基于 Metix AI 8.6 亿+ 全球人才库，对美国 22 家头部医疗 AI 公司与医院系统 AI 团队做全量画像，回答 HealthTech 招聘最难的问题：既懂临床工作流、又懂机器学习的「双语人才」存量有多少、藏在哪里、从哪来。Built on Metix AI's 860M+ global talent pool, this report fully profiles 22 leading U.S. medical-AI companies and hospital-system AI teams to answer HealthTech recruiting's hardest question: how many "bilingual" professionals—fluent in both clinical workflow and machine learning—actually exist, where they hide, and where they come from.
以下数字为 Metix AI 数据库口径（数据截至 2026 年上半年），统计对象 = 当前在职于这些医疗 AI 公司与医院系统 AI 团队、坐标美国的人才。The figures below reflect Metix AI database scope (data through H1 2026), counting talent currently employed at these medical-AI companies and hospital-system AI teams and based in the U.S.
以下基于 2025-2026 公开信源逐条核实（完整来源见研究底稿），只保留影响招聘判断的事实。The following is verified point by point against public 2025–2026 sources (full sourcing in the research file), keeping only facts that affect hiring decisions.
环境临床记录赛道 2025 年融资近 10 亿美元（Ambience $243M/估值 $12.5 亿、Suki $168M、Nabla $70M）；Abridge 估值升至 53 亿美元、深度集成 Epic。医生采用率 30-40%、领先医院达 90%。产品从「模型」走向「嵌入临床工作流」，懂工作流的人成为关键。Ambient clinical documentation raised nearly $1B in 2025 (Ambience $243M at a $1.25B valuation, Suki $168M, Nabla $70M); Abridge's valuation climbed to $5.3B with deep Epic integration. Physician adoption runs 30–40%, reaching 90% at leading hospitals. Products are shifting from "the model" to "embedding in clinical workflow," making people who understand that workflow essential.
OpenEvidence 估值 120 亿美元（40% 美国医生在用、单月 1,800 万次咨询）；Microsoft MAI-DxO 在 NEJM 病例诊断准确率报道达 85%；Google AMIE、Aidoc 31 个 FDA 清单产品。FDA 截至 2025-12 累计授权约 1,451 个 AI/ML 医疗设备（2025 单年 +48%，放射科占七成）。OpenEvidence is valued at $12B (used by 40% of U.S. physicians, with 18 million consultations a month); Microsoft MAI-DxO reportedly hit 85% diagnostic accuracy on NEJM cases; plus Google AMIE and Aidoc's 31 FDA-cleared products. Through Dec 2025 the FDA had authorized roughly 1,451 AI/ML medical devices cumulatively (+48% in 2025 alone, with radiology accounting for 70%).
Epic 预训练模型 CoMET（1.18 亿患者数据）、Cosmos 与 MyChart AI；Oracle Health 2025-11 推出 voice-first agentic EHR。EHR 巨头入场，把「临床数据 + ML」的人才需求推到新高。Epic's pretrained model CoMET (trained on data from 118 million patients), Cosmos, and MyChart AI; Oracle Health launched a voice-first agentic EHR in 2025-11. The entry of the EHR giants pushes demand for "clinical data + ML" talent to new highs.
Mayo（Halamka，250+ 算法、8 个基础模型）、UCSF（首任 chief health AI officer）、Kaiser（全系统部署 Abridge）、Mass General Brigham、Stanford RAISE-Health、Providence（1,600 医生用 ambient）。医院从采购转向自建 AI 团队，与厂商抢同一批懂临床的 ML 人才。Mayo (Halamka; 250+ algorithms, 8 foundation models), UCSF (its first chief health AI officer), Kaiser (Abridge deployed system-wide), Mass General Brigham, Stanford RAISE-Health, and Providence (1,600 physicians using ambient). Hospitals are moving from buying to building in-house AI teams, competing with vendors for the same pool of clinically literate ML talent.
统计对象 = 6,415 名美国在职人才（医疗 AI 公司为全员，医院系统/EHR 巨头为可识别 AI/数据/信息学职能子集）。Population = 6,415 current U.S. professionals (full headcount for medical-AI companies; for health systems and EHR giants, the identifiable subset in AI/data/informatics functions).
读数：医疗 AI 机构的人才结构里，纯工程与 ML/数据科学是主体，但「临床/医学事务」与「临床信息学」两个临床侧职能合计占可观比例，这正是医疗 AI 区别于通用 AI 公司的地方。下一节聚焦其中最稀缺的交叉群体。Read: In medical-AI organizations, pure engineering and ML/data science form the bulk, but the two clinical-side functions—Clinical/Medical Affairs and Clinical Informatics—together account for a meaningful share, which is exactly what sets medical AI apart from general-purpose AI companies. The next section zooms in on the scarcest cross-disciplinary group within it.
读数：与前沿 AI Lab 的「湾区一城独大」不同，医疗 AI 人才高度跟随医院系统与公司总部分布。这意味着招聘选址要贴近临床中心，而非只盯湾区。Read: Unlike frontier AI labs, where the Bay Area dominates, medical-AI talent closely tracks the footprints of health systems and company headquarters. That means recruiting should site itself near clinical centers, not fixate on the Bay Area alone.
本报告的核心。「双语人才」= 同时具备临床背景（医学/护理/药学学位、临床执业或临床职务）与机器学习能力（ML/数据科学 title 或技能）。全池仅 1,077 人，占 16.8%，即每 6.0 名医疗 AI 从业者中约 1 名。The heart of this report. "Bilingual talent" = simultaneously holding a clinical background (a medical/nursing/pharmacy degree, clinical practice, or a clinical role) and machine-learning capability (an ML/data-science title or skills). Just 1,077 people across the pool—16.8%, or roughly 1 in every 6.0 medical-AI professionals.
读数：双语人才并非都在写代码。相当一部分落在「临床/医学事务」（把临床判断带进产品）与「临床信息学」（连接两端的桥梁学科）。这决定了招聘时不能只在 ML 岗里找，临床与信息学岗位同样是双语人才的栖息地。Read: Not all bilingual talent writes code. A sizable share sits in Clinical/Medical Affairs (bringing clinical judgment into the product) and Clinical Informatics (the bridge discipline connecting both ends). The implication: don't search only ML roles—clinical and informatics positions are just as much a habitat for bilingual talent.
读数：约 54% 的双语人才履历里有过真实的医院或诊所经历，说明这个群体主要由「在临床干过、再转向 AI」的人构成，而非纯科班 ML 工程师顺带学医。影像（放射/病理）是最早转型的临床专科。Read: About 54% of bilingual talent have real hospital or clinic experience in their résumé, showing the group is made up mainly of people who worked in the clinic first and then moved into AI—not classically trained ML engineers who picked up medicine on the side. Imaging (radiology/pathology) is the earliest clinical specialty to make the switch.
读数：病理、影像、ambient 记录类公司（产品贴近临床判断）与医院系统的临床化程度最高；纯平台/基础设施类公司技术人才占比更高。临床化高的机构是双语人才密度最高的招聘池。Read: Pathology, imaging, and ambient-documentation companies (whose products sit close to clinical judgment), along with health systems, show the highest degree of clinicalization; pure platform/infrastructure companies skew more technical. The most clinicalized organizations are the recruiting pools with the highest density of bilingual talent.
回答招聘最实际的问题：要找临床×ML 双语人才，去哪些「上游」挖。下图按每个人进入当前机构前的最近一段履历归类。Answering the most practical hiring question: to find clinical×ML bilingual talent, which "upstream" sources should you mine? The chart below classifies each person by their most recent prior role before joining their current organization.
读数：全体来看最大来源是科技/医疗公司之间的横向流动；但临床执业（343 人）与学术/医学院（671 人）两条临床上游，是医疗 AI 区别于通用 AI 的特征管道，也是双语人才的主要来路。Read: Across the whole population, the largest source is lateral movement among tech and healthcare companies; but the two clinical upstreams—clinical practice (343 people) and academia/medical school (671 people)—are the signature pipelines that distinguish medical AI from general AI, and the main route for bilingual talent.
读数：双语人才的来源结构与全体明显不同，「临床执业」与「学术/医学院」两条管道的占比显著更高。这印证了核心判断：双语人才不是 ML 工程师学医学出来的，而是临床人/医学研究者跨进了 AI。招聘要去医院信息科、学术医学中心的 informatics 实验室、影像/病理科找人，而不是只在科技公司挖。Read: The source mix for bilingual talent differs markedly from the overall population—the clinical-practice and academia/medical-school pipelines carry a significantly larger share. This confirms the core thesis: bilingual talent are not ML engineers who learned medicine, but clinicians and medical researchers who crossed into AI. Recruit from hospital informatics departments, the informatics labs of academic medical centers, and radiology/pathology—not just from tech companies.
读数：2025 年现任员工入职 2217 人，是 2023 年的 3.0 倍，医疗 AI 的招聘扩张与 ambient/LLM 产品落地同步发生在 2024-2025。Read: 2217 current employees joined in 2025—3.0× the 2023 figure—so medical AI's hiring expansion ran in lockstep with the 2024–2025 rollout of ambient and LLM products.
用全量档案还原两类标杆机构的团队梯队：一家纯医疗 AI 公司、一家医院系统 AI 团队。层级按 title 推断，非官方架构；公开版人名默认模糊。Using full profiles to reconstruct the team ladders of two benchmark organizations: a pure medical-AI company and a hospital-system AI team. Levels are inferred from titles, not official org charts; names are masked by default in the public version.
纯医疗 AI 公司：临床/医学事务与 ML 并置，靠双语人才把临床判断接进产品。A pure medical-AI company: Clinical/Medical Affairs sits alongside ML, relying on bilingual talent to wire clinical judgment into the product.
医院系统 AI 团队：以临床信息学为核心，连接医生与数据科学，自建模型并嵌入 EHR 工作流。A hospital-system AI team: built around clinical informatics, connecting physicians with data science, building models in-house and embedding them in EHR workflows.
从全量画像中挑出 22 位代表性人物，展示「临床×ML 双语人才」的真实形态与履历路径。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。公开版人名默认模糊，填写信息后解锁。From the full profile set we picked 22 representative individuals to show the real shape and career paths of clinical×ML bilingual talent. Profile facts come from the Metix AI database; those marked "publicly verified" have had their current roles confirmed against 2025–2026 public sources. Names are masked by default in the public version and unlock once you submit your details.
双语人才招聘的核心张力：临床执业收入高，转入医疗 AI 多为降薪。理解这笔账，才知道用什么补。数字为 2026-06 公开市场口径，非个案承诺。The central tension in hiring bilingual talent: clinical practice pays well, and moving into medical AI usually means a pay cut. Understand the math and you'll know what to offer in compensation. Figures reflect public market scope as of 2026-06, not individual offers.
| 角色 / 路径Role / path | 典型年收入 (TC)Typical annual TC | 口径与来源Scope and source |
|---|---|---|
| 放射科医生（临床执业）Radiologist (clinical practice) | $571K（中位，2025）$571K (median, 2025) | Medscape 薪酬报告，2025 年 +9% YoY，影像科是最早接触 AI 的高薪专科Medscape compensation report; +9% YoY in 2025. Imaging is the high-paying specialty that encountered AI earliest. |
| 医疗 AI 公司 ML / 数据科学Medical-AI company ML / data science | $150K-270K | levels.fyi 口径；热门赛道（Abridge 等）可达 $320K+levels.fyi scope; hot players (Abridge, etc.) can reach $320K+ |
| Medical Director（医疗 AI 公司）Medical Director (medical-AI company) | 约 $300Kabout $300K | 医学事务负责人，常含 equityHead of medical affairs; often includes equity |
| 创业公司 CMO（首席医疗官）Startup CMO (Chief Medical Officer) | base $275-400K + equity | 成熟机构总包可 > $750K（含股权）Total package at mature companies can be > $750K (including equity) |
| 临床信息学医生（医院系统）Clinical-informatics physician (health system) | $250-400K | board 认证临床信息学，桥梁岗位Board-certified clinical informatics; a bridge role |
放射科医生临床年薪中位约 $571K，而医疗 AI 公司 ML/数据科学岗 TC 仅 $150-270K。现金机会成本巨大，因此「医生想离开临床」意愿虽高（调查 35-60%），真正全职转非临床的比例仅约 2%。主通道是兼职顾问 / advisor + 高管 title（用 equity 与影响力补现金差），而非全职转工程岗。A radiologist's median clinical pay is about $571K, while ML/data-science roles at medical-AI companies pay just $150–270K in TC. The cash opportunity cost is enormous, so although the appetite among physicians to leave the clinic is high (35–60% in surveys), only about 2% actually make a full-time move out of clinical work. The main channel is part-time consulting/advisor work plus an executive title (using equity and impact to close the cash gap), not a full-time switch to an engineering role.
吸引临床人转入的三件套：① equity 上行（早期医疗 AI 公司股权）；② 影响力规模（一个模型影响百万患者 vs 一天看几十个病人）；③ 摆脱职业倦怠与夜班（MGB 数据显示 ambient 工具使医生 burnout 降约 40%、60% 愿延长职业生涯）。对早期职业、informatics fellow、低薪专科的吸引力远高于高薪专科在职医生。The three-piece kit that draws clinicians in: ① equity upside (stock in early-stage medical-AI companies); ② scale of impact (one model touching millions of patients vs. seeing a few dozen patients a day); ③ escaping burnout and night shifts (MGB data show ambient tools cut physician burnout by about 40%, with 60% willing to extend their careers). The appeal is far greater for early-career professionals, informatics fellows, and lower-paid specialties than for high-earning practicing physicians.
把这张地图变成招聘动作：去哪找、怎么转化、用什么留。Turning this map into hiring action: where to find them, how to convert them, and what keeps them.
① 来源管道（第 5 节）指向最粗的进水管，临床执业与学术医学中心，而非科技公司；② 学术医学中心的 informatics 实验室（Stanford AIMI 已孵化 10 家医疗 AI 公司、Harvard/MGB、Mayo、Vanderbilt）是高产节点；③ 影像/病理科是最早转型的临床专科，双语密度最高；④ 高浓度机构（第 4 节）的在职者本身就是画像样本。① The source pipelines (Section 5) point to the widest intakes—clinical practice and academic medical centers, not tech companies; ② the informatics labs of academic medical centers (Stanford AIMI has already spun out 10 medical-AI companies; Harvard/MGB, Mayo, Vanderbilt) are highly productive nodes; ③ radiology/pathology are the earliest clinical specialties to pivot and carry the highest bilingual density; ④ the current staff of high-density organizations (Section 4) are themselves a profile sample.
① 优先早期职业、informatics fellow、低薪专科，而非高薪在职专科（现金差太大）；② 主推兼职顾问 / 医学顾问入口，降低转换门槛，再谈全职；③ 用 equity + 影响力规模 + 摆脱 burnout 三件套补现金差，纯薪资对标必败；④ 渠道走 AMIA、RSNA、临床信息学 fellowship 圈层与 LinkedIn 信号，而非通用技术招聘渠道。① Prioritize early-career professionals, informatics fellows, and lower-paid specialties over high-earning practicing specialists (the cash gap is too wide); ② lead with a part-time consulting / medical-advisor entry point to lower the switching barrier, then discuss full-time; ③ close the cash gap with the three-piece kit of equity + scale of impact + escaping burnout—a pure salary match is bound to lose; ④ source through AMIA, RSNA, clinical-informatics fellowship circles, and LinkedIn signals, not generic tech-recruiting channels.
① 不要只在 ML 岗找双语人才，临床/医学事务、临床信息学岗同样是栖息地（第 4.3 节）；② 用「临床信息学」岗位作为连接医生与数据科学的桥梁角色；③ 招聘选址贴近临床中心而非只盯湾区（第 3.3 节地理分散）；④ 留人靠真实临床影响力与产品话语权，这是医院系统和厂商共同争抢同一批人时的差异点。① Don't look for bilingual talent only in ML roles—Clinical/Medical Affairs and Clinical Informatics are just as much a habitat (Section 4.3); ② use the clinical-informatics role as the bridge connecting physicians with data science; ③ site recruiting near clinical centers rather than fixating on the Bay Area (the geographic dispersion in Section 3.3); ④ retention rests on real clinical impact and product influence—the differentiator when health systems and vendors fight over the same people.
本报告的检索、画像、来源管道分析全部由 Metix AI 完成。可按同样口径为任意医疗 AI 岗位生成定制人才地图：双语人才长名单导出、临床背景核验、来源管道定位、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 the search, profiling, and source-pipeline analysis in this report were done by Metix AI. Using the same methodology, we can generate a custom talent map for any medical-AI role: export the bilingual-talent long list, verify clinical backgrounds, locate source pipelines, unlock emails, and run multi-channel outreach—billed on a "pay only for qualified interviews" basis. No interview, no charge.
8.6 亿+ 全球人才画像860M+ global talent profiles1,077 名临床×ML 双语人才1,077 clinical×ML bilingual professionals临床背景核验Clinical-background verification只为合格面试付费Pay only for qualified interviews纯医疗 AI 公司（Tempus、PathAI、Viz.ai、Abridge、Ambience、Aidoc、Hippocratic、OpenEvidence、Suki、Innovaccer、Commure/Athelas、Cleerly、Cohere Health、Notable、Qventus、Regard）为全员检索；医院系统（Mayo、Kaiser、Mass General Brigham、Cleveland Clinic、HCA）与 Nuance(Microsoft) 为 AI/数据/信息学职能子集，绝对数偏保守。范围 = 档案常驻美国。Pure medical-AI companies (Tempus, PathAI, Viz.ai, Abridge, Ambience, Aidoc, Hippocratic, OpenEvidence, Suki, Innovaccer, Commure/Athelas, Cleerly, Cohere Health, Notable, Qventus, Regard) are searched at full headcount; health systems (Mayo, Kaiser, Mass General Brigham, Cleveland Clinic, HCA) and Nuance (Microsoft) reflect only the AI/data/informatics subset, so their absolute numbers are conservative. Scope = profiles based in the U.S.
满足任一即计入：① 医学/护理/药学学位（MD/DO/MBBS/RN/NP/PharmD/DNP 等）；② 临床职务（医生/护士/药师/放射/病理/医学总监/临床信息学等）；③ 履历含医院/诊所/医疗系统等临床执业机构。学位为高置信，职务或执业经历为中置信。Counted if any of the following holds: ① a medical/nursing/pharmacy degree (MD/DO/MBBS/RN/NP/PharmD/DNP, etc.); ② a clinical role (physician/nurse/pharmacist/radiology/pathology/medical director/clinical informatics, etc.); ③ a résumé including a clinical-practice institution such as a hospital/clinic/health system. A degree is high-confidence; a role or practice history is medium-confidence.
满足任一即计入：ML/数据科学/研究科学 title，或档案技能含机器学习/深度学习/NLP/计算机视觉/数据科学等。双语人才 = 同时满足临床背景与 ML 能力。Counted if any of the following holds: an ML/data-science/research-scientist title, or profile skills including machine learning/deep learning/NLP/computer vision/data science, etc. Bilingual talent = meeting both the clinical-background and ML-capability criteria.
数据截至 2026 年上半年；档案更新存在滞后，代表性人物已逐人对照公开信息复核，2025-2026 的最新职位变动以公开信源为准标注。Data through H1 2026; profile updates lag, so representative individuals have each been cross-checked against public information, and the latest 2025–2026 role changes are annotated per public sources.
| 机构Organization | 在职画像Current profiles | 临床背景Clinical background | 临床占比Clinical share | 临床×ML 双语Clinical×ML bilingual | 双语占比Bilingual share |
|---|---|---|---|---|---|
| Tempus AI | 2,067 | 498 | 24.1% | 86 | 4.2% |
| Mayo Clinic | 705 | 705 | 100.0% | 414 | 58.7% |
| Commure / Athelas | 493 | 61 | 12.4% | 13 | 2.6% |
| Abridge | 436 | 84 | 19.3% | 14 | 3.2% |
| Kaiser Permanente | 376 | 376 | 100.0% | 216 | 57.4% |
| Innovaccer | 337 | 48 | 14.2% | 5 | 1.5% |
| Cleveland Clinic | 250 | 250 | 100.0% | 137 | 54.8% |
| PathAI | 239 | 71 | 29.7% | 12 | 5.0% |
| Cleerly | 206 | 57 | 27.7% | 8 | 3.9% |
| HCA Healthcare | 190 | 18 | 9.5% | 13 | 6.8% |
| Mass General Brigham | 178 | 178 | 100.0% | 102 | 57.3% |
| Suki AI | 177 | 32 | 18.1% | 5 | 2.8% |
| Viz.ai | 165 | 56 | 33.9% | 6 | 3.6% |
| Aidoc | 155 | 44 | 28.4% | 6 | 3.9% |
| Hippocratic AI | 102 | 51 | 50.0% | 15 | 14.7% |
| Regard | 99 | 18 | 18.2% | 3 | 3.0% |
| OpenEvidence | 94 | 21 | 22.3% | 9 | 9.6% |
| Notable Health | 48 | 7 | 14.6% | 1 | 2.1% |
| Cohere Health | 47 | 11 | 23.4% | 2 | 4.3% |
| Nuance (Microsoft) | 31 | 7 | 22.6% | 2 | 6.5% |
| Ambience Healthcare | 18 | 12 | 66.7% | 8 | 44.4% |
| Qventus | 2 | 1 | 50.0% | 0 | 0.0% |
① 覆盖率Coverage：纯医疗 AI 公司为全员，医院系统/EHR 巨头为可识别 AI/数据/信息学子集（其庞大临床主体不计入），机构间比较以占比与结构为主、绝对数为辅。: pure medical-AI companies are at full headcount, while health systems and EHR giants reflect only the identifiable AI/data/informatics subset (their vast clinical core is excluded), so cross-organization comparisons rely mainly on shares and structure, with absolute numbers secondary.
② 临床背景为概率判定Clinical background is a probabilistic determination：基于学位/职务/雇主关键词；不在档案标注学位的临床人会漏检，「双语人才」为下限口径。: based on degree/role/employer keywords; clinicians who don't list a degree on their profile are missed, so "bilingual talent" is a lower-bound count.
③ 职能与层级推断Function and level are inferred：基于 title/headline 关键词，title 模糊者可能误分。: based on title/headline keywords; people with vague titles may be misclassified.
④ 数据时效Data timeliness：静态快照，近 1-2 个季度变动有滞后；代表性人物已逐人复核。: a static snapshot, with movements in the last 1–2 quarters lagging; representative individuals have each been re-checked.
⑤ 本报告基于公开职业档案聚合，各项为数据库口径，宜与机构公开编制互为参照。⑤ This report aggregates public career profiles; all figures are database-scope and are best read alongside organizations' publicly reported headcounts.