基于 Metix AI 8.6 亿+ 全球人才库，对 25 家顶级量化机构（Citadel / Jane Street / HRT / Two Sigma / Jump / D.E. Shaw / XTX / Optiver 等）与前沿 AI Lab 之间的双向人才流动做全量画像：谁在向 AI Lab 流失、哪些 Lab 在反向吸量化、两边的包差与「金手铐到期」窗口，以及量化作为华人浓度最高金融细分的华人维度分析。覆盖纽约 / 芝加哥 / 伦敦 / 新加坡 / 香港。Built on Metix AI's 860 million+ global talent pool, a full-population picture of the two-way talent flow between 25 top quant firms (Citadel / Jane Street / HRT / Two Sigma / Jump / D.E. Shaw / XTX / Optiver and more) and the frontier AI labs: who is bleeding to the AI labs, which labs are pulling quant talent the other way, the pay gap on both sides and the "golden-handcuff expiry" windows, plus a Chinese-talent analysis of quant as the finance sub-sector with the highest Chinese concentration. Covering New York / Chicago / London / Singapore / Hong Kong.
以下为 Metix AI 数据库可见口径（数据截至 2026 年上半年），统计对象 = 当前在职于 25 家顶级量化机构、坐标美国/英国/新加坡/香港/荷兰的人才，以及量化↔前沿 AI Lab 之间的双向流动。The figures below reflect what is visible in the Metix AI database (data through the first half of 2026). The population = talent currently employed at 25 top quant firms and based in the US / UK / Singapore / Hong Kong / Netherlands, plus the two-way flow between quant and the frontier AI labs.
可见口径下，现任前沿 AI Lab、拥有量化机构履历者On the visible data, people now at a frontier AI lab who carry quant-firm experience number 367 人367（其中 (of whom 233 人233为正式量化岗位经历、非学生实习）；反向（现任量化、有 AI Lab 履历）仅 held full-time quant roles rather than student internships); the reverse flow (now in quant, with AI-lab experience) is just 90 人90。即便只计正式量化岗，净流向仍约 2.6:1（含实习口径 4.1:1）。量化是 AI Lab 的净人才供给方而非相反，这把 Bloomberg「AI Lab 屋顶酒会抢 quant」的个案报道第一次落到了全量档案的方向与规模上。. Even counting full-time quant roles only, the net direction is still about 2.6:1 (4.1:1 once internships are included). Quant is a net supplier of talent to the AI labs, not the other way around — putting Bloomberg's anecdotal "AI labs throw rooftop parties to poach quants" coverage onto full-population direction and scale for the first time.
按现任 Lab 拆分：OpenAI 吸纳 159 名前量化人居首，Anthropic 94、Google DeepMind 78、xAI 32。这与 2025 年 OpenAI/Anthropic 在曼哈顿与伦敦密集面向 quant 的招聘动作一致。送出最多的是 Citadel（84）、Jane Street（80）、Two Sigma（47）。Broken out by current lab: OpenAI leads with 159 ex-quant hires, followed by Anthropic 94, Google DeepMind 78, xAI 32. This tracks OpenAI's and Anthropic's intensive 2025 quant-focused recruiting across Manhattan and London. The biggest exporters are Citadel (84), Jane Street (80) and Two Sigma (47).
量化技术池华人占 22.4%（基于姓名信号的保守口径），而流动人群（量化→AI Lab）华人占Chinese talent is 22.4% of the quant technical pool (a conservative, name-signal-based measure), while among the movers (quant → AI lab) the Chinese share is 29.7%，显著更高。量化是华人浓度最高的金融细分之一，奥数/竞赛→顶尖院校→量化→AI 的人才管线把同一批华人推向两个买家。, markedly higher. Quant is one of the finance sub-sectors with the highest Chinese concentration, and the Math Olympiad / competition → elite-university → quant → AI pipeline pushes the same cohort of Chinese talent toward both buyers.
量化机构用多年递延薪酬、3-4 年竞业/园艺假锁人；2024 年 FTC 全美竞业禁令被推翻、佛州 CHOICE 法案允许 4 年竞业，锁定反而更强。但关键结构性裂缝是：Quant firms lock people in with multi-year deferred compensation and 3-4-year non-competes / garden leave; with the FTC's nationwide non-compete ban struck down in 2024 and Florida's CHOICE Act allowing 4-year non-competes, the lock-in has only tightened. But the key structural crack is this:AI Lab 不被认定为量化「竞争对手」AI labs are not treated as quant "competitors"，于是 quant 可以在园艺假期间直接入职 OpenAI，AI Lab 成了竞业期的默认停留地。任期数据（第 6 节）据此定位高流动窗口。, so a quant can join OpenAI directly during garden leave, and the AI labs have become the default place to sit out a non-compete. The tenure data (Section 6) uses this to pinpoint the high-mobility windows.
以下基于 2024-2026 公开信源逐条核实（完整来源见研究底稿），只保留影响人才决策的事实。金额为报道口径，多为个案。The following is verified point by point against public sources from 2024-2026 (full sources in the research memo), keeping only facts that affect talent decisions. Dollar figures are as reported and are mostly individual cases.
Bloomberg（2025-08-08）报道 OpenAI、Perplexity 等直接面向投行/量化 quant 招聘；Anthropic 2025-06 在曼哈顿下东区屋顶办约 150 人的 quant 酒会、8 月在伦敦办专场。已公开核实的点名流动：Jane Street→Anthropic（Aron Thomas、James Chen 等）、Citadel/Citadel Securities→OpenAI（Zeyuan Shang、Andrey Grinshpun 等，多在园艺假期间入职）、Jane Street→OpenAI（Mark Chen，现任 OpenAI 首席研究官 CRO）。Bloomberg (2025-08-08) reported that OpenAI, Perplexity and others are recruiting quants directly from banks and quant firms; Anthropic threw a roughly 150-person quant cocktail party on a Lower East Side Manhattan rooftop in 2025-06 and ran a dedicated London session in 2025-8 (August). Named, publicly verified moves: Jane Street → Anthropic (Aron Thomas, James Chen and others), Citadel / Citadel Securities → OpenAI (Zeyuan Shang, Andrey Grinshpun and others, many joining during garden leave), Jane Street → OpenAI (Mark Chen, now OpenAI's Chief Research Officer, CRO).
入门 quant base 报道口径「高至 $300K」（不含奖金）；HFT 甚至以最高 $425K 回聘实习生防守管线。AI Lab 这边：OpenAI 研究员总包中位约 $100 万（L4 base $297K + 股票 $474K），对少数几年经验 quant 开价传可达 $300 万；OpenAI 2025 全员人均股票薪酬约 $150 万。结构变化是：Lab 已能对齐 base、用股票上行替代 quant 奖金，「不再需要降薪」。Entry-level quant base pay is reported "as high as $300K" (excluding bonus); HFT shops even re-hire interns at up to $425K to defend the pipeline. On the AI-lab side: median total comp for an OpenAI researcher is around $1 million (L4 base $297K + $474K in stock), and offers to a few quants with a few years' experience reportedly reach $3 million; OpenAI's 2025 average stock comp per head was roughly $1.5 million. The structural shift is that the labs can now match base pay and replace the quant bonus with equity upside, so candidates "no longer have to take a pay cut."
低延迟系统、大规模 GPU/推理优化、强化学习与市场微结构的思维同源，是 AI Lab 紧缺的能力。同时量化机构自己也在建 AI：XTX 自建 >2.5 万张 GPU 集群、HRT 设 HAIL「为市场建基础模型」、Two Sigma 由前 Google 的 Mike Schuster 领队、Citadel 用 RL 优化交易。Ken Griffin 2025 承认「这次 AI 是真的」、且与 AI 抢同一批数理人才。Low-latency systems, large-scale GPU / inference optimization, and a reinforcement-learning mindset that shares its roots with market microstructure are exactly the skills the AI labs are short of. At the same time the quant firms are building AI themselves: XTX has stood up a cluster of more than 25,000 GPUs, HRT founded HAIL to "build foundation models for markets," Two Sigma's effort is led by ex-Google's Mike Schuster, and Citadel uses RL to optimize trading. Ken Griffin conceded in 2025 that "this time AI is real" — and that he is fighting over the same quantitative talent as AI.
FTC 全美竞业禁令 2024-08 被法院全国性推翻；佛州 CHOICE 法案 2025-07 生效允许 4 年竞业/园艺假；Citadel 推 4 年期、SIG 3 年期，买方 sit-out 普遍 12 个月、长至 24-36 个月。真正的成本是递延薪酬没收（未归属股权即离职成本）。但 AI Lab 不被视作量化「竞争对手」，quant 得以在竞业/园艺假期间合法入职 Lab，这正是这一双向流动得以成形的法律缝隙。The FTC's nationwide non-compete ban was struck down nationwide by the courts in 2024-08; Florida's CHOICE Act took effect in 2025-07, allowing 4-year non-competes / garden leave; Citadel pushes 4-year terms, SIG 3 years, and buy-side sit-outs are typically 12 months, stretching to 24-36 months. The real cost is forfeited deferred compensation (unvested equity is the cost of leaving). But because the AI labs are not seen as quant "competitors," a quant can legally join a lab during a non-compete / garden leave — the legal loophole that lets this two-way flow take shape.
统计对象 = 29,317 名在职量化机构员工（含技术池 11,632：量化研究 / 量化开发 / ML 研究 / HFT 系统 / 数据）。Population = 29,317 current quant-firm employees (including a 11,632-person technical pool: quant research / quant dev / ML research / HFT systems / data).
读数：Citadel（5,798）、Point72（2,337）、Jane Street（2,112）规模领先。注意各机构档案维护率差异大：量化机构普遍 NDA 严、LinkedIn 维护率低，绝对数为可见下限；自营盘小机构（如 PDT、Quadrature）天然档案少。Reading: Citadel (5,798), Point72 (2,337) and Jane Street (2,112) lead on size. Note the wide variation in profile-maintenance rates: quant firms generally have strict NDAs and low LinkedIn upkeep, so absolute counts are a visible floor; small prop shops (such as PDT and Quadrature) naturally have few profiles.
读数：量化开发（7,056）与量化研究（3,320）是技术池主体，也是 AI Lab 最想要的两类（研究思维 + 大规模系统工程）。明确标注 ML/AI 研究的只有 176 人，量化机构的 ML 能力多藏在「quant researcher / dev」头衔下，title 会低估真实 AI 能力，评估必须回到作品与竞赛/论文背景。Reading: quant dev (7,056) and quant research (3,320) make up the bulk of the technical pool — and the two categories the AI labs want most (research thinking + large-scale systems engineering). Only 176 are explicitly tagged as ML / AI research; quant firms' ML capability is mostly hidden under "quant researcher / dev" titles, so titles understate true AI ability and assessment has to come back to portfolios and competition / publication backgrounds.
读数：美国 19,848 占主体（纽约 + 芝加哥两大量化中心），英国 6,108（伦敦，XTX/Qube/Marshall Wace/G-Research）；亚洲新加坡 1,198 + 香港 1,104 是增长极（Jane Street、HRT、Citadel Securities 在扩张），荷兰 1,059（Optiver/IMC 阿姆斯特丹）。Reading: the US dominates at 19,848 (the two quant hubs of New York + Chicago), the UK 6,108 (London — XTX / Qube / Marshall Wace / G-Research); in Asia, Singapore 1,198 + Hong Kong 1,104 are the growth poles (Jane Street, HRT and Citadel Securities are expanding), and the Netherlands 1,059 (Optiver / IMC in Amsterdam).
读数：创始人/高管 2,582、总监/负责人 2,067 构成可触达的资深层；初级/实习 4,451 反映量化「校招 + 实习转正」的金字塔结构，这也是 AI Lab 入门级竞争最激烈的层（HFT 用 $425K 回聘实习生防守）。Reading: founders / executives 2,582 and directors / leads 2,067 form the reachable senior layer; the 4,451 juniors / interns reflect quant's "campus hiring + intern-to-full-time" pyramid — also the layer where the AI labs compete most fiercely at entry level (HFT defends with $425K intern re-hire offers).
这是本报告的核心。可见口径下，量化 → AI LabThis is the heart of the report. On the visible data, quant → AI lab is 367 人367（含学生实习；正式量化岗 233 人），AI Lab → 量化 (including student internships; 233 in full-time quant roles), and AI lab → quant is 90 人90，净流向约 4.1:1（仅计正式岗 2.6:1）。以下逐条拆解方向、来源与去向。, a net direction of about 4.1:1 (2.6:1 counting full-time roles only). Below we unpack direction, source and destination point by point.
读数：最粗的通道是 Citadel → OpenAI（49 人）。OpenAI 与 Anthropic 是两大汇聚口，Citadel（含 Securities）与 Jane Street 是两大源头，这两家也正是公开报道中被点名最多的「净流出」机构。Reading: the thickest corridor is Citadel → OpenAI (49). OpenAI and Anthropic are the two big collectors, Citadel (incl. Securities) and Jane Street the two big sources — and the very firms most often named as "net exporters" in public coverage.
读数：OpenAI（159）领先，Anthropic（94）、Google DeepMind（78）次之。OpenAI/Anthropic 的高吸纳与其 2025 年密集面向 quant 的招聘动作（屋顶酒会、园艺假入职）直接对应。Reading: OpenAI (159) leads, with Anthropic (94) and Google DeepMind (78) next. OpenAI's and Anthropic's high intake maps directly onto their intensive 2025 quant recruiting (rooftop parties, garden-leave hires).
读数：Citadel（84）、Jane Street（80）、Two Sigma（47）是前三大源头。这与它们的规模、研究文化（Jane Street 的 ML track、Two Sigma 的生成式 AI 团队）以及被 Lab 重点招募的程度一致。Reading: Citadel (84), Jane Street (80) and Two Sigma (47) are the top three sources. This is consistent with their scale, research culture (Jane Street's ML track, Two Sigma's generative-AI team) and the degree to which the labs target them.
读数：Citadel 与 Jane Street 居右上（体量大、流出多），是人才流动的主要来源方；Two Sigma 相对其规模流出强度偏高（研究文化更接近 AI Lab）。气泡大小显示华人是各家流出的主力构成。Reading: Citadel and Jane Street sit top-right (large, high outflow) as the main sources of the flow; Two Sigma's outflow intensity runs high relative to its size (its research culture is closer to the AI labs'). Bubble size shows Chinese talent is the main component of each firm's outflow.
读数：反向流仅 90 人、约为正向的 1/4.1 折，量化整体是净流出方。反向者多来自 Google DeepMind（42，成立早、alumni 多），少数来自 OpenAI（27）。动机多为薪酬确定性与「去泡沫」，而非主流。Reading: the reverse flow is just 90 people — about 1/4.1 of the forward flow — so quant is a net exporter overall. Most reverse-movers come from Google DeepMind (42, the earliest-founded, with the most alumni), a few from OpenAI (27). The motive is usually pay certainty and "de-bubbling," not the mainstream.
Jane Street → Anthropic：Aron Thomas、James Chen、Charles Guo、Kerrick Staley（多为 MTS / 研究序列）。: Aron Thomas, James Chen, Charles Guo, Kerrick Staley (mostly MTS / research track).Citadel / Citadel Securities → OpenAI：Zeyuan Shang、Andrey Grinshpun、Eugene Tang（多在竞业/园艺假期间入职）。: Zeyuan Shang, Andrey Grinshpun, Eugene Tang (many joining during a non-compete / garden leave).Jane Street → OpenAI：Mark Chen（现任 OpenAI 首席研究官 CRO）。: Mark Chen (now OpenAI's Chief Research Officer, CRO).反向（AI → 量化侧）Reverse (AI → quant side)：Leopold Aschenbrenner 离 OpenAI 创办 Situational Awareness 基金（AI 研究背景做 AI 主题投资）。这些点名个案与本报告全量方向一致：人才主要从量化流向 AI Lab。: Leopold Aschenbrenner left OpenAI to found the Situational Awareness fund (an AI-research background turned to AI-themed investing). These named cases align with the report's full-population direction: talent flows mainly from quant to the AI labs.
来源：Bloomberg 2025-08-08、eFinancialCareers、各公司公开信息（详见研究底稿）。点名个案现职以公开信源为准。Sources: Bloomberg 2025-08-08, eFinancialCareers, and each company's public information (see the research memo for detail). For named cases, current roles are per public sources.
量化是华人浓度最高的金融细分之一。可见口径（基于姓名信号的保守识别，无受保护属性字段）：技术池华人占 22.4%，越靠近 AI Lab 浓度越高，流动人群（量化→Lab）达 29.7%。Quant is one of the finance sub-sectors with the highest Chinese concentration. On the visible data (conservative identification from name signals, no protected-attribute fields): Chinese talent is 22.4% of the technical pool, and the concentration rises the closer you get to the AI labs, reaching 29.7% among the movers (quant → lab).
读数：从全体池（16.5%）到技术池（22.4%）再到流动人群（29.7%）单调上升，说明华人不仅是量化技术主力，更是被 AI Lab 优先吸纳的那部分。这与「奥数/竞赛→顶尖院校→量化→AI」同一条管线高度吻合。Reading: a monotonic rise from the full pool (16.5%) to the technical pool (22.4%) to the movers (29.7%) shows Chinese talent is not only the backbone of quant tech but the part the AI labs absorb first. This fits the single "Math Olympiad / competition → elite university → quant → AI" pipeline closely.
读数：Two Sigma、Jump、Jane Street、Qube 等研究驱动型机构华人浓度居前；做市/交易型相对低。注意这是全员口径，技术岗的华人浓度普遍高于全员。Reading: research-driven firms such as Two Sigma, Jump, Jane Street and Qube top the Chinese concentration; market-making / trading-oriented firms run lower. Note this is a firm-wide measure — Chinese concentration in technical roles is generally higher than firm-wide.
读数（基于 1,500 名华人技术样本的教育档案）：清北 + 中科大 + 上交是中国本土主力，叠加 MIT / CMU / Berkeley / Columbia 的研究生管线。这正是数学奥赛→竞赛编程→量化→AI 共享的同一条供给链：Jane Street、Citadel、OpenAI 都赞助 IMO，抢的是同一批人。Reading (based on the education profiles of a 1,500-person Chinese technical sample): Tsinghua / Peking + USTC + SJTU are the mainland-China backbone, layered with the graduate pipelines of MIT / CMU / Berkeley / Columbia. This is the single supply chain shared by Math Olympiad → competitive programming → quant → AI: Jane Street, Citadel and OpenAI all sponsor the IMO, fighting over the same people.
量化领军Quant leaders：Peng Zhao 赵鹏（Citadel Securities CEO）、Liang Wenfeng 梁文锋（High-Flyer 幻方 → DeepSeek，量化转 AI 的最强样本）、Jian Wu（Two Sigma）。: Peng Zhao (Citadel Securities CEO), Liang Wenfeng (High-Flyer → DeepSeek, the strongest example of quant-to-AI), Jian Wu (Two Sigma).量化→AI Lab 流动人群Quant → AI lab movers：Mark Chen（Jane Street → OpenAI 首席研究官 CRO）、以及一批以 MTS/研究序列进入 OpenAI/Anthropic 的华人 quant。: Mark Chen (Jane Street → OpenAI Chief Research Officer, CRO), plus a cohort of Chinese quants who entered OpenAI / Anthropic on the MTS / research track.奥数管线The Olympiad pipeline：多位 IMO 金牌得主进入 Citadel/Jane Street/HRT，与进入 AI Lab 的金牌得主同源。: several IMO gold medalists have joined Citadel / Jane Street / HRT, from the same source pool as the medalists who join the AI labs.
背景：量化与 AI 争夺的是同一条「数学奥赛→竞赛编程→顶尖院校」的供给链，华人在这条链上的高占比是结构性的（见 5.3 院校管道），这也是华人在量化与 AI 两侧浓度同源且互通的根因。Background: quant and AI compete over the same "Math Olympiad → competitive programming → elite university" supply chain, and the high Chinese share along that chain is structural (see the 5.3 education pipeline) — the root cause of why Chinese concentration on the quant and AI sides shares a source and flows between them.
量化用多年递延薪酬 + 竞业/园艺假锁人。任期结构能反推「归属节点 / 竞业到期」的可触达窗口，这是猎头与 AI Lab recruiting 最该盯的时间表。Quant locks people in with multi-year deferred comp + non-competes / garden leave. Tenure structure lets you back out the "vesting cliff / non-compete expiry" outreach windows — the timetable headhunters and AI-lab recruiting should watch most closely.
读数：任期 2-4 年的有 7,002 人，正落在多数机构「递延薪酬大额归属 / 初始竞业期临近」的区间，这是流动性最高、最值得主动触达的人群。<1 年（7,288）多在蜜月期、且新签竞业最紧；5 年以上（6,504）是资深沉淀层，触达难但价值高。Reading: 7,002 people have 2-4 years' tenure, landing squarely in the zone where most firms hit a large deferred-comp vesting cliff / the initial non-compete nears expiry — the most mobile, highest-priority group to approach. The <1-year cohort (7,288) is mostly in the honeymoon phase with the freshest, tightest non-competes; 5+ years (6,504) is the settled senior layer, hard to reach but high-value.
| 机制Mechanism | 典型条款（报道口径）Typical terms (as reported) | 对触达的含义What it means for outreach |
|---|---|---|
| 递延薪酬 / 未归属股权Deferred comp / unvested equity | 多年分期归属Vests over multiple years | 归属节点前 3-6 个月是窗口前哨；未归属额 = 跳槽的真实成本The 3-6 months before a vesting cliff is the leading edge of the window; the unvested amount = the real cost of moving |
| 竞业 / 园艺假Non-compete / garden leave | Citadel 至 4 年 · SIG 3 年 · 买方 12-36 月Citadel up to 4 years · SIG 3 years · buy-side 12-36 months | 到期前是窗口；AI Lab 非「竞争对手」可在园艺假期间入职The run-up to expiry is the window; AI labs, not being "competitors," can be joined during garden leave |
| 非竞争法律环境Non-compete legal environment | FTC 全美禁令 2024 被推翻；佛州 4 年竞业FTC nationwide ban struck down in 2024; Florida 4-year non-compete | 锁定更强，但也更可预测；按到期排期主动触达Lock-in is stronger but also more predictable; approach on the expiry schedule |
| 入门防守Entry-level defense | HFT 最高 $425K 回聘实习生HFT re-hires interns at up to $425K | 入门层正是 Lab 竞争最激烈、机构防得最贵的层The entry level is exactly where the labs compete hardest and firms defend most expensively |
从量化→AI Lab 的流动人群中、按正式量化岗位经历（非学生实习）、级别与通道代表性精选三组共 19 人。档案事实来自 Metix AI 数据库；标注「公开核实」者为已对照公开信源确认的公众人物或 Bloomberg 点名个案。公开版人名默认模糊。A curated set of 19 people across three groups, drawn from the quant → AI-lab movers and selected for full-time quant experience (not student internships), seniority and corridor representativeness. Profile facts come from the Metix AI database; those marked "publicly verified" are public figures or Bloomberg-named cases confirmed against public sources. Names are masked by default in the public version.
量化与 AI Lab 抢同一批人，但出价结构不同：量化给确定性现金（base + 大额奖金 + 递延），AI Lab 给 base 对齐 + 股票上行。数字为 2025-2026 报道口径，多为个案。Quant and the AI labs fight over the same people but bid with different structures: quant offers certain cash (base + a large bonus + deferral), the AI labs offer matched base + equity upside. Figures are as reported for 2025-2026 and are mostly individual cases.
| 群体Group | 薪酬区间（报道口径）Pay range (as reported) | 说明Notes |
|---|---|---|
| 入门 quant（base）Entry-level quant (base) | 高至 $300KUp to $300K | Bloomberg；Jane Street/Five Rings 研究员 base ~$300K（H1B 口径）Bloomberg; Jane Street / Five Rings researcher base ~$300K (per H1B filings) |
| 顶尖暑期实习Top summer internship | ~$25K/月；回聘 offer 至 $425K~$25K/month; re-hire offers up to $425K | HFT 用高价防守入门管线（OpenAI 逼近）HFT defends the entry pipeline with high pay (OpenAI is closing in) |
| OpenAI 研究员（总包中位）OpenAI researcher (median total comp) | ~$100 万（L4 base $297K + 股票 $474K）~$1 million (L4 base $297K + $474K stock) | levels.fyi；L5 中位约 $147 万levels.fyi; L5 median about $1.47 million |
| OpenAI 对资深 quant 开价OpenAI's offer to a senior quant | 传至 $300 万Reportedly up to $3 million | 单一信源、个案，谨慎引用Single source, individual case, cite with caution |
| Anthropic SWE（levels）Anthropic SWE (levels) | ~$56 万-$78 万；资深研究员过 $100 万~$560K-$780K; senior researchers above $1 million | 含 tender 流动性Includes tender liquidity |
| Meta 超级智能（明星个案）Meta Superintelligence (star case) | 报道至 ~$1 亿级Reported into the ~$100 million range | 极端个案，RSU 结构，谨慎引用Extreme case, RSU structure, cite with caution |
过去 AI Lab 要 quant 得让其「降薪换使命」；2025 年的变化是 Lab 已能Hiring a quant used to mean asking them to "take a pay cut for the mission"; the 2025 change is that the labs can now对齐 base、用股票上行替代量化奖金match the base and replace the quant bonus with equity upside，于是「不再需要降薪」（Noam Brown 语义）。叠加 AI 的叙事与研究自由，天平向 Lab 倾斜，这正是这一流动 4.1:1 净流出的薪酬侧解释。, so candidates "no longer have to take a pay cut" (in Noam Brown's words). Layer on AI's narrative and research freedom and the balance tilts toward the labs — the comp-side explanation for this flow's 4.1:1 net outflow.
① 猎头：量化→AI 的候选人对「现金确定性 vs 股票上行」高度敏感，开场要讲清股票结构与流动性（tender）；② 量化 HR：防守要靠递延归属 + 竞业 + 入门高价，但需正视 base 已被对齐；③ AI Lab recruiting：可用「园艺假期间合法入职 + base 对齐 + 上行」三件套精准吸引 2-4 年任期的 quant。① Headhunters: quant → AI candidates are highly sensitive to "cash certainty vs equity upside," so open by spelling out the stock structure and liquidity (tender); ② quant HR: defense rests on deferred vesting + non-competes + high entry pay, but must accept that base has been matched; ③ AI-lab recruiting: use the three-piece combo of "legal hire during garden leave + matched base + upside" to target quants at 2-4 years' tenure precisely.
来源：Bloomberg、eFinancialCareers、levels.fyi、Fortune、公司公开信息（2025-2026 检索）。详见研究底稿；标注「个案/单一信源」者谨慎使用。Sources: Bloomberg, eFinancialCareers, levels.fyi, Fortune, and company public information (retrieved 2025-2026). See the research memo for detail; treat items marked "individual case / single source" with caution.
把人才流动图变成动作：猎头看可触达窗口与点名通道，量化 HR 看防守，AI Lab recruiting 看精准吸引。Turning the flow map into action: headhunters watch the outreach windows and named corridors, quant HR watches defense, AI-lab recruiting watches precision attraction.
① 重点关注 2-4 年任期 + 递延归属临近的人群（第 6 节），成功率最高；② 已验证通道（Citadel/Jane Street → OpenAI/Anthropic）候选人心理阻力低；③ 华人技术人才（技术池 22.4%、流动人群 29.7%）用校友 + 竞赛圈触达命中率高；④ 量化档案维护率低，本报告的全量画像 + 现职状态本身就是稀缺人才数据。① Focus on the 2-4-year-tenure cohort with a vesting cliff approaching (Section 6) — the highest success rate; ② proven corridors (Citadel / Jane Street → OpenAI / Anthropic) carry low candidate resistance; ③ reach Chinese technical talent (22.4% of the technical pool, 29.7% of the movers) via alumni + competition circles for a high hit rate; ④ with quant profile-maintenance low, this report's full-population picture + current-status data is itself scarce talent intelligence.
① 用本报告的「流出排行」（第 4.3 节）给自家定位，对照流出最多的同业；② 防守盯紧 2-4 年任期 + 高华人浓度技术岗（最易被 Lab 主动触达）；③ 正视 base 已被 Lab 对齐，防守要靠递延归属 + 竞业排期 + 入门高价；④ 反向吸纳（AI→量化，仅 90 人）是小而真实的机会，主来自 DeepMind alumni。① Use this report's "exporter ranking" (Section 4.3) to position yourself against the biggest-bleeding peers; ② defend by watching 2-4-year tenure + high-Chinese-concentration technical roles (the most likely to be approached by the labs); ③ accept that the labs have matched base, so defense rests on deferred vesting + non-compete scheduling + high entry pay; ④ reverse intake (AI → quant, just 90 people) is a small but real opportunity, mainly from DeepMind alumni.
① 用「园艺假期间合法入职 + base 对齐 + 股票上行」三件套吸引 2-4 年 quant；② 优先 HFT 系统/低延迟 + RL 背景（能力最迁移）；③ 复制 OpenAI/Anthropic 的「按机构办专场」打法（屋顶酒会模型）；④ 华人 quant 是最大且最可迁移的池，按院校（清北/中科大/MIT/CMU）+ 竞赛背景精准选面。① Attract quants at 2-4 years' tenure with the three-piece combo of "legal hire during garden leave + matched base + equity upside"; ② prioritize HFT-systems / low-latency + RL backgrounds (the most transferable skills); ③ replicate OpenAI's and Anthropic's "firm-by-firm dedicated event" playbook (the rooftop-party model); ④ Chinese quants are the largest and most transferable pool — screen precisely by university (Tsinghua/Peking / USTC / MIT / CMU) + competition background.
本报告的检索、画像、双向流动分析全部由 Metix AI 完成。可按同样口径为任意机构生成定制人才流动分析：全量长名单导出、按任期/竞业窗口筛选、点名通道还原、邮箱解锁与多渠道触达，并按「只为合格面试付费」计费。No interview, no charge.All of this report's search, profiling and two-way-flow analysis was done by Metix AI. We can produce a custom talent-flow analysis for any firm on the same basis: full long-list export, filtering by tenure / non-compete window, reconstruction of named corridors, email unlock and multi-channel outreach — billed on a "pay only for qualified interviews" model. No interview, no charge.
8.6 亿+ 全球人才画像860 million+ global talent profiles29,317 人量化池 + 367 人流动名单29,317-person quant pool + a 367-person flow list任期/竞业窗口定位Tenure / non-compete window targeting只为合格面试付费Pay only for qualified interviews25 家顶级量化机构（对冲基金 + 自营做市），地理 = 档案常驻地在美国/英国/新加坡/香港/荷兰。量化池 = 当前在职、且当前雇主经金融行业（Financial Services/Capital Markets 等）过滤后匹配目标机构者。前沿 AI Lab = OpenAI / Anthropic / Google DeepMind / xAI / Mistral / Meta AI(FAIR)。25 top quant firms (hedge funds + prop market makers), geography = profiles resident in the US / UK / Singapore / Hong Kong / Netherlands. Quant pool = currently employed, with a current employer that matches a target firm after a financial-services filter (Financial Services / Capital Markets, etc.). Frontier AI labs = OpenAI / Anthropic / Google DeepMind / xAI / Mistral / Meta AI (FAIR).
量化→AI Lab = 当前在职某前沿 AI Lab、且履历中有过目标量化机构（金融行业过滤）。AI Lab→量化 = 当前在职某量化机构、且履历中有过某前沿 AI Lab。一人可计入多条来源通道（sankey 边）。Quant → AI lab = currently employed at a frontier AI lab and with a target quant firm in their history (financial-services-filtered). AI lab → quant = currently employed at a quant firm and with a frontier AI lab in their history. One person can count on multiple source corridors (Sankey edges).
仅基于公开姓名信号（汉字、拼音/粤拼/威妥玛姓氏库、姓名结构），Based only on public name signals (Chinese characters, pinyin / Jyutping / Wade-Giles surname libraries, name structure), and不使用任何受保护属性字段using no protected-attribute fields whatsoever。分高/中置信，主口径 = 高 + 中。教育院校管道基于 1,500 名华人技术样本的公开教育档案。这是基于公开职业信号的族裔/侨界人才市场分析，非个人属性判定。. Split into high / medium confidence, with the main measure = high + medium. The education-pipeline analysis is based on the public education profiles of a 1,500-person Chinese technical sample. This is an ethnic / diaspora talent-market analysis built on public professional signals, not a determination of individual attributes.
角色按 title/headline 归类（量化研究/开发/交易/ML 研究/HFT 系统/数据/高管/职能）；技术池 = 研究+开发+ML+HFT+数据。任期 = 现职在该机构的在岗月数。均为概率推断，title 会低估量化机构内的真实 ML 能力。Roles are classified by title / headline (quant research / dev / trading / ML research / HFT systems / data / executive / functional); the technical pool = research + dev + ML + HFT + data. Tenure = months in the current role at the firm. All are probabilistic inferences, and titles understate the true ML capability inside quant firms.
| 机构Firm | 在职池Current pool | 技术池Technical pool | 华人Chinese | 华人占比Chinese share | →AI Lab |
|---|---|---|---|---|---|
| Citadel | 5,798 | 2,276 | 1,100 | 19.0% | 84 |
| Point72 | 2,337 | 684 | 339 | 14.5% | 8 |
| Jane Street | 2,112 | 753 | 389 | 18.4% | 80 |
| Susquehanna (SIG) | 2,085 | 777 | 226 | 10.8% | 24 |
| Two Sigma | 1,700 | 827 | 427 | 25.1% | 47 |
| Optiver | 1,572 | 597 | 194 | 12.3% | 15 |
| DRW | 1,544 | 607 | 195 | 12.6% | 9 |
| Jump Trading | 1,335 | 819 | 259 | 19.4% | 29 |
| IMC Trading | 1,323 | 562 | 204 | 15.4% | 15 |
| Squarepoint | 1,284 | 800 | 247 | 19.2% | 4 |
| Qube Research | 1,182 | 367 | 217 | 18.4% | 1 |
| D. E. Shaw | 1,113 | 174 | 131 | 11.8% | 23 |
| Hudson River Trading | 1,004 | 594 | 211 | 21.0% | 37 |
| G-Research | 818 | 498 | 9 | 1.1% | 9 |
| Virtu Financial | 724 | 195 | 115 | 15.9% | 3 |
| Tower Research Capital | 634 | 245 | 127 | 20.0% | 15 |
| Marshall Wace | 588 | 200 | 67 | 11.4% | 1 |
| AQR Capital | 552 | 75 | 81 | 14.7% | 9 |
| Akuna Capital | 309 | 169 | 76 | 24.6% | 9 |
| Millennium | 295 | 88 | 39 | 13.2% | 1 |
| Five Rings | 243 | 111 | 93 | 38.3% | 15 |
| Renaissance Technologies | 211 | 46 | 31 | 14.7% | 0 |
| XTX Markets | 198 | 74 | 18 | 9.1% | 0 |
| PDT Partners | 185 | 10 | 30 | 16.2% | 2 |
| Quadrature | 171 | 84 | 7 | 4.1% | 1 |
① 覆盖率/维护率Coverage / maintenance rate：量化机构 NDA 严、LinkedIn 维护率低（样本实测：姓名/履历 100%、教育约 58%、语言约 24%），绝对数为可见下限；机构间比较以占比与结构为主、绝对数为辅。: quant firms have strict NDAs and low LinkedIn upkeep (sample-measured: name / work history 100%, education about 58%, languages about 24%), so absolute counts are a visible floor; cross-firm comparison relies primarily on shares and structure, with absolute counts secondary.
② 实体歧义Entity ambiguity：Citadel 含对冲基金与 Citadel Securities；同名公司经金融行业过滤排除（如 Citadel 广播/学院已剔除）。: Citadel covers both the hedge fund and Citadel Securities; same-named companies are excluded by the financial-services filter (e.g. Citadel Broadcasting / colleges have been removed).
③ 华人识别为概率判定Chinese identification is probabilistic：基于姓名信号、无受保护属性字段；使用西文名且无中文信号的华裔会漏检，故为保守下限。: based on name signals, with no protected-attribute fields; ethnically Chinese people who use Western names with no Chinese signal are missed, so this is a conservative floor.
④ 人才流动为履历推断The flow is inferred from work histories：基于公开任职记录的先后关系，不含离职原因；点名个案现职以公开信源为准、使用前建议二次确认。: based on the sequence of public employment records, with no reason for leaving; for named cases, current roles are per public sources and should be re-confirmed before use.
⑤ 研究底稿Research memos：两份带全部信源 URL 的研究备忘录（行业格局 / 人才生态）与本报告同目录交付。: two research memos with all source URLs (industry landscape / talent ecosystem) are delivered in the same directory as this report.