基于 Metix AI 8.6 亿+ 全球人才库，对 36 家北美自动驾驶与智能驾驶公司做全量检索与逐档画像，叠加 2025-2026 行业公开信息深度调研，系统梳理这一领域的华人算法工程师与技术管理者全集。Drawing on Metix AI's 860 million+ global talent pool, we ran a full census and profile-by-profile mapping of 36 North-American autonomous- and assisted-driving companies, layered with deep 2025-2026 public-source research, to systematically chart the full universe of ethnic-Chinese algorithm engineers and technical leaders in this field.
以下数字为 Metix AI 数据库数据库口径（数据时点约 2026 年上半年），指「当前在职于目标公司、坐标美国或加拿大、从事智驾算法及相关软件职能」的人才。The figures below follow Metix AI database scope (data as of roughly H1 2026), counting talent “currently employed at a target company, based in the US or Canada, working in AV algorithms and related software functions.”
目标公司算法池中华人占比 22.8%，远高于华人在美国 STEM 从业者中的整体比例。Waymo、NVIDIA (汽车)、Zoox 三家聚集了 45% 的华人智驾算法人才，头部集中度高。Ethnic-Chinese make up 22.8% of the target companies' algorithm pool, far above the overall ethnic-Chinese share of the US STEM workforce. Waymo, NVIDIA (Automotive) and Zoox together account for 45% of ethnic-Chinese AV algorithm talent — a high degree of concentration at the top.
60.7% 的华人智驾算法人才坐标湾区。西雅图、南加州、密歇根、匹兹堡、奥斯汀构成第二梯队，但单极体量均不足湾区的零头，跨城寻访必须以湾区为主战场。60.7% of ethnic-Chinese AV algorithm talent is based in the Bay Area. Seattle, Southern California, Michigan, Pittsburgh and Austin form the second tier, but no single one comes close to even a fraction of the Bay Area — cross-city sourcing must treat the Bay Area as the main battleground.
本科主力来自清华、上交、浙大、中科大等 C9 院校，研究生集中于 CMU、UMich、Stanford、Berkeley 等。华人池 PhD 占比 25.9%（全池 15.4%）。顶会论文与校友脉络是理解这一人才群体的重要线索。The undergraduate core comes from C9 schools such as Tsinghua, SJTU, Zhejiang University and USTC, with graduate degrees concentrated at CMU, UMich, Stanford and Berkeley. PhDs make up 25.9% of the ethnic-Chinese pool (vs. 15.4% for the full pool). Top-conference papers and alumni networks are key threads for understanding this talent group.
出清结束后行业重新大举吸纳（详见 3.8）；Waymo 在职华人 39.3% 已过 4 年 vesting cliff；H-1B 10 万美元附加费 2026-06-08 被联邦法院撤销。多重信号叠加，2026 年下半年是这一人才市场流动性显著回升的阶段。After the shakeout ended, the industry resumed large-scale hiring (see 3.8); 39.3% of Waymo's current ethnic-Chinese employees are past the 4-year vesting cliff; the $100,000 H-1B surcharge was struck down by a federal court on 2026-06-08. With these signals stacking up, H2 2026 is the phase where liquidity in this talent market rebounds markedly.
以下基于 2025-2026 公开信源逐条核实，只保留影响人才决策的事实。The following is verified point by point against 2025-2026 public sources, keeping only facts that affect talent decisions.
Waymo 2026-03 达每周 50 万次付费行程、覆盖 10 城，2026-02 完成 160 亿美元融资（估值 1,260 亿美元）；Tesla 2026-01 起在 Austin 提供无监督 Robotaxi 并扩至 Dallas/Houston，FSD 模型拟从约 1B 参数扩到 10B；Zoox 在 Las Vegas/SF 累计载客 35 万+ 人次、待收费审批；Nuro 绑定 Uber-Lucid（至少 3.5 万辆承诺）。需求侧从「研究突破」转向「长尾场景攻坚 + 数据引擎效率」，资深算法人才依旧是最硬的通货。Waymo reached 500,000 paid trips per week across 10 cities in 2026-03, and closed a $16 billion round in 2026-02 (valuation $126 billion); Tesla began offering unsupervised Robotaxi in Austin from 2026-01 and expanded to Dallas/Houston, with the FSD model set to scale from roughly 1B to 10B parameters; Zoox has carried 350,000+ cumulative riders in Las Vegas/SF and awaits fare approval; Nuro is locked into the Uber-Lucid deal (at least 35,000 vehicles committed). Demand has shifted from “research breakthroughs” to “long-tail scenario grind + data-engine efficiency,” and senior algorithm talent remains the hardest currency.
Cruise 2024-12 关停、2025-02 裁撤约 50%（约千人），80%+ 工程并入 GM Super Cruise；Motional 2024 裁 40% 后于 2026-01 以 AI-first 路线重启；TuSimple 退市转型 CreateAI；Plus 的 SPAC 上市计划于 2026-04 终止；Torc 关闭两基地收缩至 DFW + Ann Arbor。出清释放的工程师大部分已被吸收，但「再被关停一次」的心理阴影让这批人对外部机会保持开放。Cruise shut down in 2024-12 and cut roughly 50% (about a thousand people) in 2025-02, with 80%+ of engineering folded into GM Super Cruise; Motional, after a 40% cut in 2024, relaunched on an AI-first path in 2026-01; TuSimple delisted and pivoted to CreateAI; Plus's SPAC listing plan was terminated in 2026-04; Torc closed two sites and contracted to DFW + Ann Arbor. Most of the engineers freed up by the shakeout have already been absorbed, but the psychological shadow of “getting shut down once more” keeps this cohort open to external opportunities.
NVIDIA 开源 32B 推理型 VLA（Alpamayo 2 Super，2026-06）；PlusAI 把 10B VLA 蒸馏为 0.5B 车端模型；Motional 用端到端重写全栈；Wayve 以纯 E2E 路线拿下 12 亿美元 D 轮（估值 86 亿，Nissan 量产绑定）。「会大模型的智驾人」与「会智驾的大模型人」同时变稀缺，传统模块化背景的工程师正经历向基础模型方向的能力迁移。NVIDIA open-sourced a 32B reasoning VLA (Alpamayo 2 Super, 2026-06); PlusAI distilled a 10B VLA into a 0.5B on-vehicle model; Motional rewrote its full stack end-to-end; Wayve raised a $1.2 billion Series D on a pure-E2E path (valuation $8.6 billion, Nissan production tie-up). “AV people who know large models” and “large-model people who know AV” are both becoming scarce at once, and engineers from a traditional modular background are undergoing a capability shift toward foundation models.
前 Cruise 创始人的 The Bot Company 冲 40 亿美元估值；前 Waymo 团队创办 Bedrock Robotics（8,000 万美元）；Physical Intelligence 估值 56 亿；OpenAI 重启 robotics 招聘；Tesla 内部 Autopilot 与 Optimus 已统归一人。AV 感知/规划技能向机器人迁移的路径已被验证，从业者的职业选择面随之扩大，行业整体流动性上升。The Bot Company, from a former Cruise founder, is chasing a $4 billion valuation; a former Waymo team founded Bedrock Robotics ($80 million); Physical Intelligence is valued at $5.6 billion; OpenAI has restarted robotics hiring; inside Tesla, Autopilot and Optimus now report to one person. The path for migrating AV perception/planning skills to robotics has been validated, broadening practitioners' career options and lifting overall industry liquidity.
Commerce 关联车辆规则（2025-01 最终规则）自 MY2027 起禁止中国开发的智驾软件进入美国市场，小马智行/文远知行在美商业化路径被锁死，扩张重心转向中东/东南亚/欧洲，美国团队转纯研发职能。但规则不禁止研发中心：理想 2025 年新建硅谷 AI 实验室，小鹏北美团队约百人并获追加算力，蔚来保留湾区团队。结果是湾区出现「中美公司同场竞价同一批华人算法人才」的新格局，本土与中国系雇主在同一人才池中展开竞争。The Commerce Department's Connected Vehicle rule (final rule 2025-01) bars China-developed AV software from the US market starting MY2027, locking down the US commercialization path for Pony.ai and WeRide and shifting their expansion focus to the Middle East/Southeast Asia/Europe while their US teams pivot to pure R&D. But the rule does not bar R&D centers: Li Auto built a new Silicon Valley AI lab in 2025, XPeng's North-American team of about a hundred received additional compute, and NIO retained its Bay Area team. The result is a new dynamic in the Bay Area where “US and Chinese companies bid for the same pool of ethnic-Chinese algorithm talent side by side,” with domestic and China-system employers competing in the same talent pool.
以下全部图表的统计对象 = 2,693 名华人智驾算法人才（高置信 2,566 + 中置信 127）。All charts below count the same 2,693 ethnic-Chinese AV algorithm professionals (2,566 high-confidence + 127 medium-confidence).
读数：千人级体量且浓度高于均值的只有 Waymo（34.2%）与 NVIDIA（31.2%）两家，是「又大又浓」的双极；Nuro、Cruise 关停后群体、Tesla 在数百人体量上同样高于均值；Zoox 体量第二大但浓度（19.4%）略低于均值。左上的小鹏美研、Pony 美研、AutoX 是「小而极浓」，单一群体即高度集中。右下的大池低浓公司（GM/Ford/Lucid/Qualcomm）是规模型人才来源，左上高浓小池则是高密度的细分聚集。Read: the only two firms at thousand-person scale and above-average density are Waymo (34.2%) and NVIDIA (31.2%) — the “big and dense” twin poles; Nuro, the Cruise post-shutdown cohort and Tesla are likewise above average at several-hundred-person scale; Zoox is second-largest by size but slightly below average in density (19.4%). In the upper left, XPeng US R&D, Pony US R&D and AutoX are “small and extremely dense,” with a single group highly concentrated. The large-pool, low-density firms in the lower right (GM/Ford/Lucid/Qualcomm) are scale-type talent sources, while the high-density small pools in the upper left are dense niche clusters.
读数：旧金山市区高度集中（Waymo/Zoox/Cruise 系聚集），Mountain View 是 Waymo 大本营，Santa Clara/San Jose 对应 NVIDIA 与小鹏美研，Fremont 对应 Pony.ai 与 Tesla 工厂圈。这一分布勾勒出湾区智驾人才的地理重心。Read: downtown San Francisco is highly concentrated (Waymo/Zoox/Cruise clusters), Mountain View is Waymo's home base, Santa Clara/San Jose map to NVIDIA and XPeng US R&D, and Fremont maps to the Pony.ai and Tesla-factory orbit. This distribution outlines the geographic center of gravity of Bay Area AV talent.
读数：感知与规划控制两大传统模块占据存量大头，也是当前向端到端范式迁移最活跃的群体；明确从事端到端/基础模型方向者仍是少数，是各家共同争抢的稀缺画像；仿真评估与数据引擎是 robotaxi 规模化运营的放大器职能，需求旺盛而供给最为分散。Read: the two traditional modules — perception and planning-control — hold the bulk of the stock and are also the cohorts most actively migrating to the end-to-end paradigm; those explicitly working on end-to-end/foundation models remain a minority and are the scarce profile everyone competes for; simulation-evaluation and the data engine are amplifier functions for robotaxi scale operations — in strong demand but with the most fragmented supply.
读数：Waymo 是唯一各职能厚度均匀的公司；NVIDIA 集中在感知、ML 综合与数据引擎，规控相对偏薄（平台属性所致）；Zoox 的规划与控制是全行业最厚的单一职能；小鹏美研高度集中在端到端/基础模型，与其「世界基座模型」公开叙事吻合。Read: Waymo is the only company with even thickness across every function; NVIDIA concentrates in perception, ML general and the data engine, relatively thin on planning-control (a function of its platform nature); Zoox's planning and control is the single thickest function in the whole industry; XPeng US R&D is highly concentrated in end-to-end/foundation models, consistent with its public “world foundation model” narrative.
读数：Staff/Principal 级 339 人与带队管理层 305 人构成能独立带方向的资深骨干层；中级及以下体量最大，是行业的人才基本盘。注意 title 通胀差异：Tesla 不设 Staff 序列、NVIDIA 的 Senior 含金量横跨多级，跨公司比较需用年限与作品校准。Read: the 339 Staff/Principal-level engineers and 305 people-leading managers form the senior backbone able to own a direction independently; mid-level and below is the largest band and the industry's talent base. Note title-inflation differences: Tesla has no Staff track, and NVIDIA's “Senior” spans several levels in substance — cross-company comparison needs to be calibrated against years of experience and body of work.
读数：「国内 C9 本科 + 北美 Top CS/Robotics 硕博」是绝对主流路径，华人池 PhD 占比 25.9%（全池 15.4%）。寻访动作上，清华/上交/浙大/中科大校友会、CMU RI 与 UMich 系校友圈、顶会（CVPR/ICRA/CoRL/NeurIPS）华人作者网络，共同构成这一人才群体的学术脉络。Read: “C9 undergraduate in China + top North-American CS/Robotics master's or PhD” is the absolutely dominant path, and PhDs make up 25.9% of the ethnic-Chinese pool (vs. 15.4% for the full pool). For sourcing, the Tsinghua/SJTU/Zhejiang/USTC alumni associations, the CMU RI and UMich alumni circles, and the ethnic-Chinese author networks at top conferences (CVPR/ICRA/CoRL/NeurIPS) together form this group's academic lineage.
读数：45.5% 的华人智驾算法人才有中国教育或中国科技公司履历，意味着相当一部分人同时具备中美两地的职业选择，是本土与中国系雇主（如理想/小鹏硅谷实验室）共同关注的群体。Cruise/Apple/Argo 等已关停项目的前员工如今散布在各家公司，构成一条清晰的人才再流动脉络。Read: 45.5% of ethnic-Chinese AV algorithm talent has Chinese education or a Chinese tech-company background, meaning a sizable share has career options on both sides of the Pacific — a group of common interest to both domestic and China-system employers (such as the Li Auto/XPeng Silicon Valley labs). Former employees of already-shut-down projects like Cruise/Apple/Argo are now scattered across companies, forming a clear thread of talent re-circulation.
读数：三条结构性来源清晰可见。① 高校与科研院所是最大单一来源（NVIDIA 78 人、Waymo 55 人、Zoox 28 人），校招与实习转正是华人进入智驾的第一入口；② Google 系向 Waymo 输送 108 人，集团内部转岗是其最主要的人才通道；③ Cruise 关停后 24 人流向 Waymo（另见前 Cruise 去向图），具中国大厂/新势力背景者 34 人进入 NVIDIA，跨太平洋履历在平台型公司接受度最高。Read: three structural sources are clearly visible. ① Universities and research institutes are the single largest source (NVIDIA 78, Waymo 55, Zoox 28) — campus hiring and intern conversion are the first entry point for ethnic-Chinese into AV; ② the Google system feeds 108 people into Waymo, with intra-group transfer its primary talent channel; ③ post-shutdown, 24 people moved from Cruise to Waymo (see also the former-Cruise destinations chart), while 34 with a Chinese tech-giant/EV-startup background entered NVIDIA — trans-Pacific résumés are most readily accepted at platform-type companies.
读数：2025 年入职 610 人，是数据库口径下的历史最高，验证了「出清结束、重启吸纳」的判断（Waymo 105 人、NVIDIA 89 人居前）；2023 年（行业低谷）仅 367 人；Cruise 线 2024 年后归零；小鹏美研 2024-2025 两年合计入职 69 人，中国系硅谷加码在数据上坐实；Tesla 2025 年 36 人，对应 Robotaxi 冲刺扩编。Read: 2025 saw 610 hires, the highest on record under database scope, confirming the “shakeout over, hiring restarted” read (Waymo 105, NVIDIA 89 in the lead); 2023 (the industry trough) had only 367; the Cruise line goes to zero after 2024; XPeng US R&D hired 69 in 2024-2025 combined, confirming the China-system Silicon Valley buildup in the data; Tesla's 36 in 2025 corresponds to its Robotaxi sprint expansion.
读数：右上是「资深存量区」，Waymo（任期中位 38 个月，39% 已满 4 年）和 NVIDIA（35 个月，36%）沉淀了全行业最厚的资深员工层，多数人初始股票已归属完毕；左下是「新军区」，Applied（中位仅 14 个月）、小鹏美研（19 个月）、Rivian（18 个月）处于快速扩张期，团队普遍年轻；Zoox 居中（26 个月），2024 年入职的大批新人任期将在 2027 年集中满四年；右上孤点小马美研（中位 61 个月、62% 过 4 年）印证了「收缩后留下的多是资深员工」。Read: the upper right is the “senior-stock zone” — Waymo (median tenure 38 months, 39% past 4 years) and NVIDIA (35 months, 36%) have accumulated the industry's thickest senior layer, with most people's initial equity fully vested; the lower left is the “new-recruit zone” — Applied (median just 14 months), XPeng US R&D (19 months) and Rivian (18 months) are in rapid expansion with generally young teams; Zoox sits in the middle (26 months), and the large 2024 hiring class will collectively hit four years in 2027; the lone upper-right outlier Pony US R&D (median 61 months, 62% past 4 years) confirms that “those left after the contraction are mostly senior staff.”
2024 年起的净流动：流入 = 2024 年起入职并在职者，流出 = 2024 年起离开者（含已回国/转行）。净值作方向参考，目的地结构与回国率是更稳健的读法。Net flow since 2024: inflow = those hired since 2024 and still employed; outflow = those who left since 2024 (including those who returned to China/changed fields). Treat the net value as a directional reference; the destination mix and return-to-China rate are the more robust reads.
读数：Waymo 流入 249 人、可见流出 167 人，是行业唯一的强净流入极；Cruise 是最大净流出源（关停清退）；小马/文远美研进出规模都较小，人员相对稳定。注意流出侧数值偏保守，但方向排序可信。Read: Waymo, with 249 inflows and 167 visible outflows, is the industry's only strong net-inflow pole; Cruise is the largest net-outflow source (shutdown separations); Pony/WeRide US R&D have small in/out volumes and relatively stable headcount. Note that the outflow side runs conservative, but the directional ordering is reliable.
读数（Metix AI 数据库口径）：① 关停人才被行业回收：Cruise 华人离职者前五去向为 Meta、Waymo、NVIDIA、Zoox 与小鹏；TuSimple 系约 25 人汇入 Bot Auto，另有约 32 人进入 TikTok/字节系。② 前沿 AI Lab 的吸引力显现：Waymo 华人流出中 OpenAI 与 DeepMind 合计约 30 人，叠加 Meta，研究序列的流向清晰。③ 新生力量：新晋 robotaxi 公司 Tensor Auto 已吸纳约 10 名来自 Waymo 的华人工程师。④ 可见回国约 57 人（占样本约 1.6%，回国者常停更档案，实际偏高），绝大多数人留在北美生态内流动，湾区仍是这一人才市场的绝对中心。Read (Metix AI database scope): ① shutdown talent gets recycled by the industry: the top five destinations for Cruise's ethnic-Chinese leavers are Meta, Waymo, NVIDIA, Zoox and XPeng; about 25 from the TuSimple system flowed into Bot Auto, with another ~32 going into the TikTok/ByteDance system. ② The pull of frontier AI Labs is showing: among Waymo's ethnic-Chinese outflows, OpenAI and DeepMind together account for about 30, and adding Meta makes the research-track flow clear. ③ Fresh forces: the new robotaxi entrant Tensor Auto has already absorbed about 10 ethnic-Chinese engineers from Waymo. ④ About 57 visibly returned to China (roughly 1.6% of the sample; returnees often stop updating their profiles, so the real figure is higher), with the vast majority staying mobile within the North-American ecosystem — the Bay Area remains the absolute center of this talent market.
每张卡片：Metix AI 数据库可见的北美在职/算法池/华人算法数，叠加 2025-2026 公司动态与人才观察。Each card: Metix AI-database-visible North-American current/algorithm-pool/ethnic-Chinese-algorithm counts, layered with 2025-2026 company developments and talent observations.
现状Status 断层领先者，刚完成 160 亿美元融资，2026 年向 20+ 城铺开，周订单目标 100 万。Runaway leader, having just closed a $16 billion round, rolling out to 20+ cities in 2026 with a weekly-order target of 1 million.
人才池Talent pool 人才密度全行业最高，感知/行为/仿真/数据引擎建制完整，华人 leader 包括感知方向工程总监 Congcong Li 等。Highest talent density in the industry, with complete perception/behavior/simulation/data-engine staffing; ethnic-Chinese leaders include perception engineering director Congcong Li, among others.
人才动态Talent dynamics 行业人才高地。资深层在大组织内面临晋升节奏放缓与决策链拉长，部分追求「从零到一」的工程师会向更早期平台流动，运营波动期前后流动性通常上升。An industry talent peak. Within a large organization, the senior layer faces a slowing promotion pace and lengthening decision chains, so some engineers chasing “zero-to-one” drift toward earlier-stage platforms; liquidity typically rises around operational-volatility periods.
现状Status Amazon 全资，Las Vegas/SF 双城运营，累计载客 35 万+，等收费批准，同时向 Austin/Miami 扩。Wholly owned by Amazon, operating in Las Vegas/SF, 350,000+ cumulative riders, awaiting fare approval, while expanding to Austin/Miami.
人才池Talent pool 算法团队规模仅次于 Waymo，规划与仿真职能比重高，华人工程师网络成熟。The algorithm team is second in size only to Waymo, weighted toward planning and simulation functions, with a mature ethnic-Chinese engineer network.
人才动态Talent dynamics 商业化进度迟于 Waymo，叠加 Amazon RSU 四年归属周期，早期入职的资深员工正进入职业再选择的窗口期。Commercialization lags Waymo, and combined with Amazon's four-year RSU vesting cycle, early-joining senior staff are entering a career-reassessment window.
现状Status 纯视觉 E2E 路线，Robotaxi 已在 Austin 无监督运营并扩 Dallas/Houston，FSD 拟扩到 10B 参数。Pure-vision E2E path; Robotaxi already runs unsupervised in Austin and has expanded to Dallas/Houston, with FSD set to scale to 10B parameters.
人才池Talent pool Autopilot/AI 团队华人骨干密集，工程总监 Phil Duan（段鹏飞）为最高级别华人 leader 之一。The Autopilot/AI team is dense with ethnic-Chinese core engineers, and engineering director Phil Duan is one of the highest-level ethnic-Chinese leaders.
人才动态Talent dynamics 高强度文化下人员流动本就频繁，Optimus 与 FSD 的资源分配带来方向选择；Tesla 的纯视觉大模型履历在行业内含金量高，其离职者是各家关注的人才来源。Turnover is already frequent under the high-intensity culture, and resource allocation between Optimus and FSD forces directional choices; a Tesla pure-vision large-model résumé carries high value in the industry, making its leavers a talent source everyone watches.
现状Status 汽车业务 FY2026 收入 23 亿美元（+39%），吴新宙统管，Alpamayo 开源 VLA + DRIVE Hyperion 生态卡位。The automotive business posted FY2026 revenue of $2.3 billion (+39%), run by Xinzhou Wu, with Alpamayo open-sourcing a VLA + the DRIVE Hyperion ecosystem staking out position.
人才池Talent pool 汽车线华人浓度全行业最高，从芯片到模型到仿真全栈职能齐备。The automotive line has the highest ethnic-Chinese density in the industry, with a full stack of functions from chips to models to simulation.
人才动态Talent dynamics 做平台而非整车，部分工程师希望算法直接上车落地；股价高位使股票 refresh 空间收窄，四年期满者的流动意愿上升。As a platform rather than a full-vehicle maker, some engineers want their algorithms to ship directly on vehicles; the high share price narrows room for equity refreshes, raising the propensity to move among those past four years.
现状Status 已于 2024-12 关停 robotaxi，2025-02 裁员 50%，余部并入 GM。数据库中仍挂 Cruise 的档案属未更新群体。Shut down its robotaxi in 2024-12, cut 50% of staff in 2025-02, with the remainder folded into GM. Profiles still tagged Cruise in the database are an un-updated cohort.
人才池Talent pool 这批仍以 Cruise 为当前归属的人才，实际多已并入 GM Super Cruise、转入其他公司或处于职业过渡，是关停后人才再流动的真实写照。This talent still attributed to Cruise has, in practice, mostly folded into GM Super Cruise, moved to other companies, or is in career transition — a true reflection of post-shutdown talent re-circulation.
人才动态Talent dynamics 这一群体多处于开放求职状态，是行业内活跃度最高的可流动人才。This group is mostly in active job-seeking mode and is the most active, mobile talent in the industry.
现状Status 接收 Cruise 80%+ 工程人才，Mountain View 为核心，eyes-off 系统瞄准 2028 年 Cadillac Escalade IQ。Took in 80%+ of Cruise's engineering talent, with Mountain View as the core; its eyes-off system targets the 2028 Cadillac Escalade IQ.
人才池Talent pool Super Cruise 团队里有大量前 Cruise L4 背景的华人算法工程师。The Super Cruise team has a large number of ethnic-Chinese algorithm engineers with a former Cruise L4 background.
人才动态Talent dynamics 从 robotaxi 转向量产 L2+/L3 带来定位落差，重回 L4 一线的机会对这批工程师有天然吸引力。The shift from robotaxi to mass-production L2+/L3 brings a status drop, so a chance to return to the L4 front line has a natural pull for these engineers.
现状Status 转型 AV 技术授权，绑定 Uber-Lucid 三方计划（至少 3.5 万辆），Uber 承诺近 5 亿美元里程碑付款，2026 年内目标全无人载客。Pivoted to AV technology licensing, locked into the Uber-Lucid three-way plan (at least 35,000 vehicles), with Uber committing nearly $500 million in milestone payments; the target is fully driverless rides within 2026.
人才池Talent pool 联合创始人兼 CEO 朱佳俊是华人智驾创业最高标杆，算法团队精干，感知与行为方向扎实。Co-founder and CEO Jiajun Zhu is the highest benchmark in ethnic-Chinese AV entrepreneurship; the algorithm team is lean, solid in perception and behavior.
人才动态Talent dynamics 估值相对温和且 licensing 模式处于验证期，对看重早期股权空间与华人创始团队文化的工程师有吸引力。With a relatively modest valuation and a licensing model still in its validation phase, it appeals to engineers who prize early-stage equity upside and an ethnic-Chinese founding-team culture.
现状Status 无人卡车商业化第一名：2025-05 起 Dallas-Houston driver-out 常态运营，2026 年底目标 200+ 台。No. 1 in driverless-truck commercialization: regular Dallas-Houston driver-out operations since 2025-05, targeting 200+ trucks by end-2026.
人才池Talent pool 规划/仿真/安全案例方向积累深；前软件 SVP 李燕冰（现 Datadog CPO）离开后，公开层面已无华人 SVP 级高管。Deep accumulation in planning/simulation/safety-case directions; since former software SVP Yanbing Li (now Datadog CPO) departed, there is no publicly known ethnic-Chinese SVP-level executive.
人才动态Talent dynamics 无人卡车场景对追求快速落地与强工程能力的工程师有吸引力，团队分布于 Pittsburgh 与湾区双中心。The driverless-truck setting appeals to engineers who prize fast deployment and strong engineering, with the team split across the dual hubs of Pittsburgh and the Bay Area.
现状Status 估值 150 亿美元，从仿真工具商扩张为全栈 L2+ 供应商 + 防务 AI 双线，与 NVIDIA Cosmos 打通。Valued at $15 billion, expanded from a simulation-tooling vendor into a dual track of full-stack L2+ supplier + defense AI, integrated with NVIDIA Cosmos.
人才池Talent pool 工程师画像偏「工具链 + 仿真 + 客户工程」，华人占比可观，节奏快文化强。The engineer profile leans “toolchain + simulation + customer engineering,” with a sizable ethnic-Chinese share and a fast-paced, strong culture.
人才动态Talent dynamics 现金薪酬低于一线 robotaxi，私司股权的价值兑现取决于上市预期；其工程师普遍兼具算法与交付能力。Cash compensation is below the top-tier robotaxi players, and the value realization of private-company equity depends on IPO expectations; its engineers generally combine algorithm and delivery skills.
现状Status 纯端到端旗手，12 亿美元 D 轮（估值 86 亿，NVIDIA/微软/Uber/三车企参投），Nissan 量产绑定，湾区持续扩张。The pure-end-to-end standard-bearer, with a $1.2 billion Series D (valuation $8.6 billion, backed by NVIDIA/Microsoft/Uber/three automakers), a Nissan production tie-up, and ongoing Bay Area expansion.
人才池Talent pool 美国团队较新、扩张中，E2E/基础模型方向浓度全行业最高。The US team is newer and expanding, with the highest E2E/foundation-model density in the industry.
人才动态Talent dynamics 对坚定看好纯端到端路线的工程师是首选目的地，也是端到端方向人才竞争中绕不开的一方。For engineers firmly bullish on the pure-end-to-end path it is the destination of choice, and it is an unavoidable player in the competition for end-to-end talent.
现状Status Hyundai 主导，2026-01 宣布 AI-first 重启，端到端重写全栈，目标 2026 年底 Las Vegas 无人商业化。Hyundai-led; announced an AI-first relaunch in 2026-01, rewriting the full stack end-to-end, targeting driverless commercialization in Las Vegas by end-2026.
人才池Talent pool 2024 年裁员 40% 后团队精简，留下者多为核心骨干，Boston/Pittsburgh/LA/Vegas 多点分布。After a 40% layoff in 2024 the team is lean, with those remaining mostly core backbone, spread across Boston/Pittsburgh/LA/Vegas.
人才动态Talent dynamics 重启期通常伴随更灵活的级别与股权空间，但稳定性疑虑仍在，员工对更稳健的端到端平台有潜在偏好。A relaunch phase usually comes with more flexible levels and equity room, but stability doubts persist, leaving employees with a latent preference for a more stable end-to-end platform.
现状Status 纳斯达克上市，美国业务收入占比Nasdaq-listed; share of revenue from the US business <1%，Fremont 研发中心保留，商业化重心在中国与中东。
人才池Talent pool 美国团队收缩后剩余成员以资深研发为主，普遍面临「留美做研发」与「回国进核心」的选择。After the US team's contraction, the remaining members are mostly senior R&D, generally facing a choice between “staying in the US to do R&D” and “returning to China for a core role.”
人才动态Talent dynamics 其在美员工对身份支持与业务确定性的诉求显著高于行业均值，是观察中国系在美研发人才心态的样本。Its US employees' demand for immigration support and business certainty is markedly above the industry average — a sample for observing the mindset of China-system US R&D talent.
现状Status 纳斯达克上市，San Jose 设研发，与 Uber 绑定 15 城计划，商业化在中东/东南亚兑现。Nasdaq-listed, with R&D in San Jose, tied to Uber on a 15-city plan, with commercialization realized in the Middle East/Southeast Asia.
人才池Talent pool 美国团队定位纯研发前哨，体量小而精。The US team is positioned as a pure R&D outpost — small but elite.
人才动态Talent dynamics 与小马类似，身份与业务确定性是其在美员工最关注的议题。As with Pony, immigration and business certainty are the issues its US employees care about most.
现状Status 理想 2025 年新建硅谷 AI 实验室；小鹏北美（硅谷 + 圣迭戈）约百人承担算法核心工作并获追加算力；蔚来保留湾区团队。Li Auto built a new Silicon Valley AI lab in 2025; XPeng North America (Silicon Valley + San Diego), about a hundred people, carries core algorithm work and received additional compute; NIO retained its Bay Area team.
人才池Talent pool 三家直接以国内大模型/智驾叙事 + 有竞争力的包在湾区人才竞争，目标画像与本报告完全重叠。All three compete directly for Bay Area talent with a domestic large-model/AV narrative + competitive packages, and their target profile fully overlaps with this report.
人才动态Talent dynamics 它们为华人工程师提供了「不回国的中国机会」，offer 节奏快、决策链短，是本土雇主在湾区的直接竞争方。They offer ethnic-Chinese engineers a “Chinese opportunity without returning to China,” with fast offer cadence and short decision chains — direct competitors to domestic employers in the Bay Area.
现状Status Kodiak 已上市（28 台客户自有无人卡车）；Plus 的 SPAC 终止后维持私有，走 OEM 前装 + VLA 蒸馏路线；Bot Auto（侯晓迪）完成 hub-to-hub 无人验证；Torc 收缩至 DFW/Ann Arbor 但 driver-out 仍排 2026；Gatik 绑定 Loblaw 50 台；Waabi 以世界模型路线推进 Volvo 量产合作；Stack AV 低调。Kodiak is listed (28 customer-owned driverless trucks); after its SPAC fell through, Plus stayed private on an OEM-factory-fit + VLA-distillation path; Bot Auto (Xiaodi Hou) completed driverless hub-to-hub validation; Torc contracted to DFW/Ann Arbor but still has driver-out slated for 2026; Gatik is locked into Loblaw for 50 trucks; Waabi advances a Volvo production collaboration on a world-model path; Stack AV keeps a low profile.
人才池Talent pool 卡车赛道华人体量小而精，Bot Auto/Plus 为华人创始人公司，文化亲和度高。Ethnic-Chinese in the truck space are small in number but elite; Bot Auto/Plus are ethnic-Chinese-founded companies with high cultural affinity.
人才动态Talent dynamics 赛道波动较大（Plus 的 SPAC 上市于 2026-04 终止、Torc 收缩重组），从业者对稳定性更敏感；另一面，Aurora/Kodiak 的无人化运营经验在 robotaxi 公司眼中具有明显溢价。The space is fairly volatile (Plus's SPAC listing terminated in 2026-04, Torc contracted and restructured), so practitioners are more sensitive to stability; on the other side, Aurora/Kodiak's driverless-operations experience carries a clear premium in the eyes of robotaxi companies.
范围 = Waymo 华人算法人才中的感知 / 端到端·基础模型 / 仿真评估三条线，共 243 人。层级依据 title 推断，Scope = three lines within Waymo's ethnic-Chinese algorithm talent — perception / end-to-end·foundation models / simulation-evaluation — totaling 243 people. Levels are inferred from titles, and汇报关系未经官方证实reporting relationships are not officially confirmed，用于呈现团队梯队结构概览。, used to present an overview of the team's tier structure.
说明：总监层勾勒出该方向的资深坐标，带队层反映中坚力量的分布，IC 厚度体现团队的整体规模与梯队结构。Note: the director layer sketches the senior coordinates of this direction, the people-leading layer reflects the distribution of the core force, and IC depth captures the team's overall scale and tier structure.
从 2,693 人名单中按级别、方向与履历强度精选出三组代表性人物。档案事实来自 Metix AI 数据库；标注「公开核实」者已对照 2025-2026 公开信源确认现职。LinkedIn 链接可直接点击，邮箱可经 Metix AI 平台解锁。Three sets of representative figures, hand-picked from the 2,693-person list by level, direction and résumé strength. Profile facts come from the Metix AI database; those marked “publicly verified” have had their current role confirmed against 2025-2026 public sources. LinkedIn links are directly clickable, and emails can be unlocked via the Metix AI platform.
总包 TC = base + 股票年化 + 奖金。levels.fyi 为 2026-06-10 检索值；H1B 数据库只含 base，两套口径不可混用。来源链接附后。Total compensation TC = base + annualized equity + bonus. levels.fyi figures are as retrieved on 2026-06-10; the H1B database contains base only, and the two scopes must not be mixed. Source links follow.
| 公司Company | 中位 TCMedian TC | Senior 级参考Senior-level reference | Staff / Principal 级参考Staff / Principal-level reference | 备注Notes |
|---|---|---|---|---|
| Waymo | $396K | L5 约 $375-500KL5 about $375-500K | L6 居间，L7 可达 $900K+L6 in between, L7 can reach $900K+ | 同业最高，现金 + GUS 股票Highest among peers, cash + GUS equity |
| NVIDIA | $390K | MLE 口径 $205-331K+MLE scope $205-331K+ | IC7 上限 $1.04M+IC7 cap $1.04M+ | 股票增值使老员工纸面包极高Equity appreciation makes veterans' paper packages very high |
| Tesla | $380K (MLE) | Sr Autopilot MLE base 中位 $212K (H1B 2025)Sr Autopilot MLE base median $212K (H1B 2025) | SWE 全距 $137-767K+SWE full range $137-767K+ | 低 base + 股票杠杆文化Low base + equity-leverage culture |
| Aurora | $319K | P5/P6 居间P5/P6 in between | P8/P9 上限 $769K+P8/P9 cap $769K+ | 上市公司，股票流动性好Listed company, good equity liquidity |
| Nuro | $280K | L5 居间L5 in between | L6 上限 $548K+L6 cap $548K+ | 私司，估值约 60 亿，期权弹性Private company, valuation about $6 billion, option upside |
| Zoox | $270K | 约 $300KAbout $300K | L6 上限 $500K+L6 cap $500K+ | Amazon RSU 体系Amazon RSU system |
| Applied Intuition | $219K | $315K+ | 无公开分级样本No public level breakdown sample | 现金偏低，私司股权为主要变量Cash on the low side, private-company equity the main variable |
| Pony.ai / WeRide 美国Pony.ai / WeRide US | 无公开样本No public sample | levels.fyi 无足量数据，H1B base 个案为主Not enough levels.fyi data; mostly individual H1B base cases | 谈判以个案为准Negotiation is case-by-case | |
OpenAI L5 总包约 $1.15M，Anthropic 中位约 $600K，对 AV Senior（$300-500K）形成 1.5-3 倍压差，研究序列差距更大。这一结构性薪酬差是 AV 算法人才向前沿 AI Lab 流动的主要经济动因；AV 公司的相对优势在于基础模型的实景落地、算法上车的兑现感与更密集的带队机会。OpenAI L5 total comp is about $1.15M and Anthropic's median about $600K, creating a 1.5-3x premium over AV Senior ($300-500K), with an even larger gap on the research track. This structural pay gap is the main economic driver of AV algorithm talent flowing to frontier AI Labs; AV companies' relative advantage lies in real-world deployment of foundation models, the tangible payoff of shipping algorithms on vehicles, and more abundant team-leading opportunities.
国内 senior AI 工程师基准约 ¥80-120 万/年，头部智驾高位个案可达 ¥150 万+（历史人才争夺潮口径）；理想/小鹏硅谷实验室则直接以美元包就地竞价。这类机会通常现金占比高、决策快，与本土雇主的股权预期价值、身份支持周期形成直接对照。Senior AI engineers in China benchmark at roughly ¥800,000-1,200,000/year, with top AV cases reaching ¥1,500,000+ at the high end (per the historical talent-war scope); the Li Auto/XPeng Silicon Valley labs, meanwhile, bid on the spot directly in dollar packages. Such opportunities typically have a high cash share and fast decisions, standing in direct contrast to domestic employers' expected equity value and immigration-support timelines.
来源：levels.fyi 各公司页（2026-06-10 检索）、h1bdata.info、OpenAI/Anthropic 薪酬公开梳理、36kr/未来汽车日报（中国包个案）、PBS/Clark Hill（H-1B 判决）、Boundless/USCIS（绿卡数据）。详见附录备忘录。Sources: levels.fyi company pages (retrieved 2026-06-10), h1bdata.info, public roundups of OpenAI/Anthropic compensation, 36kr/Future Auto Daily (Chinese-package cases), PBS/Clark Hill (H-1B ruling), Boundless/USCIS (green-card data). See the appendix memo for details.
同一组数据，对招聘方与从业者各有不同含义。The same set of data means different things to recruiters and to practitioners.
这一人才群体高度集中于湾区、头部三家公司与少数院校脉络，意味着精准而非广撒是更有效的方式。资深骨干层（Staff/Principal 与工程管理）是能独立带方向的稀缺资源；端到端与基础模型方向供给最紧、竞争最激烈。在与前沿 AI Lab 及中国系在美实验室的同台竞争中，AV 公司的差异化优势在于算法的实景落地与兑现感、更密集的带队机会，以及对国际人才稳定、透明的身份支持。This talent group is highly concentrated in the Bay Area, the top three companies, and a handful of school lineages, meaning precision rather than broad casting is the more effective approach. The senior backbone (Staff/Principal and engineering management) is a scarce resource able to own a direction independently; supply is tightest and competition fiercest in end-to-end and foundation models. In head-to-head competition with frontier AI Labs and China-system US labs, AV companies' differentiated advantages are the real-world deployment and tangible payoff of their algorithms, more abundant team-leading opportunities, and stable, transparent immigration support for international talent.
端到端范式转型为传统规控与感知工程师打开了向基础模型迁移的通道；具身智能与机器人提供了相邻的职业出口；中国系在美实验室则提供了「不回国的中国机会」。任期满四年的资深员工、经历过项目关停的工程师，往往处于职业再选择的窗口期，市场流动性整体回升。The end-to-end paradigm shift has opened a channel for traditional planning-control and perception engineers to migrate to foundation models; embodied AI and robotics offer an adjacent career exit; China-system US labs offer a “Chinese opportunity without returning to China.” Senior staff past four years of tenure and engineers who have lived through project shutdowns are often in a career-reassessment window, with overall market liquidity rebounding.
合规提示。Compliance note.智能驾驶是商业机密诉讼高发领域（Waymo v. Uber 先例）。无论招聘方还是候选人，都应确保不携带前雇主的代码、数据与文档；竞业与知识产权条款建议逐条经法律审阅；涉及受出口管制的技术岗位，提前确认合规口径。Autonomous driving is a field with frequent trade-secret litigation (the Waymo v. Uber precedent). Both recruiters and candidates should ensure no code, data or documents from a former employer are carried over; non-compete and IP clauses should be reviewed clause by clause by legal counsel; for technical roles involving export-controlled technology, confirm the compliance scope in advance.
本报告的检索、画像与名单全部由 Metix AI 完成。Mira 可在此基础上自动执行：完整人才名单、联系方式与多渠道联系、面试排期，并按「只为合格面试付费」计费。No interview, no charge.The search, profiling and lists in this report were all produced by Metix AI. On top of this, Mira can automatically execute: the full talent list, contact details and multi-channel outreach, and interview scheduling, billed on a “pay only for qualified interviews” basis. No interview, no charge.
8.6 亿+ 全球人才画像860 million+ global talent profiles2,693 人华人智驾长名单即取即用The 2,693-person ethnic-Chinese AV long list, ready to use邮箱一键解锁One-click email unlock只为合格面试付费Pay only for qualified interviews纳入职能Functions included：感知 / 预测 / 规划与控制 / 定位与建图 / 端到端·基础模型 / 仿真与评估 / 数据引擎·ML 平台 / ML·AI 综合 / 自动驾驶软件（综合）。判定依据为当前职位 title、headline 与技能标签；「自动驾驶软件（综合）」指含 autonomy/self-driving/ADAS 等但未标注细分方向者。: perception / prediction / planning and control / localization and mapping / end-to-end·foundation models / simulation and evaluation / data engine·ML platform / ML·AI general / autonomous-driving software (general). The determination is based on current job title, headline and skill tags; “autonomous-driving software (general)” refers to those containing autonomy/self-driving/ADAS, etc. but with no sub-direction labeled.
明确排除Explicitly excluded：车辆运营与安全员、硬件/机械/制造、产品/项目管理、销售/市场/职能岗、普通 IT 与企业系统；嵌入式/系统软件单独标记，不计入算法核心池。: vehicle operations and safety drivers, hardware/mechanical/manufacturing, product/project management, sales/marketing/functional roles, general IT and enterprise systems; embedded/system software is tagged separately and not counted in the core algorithm pool.
华人识别Ethnic-Chinese identification：五信号交叉验证，即姓名族裔模型（可区分韩/越/日等相近姓名体系）、姓名含汉字、中文语言能力、中国院校教育（剔除国际学校与交换项目）、多拼写体系姓氏库（中韩共用姓氏单独降权）。含汉字或任意两信号交叉 = 高置信；仅单一中等信号 = 中置信；仅语言或仅模糊姓氏不判定。主口径 = 高置信 + 中置信，关键结论同时给出高置信敏感性数字。: five-signal cross-validation — a name-ethnicity model (able to distinguish neighboring naming systems such as Korean/Vietnamese/Japanese), names containing Chinese characters, Chinese-language ability, education at Chinese schools (excluding international schools and exchange programs), and a multi-romanization surname library (shared Chinese-Korean surnames are down-weighted separately). Containing Chinese characters or a cross of any two signals = high confidence; only a single medium signal = medium confidence; language alone or an ambiguous surname alone is not classified. The main scope = high confidence + medium confidence, with key conclusions also giving a high-confidence sensitivity figure.
方法概述Methodology overview：基于 Metix AI 人才库对 36 家目标公司完成全量检索与逐档画像，经雇主归一、职能分类、华人多信号判定与聚合分析后，重点人选逐人对照公开信息复核。: ran a full census and profile-by-profile mapping of the 36 target companies against the Metix AI talent pool, then, after employer normalization, function classification, multi-signal ethnic-Chinese determination and aggregate analysis, reviewed key profiles one by one against public information.
数据时效Data recency：人才库数据时点约为 2026 年上半年；对吴新宙（NVIDIA）、Congcong Li（Waymo）等已知人物抽检，在职公司与头衔与公开信息一致；仍标注 Cruise 在职的档案属「未更新」群体，单独标记而非剔除。: the talent-pool data is as of roughly H1 2026; spot checks of known figures such as Xinzhou Wu (NVIDIA) and Congcong Li (Waymo) found their current company and title consistent with public information; profiles still tagged as employed at Cruise are an “un-updated” cohort, flagged separately rather than removed.
AV 原生公司（Waymo/Zoox/Nuro 等）为全员工检索；巨头公司（Tesla/NVIDIA/Qualcomm/GM/Ford 等，表中带业务线标注者）为「自动驾驶相关」定向检索口径，非该公司整体员工数，且可能含少量相邻业务线人员。AV-native companies (Waymo/Zoox/Nuro, etc.) were searched across all employees; giant companies (Tesla/NVIDIA/Qualcomm/GM/Ford, etc., those with a business line annotated in the table) use an “autonomous-driving-related” targeted-search scope — not the company's total headcount, and possibly including a small number of people from adjacent business lines.
| 公司Company | 北美在职画像North-American current profiles | 算法池Algorithm pool | 华人算法Ethnic-Chinese algorithm | 华人占比Ethnic-Chinese share | 华人池 PhD 率PhD rate of the ethnic-Chinese pool |
|---|---|---|---|---|---|
| Zoox | 3,812 | 1,404 | 273 | 19.4% | 24.2% |
| Waymo | 2,895 | 1,414 | 484 | 34.2% | 31.4% |
| NVIDIA (汽车/DRIVE)NVIDIA (Automotive/DRIVE) | 1,848 | 1,457 | 454 | 31.2% | 26.2% |
| Aurora | 1,693 | 558 | 92 | 16.5% | 20.7% |
| Lucid (ADAS) | 1,550 | 557 | 64 | 11.5% | 21.9% |
| Cruise (关停后)Cruise (post-shutdown) | 1,360 | 583 | 144 | 24.7% | 20.8% |
| GM (Super Cruise) | 1,289 | 801 | 66 | 8.2% | 24.2% |
| Applied Intuition | 1,061 | 558 | 120 | 21.5% | 12.5% |
| Rivian (Autonomy) | 1,024 | 412 | 72 | 17.5% | 29.2% |
| Ford (ADAS) | 912 | 515 | 45 | 8.7% | 37.8% |
| Tesla (Autopilot/AI) | 899 | 494 | 120 | 24.3% | 11.7% |
| Nuro | 841 | 338 | 106 | 31.4% | 22.6% |
| Torc Robotics | 831 | 436 | 34 | 7.8% | 44.1% |
| Motional | 715 | 312 | 44 | 14.1% | 38.6% |
| Qualcomm (Ride/ADAS) | 707 | 395 | 70 | 17.7% | 25.7% |
| Latitude AI (Ford) | 565 | 296 | 38 | 12.8% | 23.7% |
| May Mobility | 286 | 124 | 11 | 8.9% | 18.2% |
| Stack AV | 243 | 131 | 12 | 9.2% | 16.7% |
| Waabi | 242 | 132 | 35 | 26.5% | 34.3% |
| Wayve 美国Wayve US | 233 | 120 | 24 | 20.0% | 41.7% |
| XPeng 美国XPeng US | 217 | 141 | 117 | 83.0% | 35.0% |
| Woven by Toyota 美国Woven by Toyota US | 217 | 126 | 32 | 25.4% | 34.4% |
| Momenta 美国Momenta US | 195 | 39 | 15 | 38.5% | 20.0% |
| Avride | 160 | 62 | 0 | 0.0% | 0% |
| NIO 美国NIO US | 158 | 32 | 23 | 71.9% | 39.1% |
| Gatik | 148 | 58 | 8 | 13.8% | 12.5% |
| Pony.ai 美国Pony.ai US | 129 | 63 | 58 | 92.1% | 20.7% |
| Kodiak AI | 87 | 39 | 4 | 10.3% | 25.0% |
| Helm.ai | 74 | 47 | 12 | 25.5% | 25.0% |
| WeRide 美国WeRide US | 70 | 45 | 40 | 88.9% | 15.0% |
| Bot Auto | 70 | 33 | 24 | 72.7% | 33.3% |
| Mobileye 美国Mobileye US | 66 | 16 | 3 | 18.8% | 0.0% |
| Plus (PlusAI) | 44 | 23 | 11 | 47.8% | 27.3% |
| AutoX | 44 | 23 | 19 | 82.6% | 15.8% |
| Li Auto 美国Li Auto US | 39 | 8 | 8 | 100.0% | 0.0% |
| Baidu 美国Baidu US | 13 | 12 | 11 | 91.7% | 45.5% |
① 数据时效Data recency：数据库为静态数据（约 2026 年上半年），近 1-2 个季度的人事变动存在滞后；重点人选已逐人对照公开信息复核，长名单使用前建议二次核对现职。: the database is static data (around H1 2026), and personnel changes in the most recent 1-2 quarters lag; key profiles have been reviewed one by one against public information, and the long list should be re-checked for current roles before use.
② 覆盖率Coverage：本报告基于公开职业档案聚合，规模型公司的覆盖度高于小型团队；各项数字为数据库口径，宜与公司公开编制互为参照。: this report is based on aggregation of public professional profiles, and coverage is higher for scale-type companies than for small teams; all figures are database scope and are best cross-referenced with companies' publicly disclosed headcount.
③ 职能分类Function classification：基于 title/headline/技能关键词规则，title 模糊者可能误分；「软件工程（其他）」中仍可能藏有算法工程师（标题不写方向的公司，如 Tesla，实际算法人数会被低估）。: based on title/headline/skill-keyword rules, those with ambiguous titles may be misclassified; “software engineering (other)” may still hide algorithm engineers (at companies that do not state a direction in titles, such as Tesla, the actual algorithm headcount will be underestimated).
④ 华人识别为概率判定Ethnic-Chinese identification is a probabilistic determination：使用西文名且无任何中国信号的华裔（尤其二代移民）会漏检；姓名模型对个别非华人存在误检，中置信层已隔离此类风险。主口径 2,693 人 = 高置信 2,566 + 中置信 127，做敏感性分析时可仅采用高置信数。: ethnic-Chinese who use a Western name and have no Chinese signal at all (especially second-generation immigrants) will be missed; the name model misclassifies a few non-ethnic-Chinese, and the medium-confidence layer isolates this risk. The main scope of 2,693 = 2,566 high confidence + 127 medium confidence, and for sensitivity analysis the high-confidence number alone can be used.
⑤ Cruise 群体The Cruise cohort：Cruise 已于 2024-12 关停，仍以其为当前归属的档案对应关停后处于职业过渡的人才，单独成列呈现而非剔除。: Cruise shut down in 2024-12, and profiles still attributed to it correspond to post-shutdown talent in career transition — presented as a separate column rather than removed.
⑥ 流出侧样本（3.9 节）Outflow-side sample (Section 3.9)：离职者检索受样本上限约束，流出数偏保守；「回国」按档案地址判定，回国者常停更档案，回国率为低估。组织拼图（4.1 节）依据公开职位 title 归类，为非官方结构概览，属示意性整理而非事实断言。: the search for leavers is constrained by a sample cap, so the outflow count runs conservative; “returned to China” is determined by profile address, and since returnees often stop updating their profiles, the return-to-China rate is underestimated. The org jigsaw (Section 4.1) is grouped by public job titles, an unofficial structural overview — an illustrative arrangement rather than a factual assertion.