarterly driverless rides equal about two hours of DiDi (※editor’s arithmetic).
绿灯 = 完成时(调度/分拣/港口)Green = completed (dispatch, sorting, ports)
黄灯 = 进行时(干线/末端/Robotaxi)Yellow = under way (trunks, last-mile, Robotaxi)
红灯 = 硬骨头(责任/百米/蓄水池)Red = hard bones (liability, 100 m, the pool)
渗透% = 评估值;A–D = 证据分级Penetration = estimates; A–D = evidence grades
9.77 亿单2026-05 全国网约车月订单,聚合平台占约三成(交通运输部监管月报 · A)——被算法调度的人流体量National ride-hailing orders in May 2026, aggregators near a third (MOT monthly report · A) — the human flow under algorithmic dispatch
2200 万次萝卜快跑累计出行(至 2026-04,百度 IR · B;累计含早期带安全员期)——无人化的真实进度Apollo Go cumulative rides (to Apr 2026, Baidu IR · B; includes early safety-officer era) — unmanning’s true odometer
23.3 → 2.1 元Robotaxi 每公里成本 2019→2026(C · 末值为预期,口径不明)——黄金交叉的逼近速度Robotaxi cost per km, 2019→2026 (C · final value a forecast, basis unclear) — the golden cross approaching
62.8%网约车司机中家庭唯一就业者占比(新就业形态研究中心 2025 · B)——就业缓冲垫的重量Ride-hailing drivers who are their family’s sole earner (2025 survey · B) — the weight of the employment buffer
⚠️ 口径裁判(先读):① 9.77 亿=监管系统订单信息,与平台自报不同盘;② 2200 万次=累计口径(含早期带安全员),与「320 万次/季全无人」分开写;③ 成本曲线口径不明(折旧/远程安全员/保险是否计入),2.1 元为预期非实测,两份来源同数疑同源;④ 62.8%/77%/7623 元=受访样本非普查;⑤ 司机规模全用「量级」:驾驶员证 750–800 万本≠在岗数,卡车司机 1700–3000 万随口径漂移,骑手「单平台 600 万」与「行业千万级」勿混;⑥ 抽成 30%/27%/14–18.8% 为公示帽/承诺帽/平台自报均值三口径,勿混写;⑦ 武汉「盈利」=百度 2024 年预期(D),四份来源均无 2026 复核;⑧ 渗透%为编者评估值;⑨ 原一份来源句中截断且引文列表丢失(其独有数字曾降 C)——已于 2026-07 重跑修订、完整收束且引文恢复(存量降级维持,新增数字按新源分级);另一份自认离线且明示转录他源数字(不作双源加分)。⚠️ The basis rulings (read first): ① 9.77B comes from the regulator’s system, a different pot from platform self-reports; ② 22M is a cumulative basis (early safety-officer rides included), kept apart from «3.2M driverless per quarter»; ③ the cost curve’s basis is unclear (depreciation, remote officers, insurance), the 2.1 is a forecast, not a measurement, and the two sources carrying it are likely one; ④ 62.8%/77%/¥7,623 are survey samples, not a census; ⑤ driver counts stay at magnitudes: 7.5–8M licences ≠ active drivers, truckers drift between 17M and 30M by basis, and one platform’s 6M riders never equals the industry’s tens of millions; ⑥ 30%/27%/14–18.8% are the posted cap, the pledged cap and platform-reported averages — three bases, never merged; ⑦ Wuhan «profitability» is Baidu’s 2024 forecast (D), unverified by any source in 2026; ⑧ penetration %s are editorial estimates; ⑨ one source once truncated mid-sentence losing its references (unique figures were capped C) — re-run in Jul 2026, fully closed with citations restored (legacy caps stand, new figures graded on the new source); another wrote offline transcribing numbers from a sibling (no double-source credit).
中心装置 · 一次位移与封闭度阶梯THE CENTRAL DEVICE · ONE MOVEMENT & THE ENCLOSURE LADDER
方向盘还在人手里,方向早就不在了The wheel is still in human hands; the direction left long ago
本图的最小分析单元是一次位移:把一单位人或货从 A 移到 B,并完成交接与计费。这个行业的三条天性决定一切判断:时空不可库存(空驶即蒸发,所以调度是全行业最高价值的算法)、强网络效应(天然平台化、寡头化)、强监管重资产担责重(技术就绪≠可以上路)。于是 AI 的进攻分成两场:组织位移(调度室)已经打完,执行位移(驾驶座)按封闭度一格一格推进。This map’s atomic unit is one movement: one unit of person or freight from A to B, with handover and billing complete. Three native traits of the trade decide every judgment: time-space cannot be inventoried (an empty seat evaporates — which makes dispatch the industry’s highest-value algorithm), network effects run strong (native platformisation and oligopoly), and heavy regulation, heavy assets, heavy liability (technical readiness ≠ road permission). So AI’s offensive split in two: the organisational movement (the dispatch room) is already won; the executional movement (the driver’s seat) advances cell by cell, by enclosure.
组织位移 · 完成时 🟢Organisation · completed 🟢
派单、定价、路径、结算、数据五权全部算法化——平台同时握需求入口、供给排序、价格信号与结算规则,劳动者隔着黑箱。Dispatch, pricing, routing, settlement, data — all five powers algorithmic; the platform holds demand, supply ranking, price signals and settlement rules at once, with the worker behind a black box.
执行位移 · 进行时 🟡Execution · under way 🟡
分拣仓内 ~80%、港口 ~70%、干线 ~15%、末端 ~5%、城市 Robotaxi <1%——全图分化最大的一段,按封闭度排队。Sorting ~80%, ports ~70%, trunks ~15%, last-mile ~5%, city Robotaxi <1% — the map’s widest divergence, queued by enclosure.
社会账本 · 压舱层 🔴The social ledger 🔴
网约车不是零工经济,是就业缓冲垫:62.8% 的司机是家庭唯一就业者——政策必答题不是「技术成熟了吗」,而是「被替代的人去哪」。Ride-hailing is not a gig economy but an employment buffer: 62.8% of drivers are their family’s sole earner — the policy question is never «is the tech ready» but «where do the replaced go».
封闭度阶梯 · 全页最稳的预测骨架(越往下越开放,越开放越慢)THE ENCLOSURE LADDER · THE PAGE’S STEADIEST FORECAST (THE MORE OPEN, THE SLOWER)
矿山 · 港口 · 园区Mines, ports, campuses
封闭·已商业化:小松 AHS 自 2008,常态化去安全员——比城市自动驾驶早十余年(互链 farm/build)。Enclosed, commercial: Komatsu’s AHS since 2008, safety officers routinely removed — a decade ahead of the city (cross-linked to farm/build).
高速干线Highway trunks
半封闭·规模化拐点:2026-07 京东×重汽×嬴彻 L4 重卡示范(31 公里/满载 43 吨,B);现实形态=hub-to-hub 人机接力。Semi-enclosed, at the scaling gate: the July 2026 JD×Sinotruk×Inceptio L4 demo (31 km, 43 t laden, B); the real shape is hub-to-hub human-machine relay.
城配轻卡Urban light trucks
进行时:路线半固定、路况混杂——比干线开放一档,比末端封闭一档。Under way: semi-fixed routes in mixed traffic — one notch opener than trunks, one notch tighter than the kerb.
末端无人配送Last-mile delivery pods
小批量示范期(新石器/九识/白犀牛);最后一百米是红灯:上楼、门禁、老人、活鲜——TomTom 拆成找地址/找车位/进楼/找对门四个卡点,每个都能吃掉干线省下的分钟(B);单票几分钱利润养不起几万元的机器人。低空支线初露:济南 7 公里航线 10 分钟(较地面提效 60%+)、阳澄湖大闸蟹 5 分钟出湖(B)——但分级空域准入与审批通道还在建,规模化是政策件不是技术件。Small-batch pilots (Neolix, Jiushi, White Rhino); the last hundred metres stays red: stairs, entry systems, the elderly, live goods — TomTom splits it into four choke points (address, parking, entry, the right door), each able to eat the minutes the trunk saved (B); a few fen of margin per parcel cannot feed a robot costing tens of thousands. A low-altitude branch appears: Jinan’s 7-km air route in 10 minutes (60%+ faster than ground), Yangcheng Lake crabs ashore in 5 (B) — yet tiered airspace access and approval lanes are still being built: scaling is a policy artefact, not a technical one.
城市 Robotaxi(ODD 内)City Robotaxi (inside the ODD)
围栏内进行时:累计 2200 万次(B),但「进了 N 城」≠「N 城打得到」——围栏内是城市,围栏外是新闻稿。Under way inside the fence: 22M cumulative rides (B), yet «live in N cities» ≠ hailable there — inside the fence is a city, outside it a press release.
全城开放 · 航空铁路全责任Open city · aviation & rail, full liability
红灯:长尾无穷+责任链极长;航空铁路的安全冗余要求可解释可审计——黑箱不能出庭;完全无人客机在可预见未来不具现实性。Red: an infinite long tail and the longest liability chain; aviation and rail demand explainable, auditable systems — a black box cannot take the stand; the fully unmanned airliner stays unreal for any foreseeable future.
祛魅算术※:萝卜快跑一个季度的全无人订单(320 万次)≈ 滴滴中国出行两个小时的单量(日均 3940 万单)——无人化是真的,「替代」还早得很;两组数字均出自 IR(B),对比口径已注明。The demystifying arithmetic※: Apollo Go’s quarterly driverless rides (3.2M) ≈ two hours of DiDi’s China mobility volume (39.4M orders a day) — the unmanning is real, the «replacement» is far; both figures from IR filings (B), comparison basis noted.
Reading the MapReading the Map
从这张图带走的五条规律Five patterns to take away
立场声明:本页是批判性、祛魅的行业结构分析——四份深度研究交叉整理(原截断源已于 2026-07 重跑补全、引文恢复,存量降级维持;一份自认离线且明示转录他源数字,已声明并降级;框架共识部分源于委托预设论纲,已如实声明);厂商成本与效率数字一律 C/D 级,渗透%为评估值。本页提供行业结构信息,不构成职业、投资或采购建议;对司机骑手不猎奇不俯视——被算法化的是调度,不是人的价值;「比特位移」(运营商与算力网络)按主题裁剪未展开,素材留档待未来立图。核心判断一句话:AI 先拿走派单权,后拿方向盘——而油门,踩在政府脚下。
Stance: a critical, demystifying structural analysis — cross-compiled from four deep-research reports (the once-truncated source re-run and completed in Jul 2026 with citations restored, legacy downgrades kept; one offline and openly transcribing a sibling’s numbers — declared and downgraded; where the four converge partly echoes the commissioning brief, declared honestly); vendor cost and efficiency figures are C/D throughout, penetration %s are estimates. Industry-structure information only — not career, investment or procurement advice; no voyeurism toward drivers and riders — what got algorithmised is the dispatch, never the person’s worth; the «bit movement» (carriers and compute networks) is trimmed by scope, its material archived for a future map. The core judgment in one line: AI took the dispatch first and the wheel last — and the throttle sits under the government’s foot.