初创公司尽调
尽调报告 healthcare / biotech growth-stage private unicorn 2026-06-19

Owkin

面向生物制药 R&D 的联邦学习与 agentic AI 平台,由 Sanofi 和 BMS 合作关系锚定

Owkin 拿到了顶级药企的真实战略验证,也有差异化的联邦数据入口,K Pro 商业化信号也在改善;但财务长期不透明,新股估值又没有刷新,今天更适合观察而不是买入。

封面要素

最近披露估值 01
1 USD B [CV004]
累计融资 02
304.1 USD M [CV003]
FY2024 收入 03
86 USD M [CV006]
FY2024 收入增速 04
83 % [CV006]
2024 年末现金 05
200 USD M [CV008]

公司概况

Owkin 是一家总部在巴黎的 AI 生物技术公司,2016 年由肿瘤科医生 Thomas Clozel 和机器学习科学家 Gilles Wainrib 创立。公司早期在学术医院网络中搭建保护隐私的联邦学习壁垒,随后扩展到多模态病理 / 基因组模型,并在 2025-2026 年推出由 Owkin Zero 生物推理模型驱动的 agentic AI 平台 K Pro。Owkin 已融资约 $304.1 million,2021 年估值达到 $1 billion,并加深了与 Sanofi、BMS 和 AstraZeneca 的战略关系;但公司财务仍不透明,最新商业牵引也尚未经过新一轮一级市场融资验证。

官网
owkin.com
成立时间
2016-01-01
创始人
Thomas Clozel, Gilles Wainrib
创立地点
Paris, France
总部
Paris, France
产品
Owkin 销售面向生物制药 R&D 的 AI 基础设施和应用,核心是 K Pro、Owkin Zero、联邦学习工作流,以及多模态病理 / 基因组工具。原诊断产品组合(包括 MSIntuit CRC 和 RlapsRisk BC)已在 2026 年 3 月拆分至 Waiv。
客户
大型制药公司、转化研究团队,以及为联邦数据网络贡献数据的学术医疗中心。
商业模式
多年期企业软件与研究协作模式,结合药企平台许可、基于里程碑的 R&D 付款,以及战略共同开发协议。
阶段
growth-stage private unicorn
融资情况
迄今已融资约 $304.1 million;母公司最近披露的一级市场融资是 2022 年 6 月 Bristol-Myers Squibb 投入的 $80 million Series B 延展轮,此前是 2021 年 Sanofi 领投的 $180 million Series B。据报道,公司在 2025 年末寻求新一轮 $200-300 million 融资,但截至 2026 年 6 月尚未确认交割。
[CO001, CO002, CO006, CO007, CO016, CO018, CO020, CO021]

执行摘要

主要优势

  • Sanofi、Bristol-Myers Squibb 和 AstraZeneca 给了深度战略验证,合作是多年平台关系,不是一次性试点。
  • 通过大型医院网络接入联邦学习数据,形成差异化数据护城河,比纯软件 UX 更难复制。
  • 2025-2026 年产品重心收敛到 K Pro 和 Owkin Zero,相比传统诊断业务,可能提高经常性软件收入质量。

主要风险

  • 最后一次确认的新股估值仍是 2021 年 $1 billion 标记,投资人缺少转型后业务的当前价格信号。
  • 收入质量、毛利率、烧钱速度和客户集中度仍不透明,无法严谨承销下行情景和现金跑道。
  • Sanofi 同时是 Owkin 的锚定投资人、旗舰客户和声誉背书方,集中度风险过高。
  • 2025 年裁员和员工对管理执行的批评,让人担心高风险战略转型期的运营纪律。

未决问题

  • 当前经审计财务报表、烧钱速度和 24 个月现金流预测未公开。
  • 合同层面的经常性收入、续约条款和按收入计算的客户集中度未披露。
  • 自 2021/2022 年 Series B 轮以来,没有确认的新钱估值发布。
  • 董事会构成、继任规划和治理控制披露仍很薄。

目录

Chapter 01

01公司概览

1.1 身份与商业模式

Owkin 是一家法国 - 美国私营 AI 生物技术公司,2016 年创立于巴黎,并按法国法律(SAS)注册。公司总部仍在法国巴黎,同时在美国纽约有重要布局。公司将使命描述为实现 “Biological Artificial Superintelligence”(BASI)——一种能够在人类认知之外建模并工程化生物学的 AI,用来加速药物发现和疗法开发。 核心商业模式是面向大型制药公司的企业 SaaS 和长期 R&D 协作协议。Owkin 将 AI 平台嵌入药企在发现、转化和临床阶段的 R&D 工作流,通过多年期战略合作收费,而不是单纯的交易型授权。关键客户包括 Sanofi、Bristol-Myers Squibb、AstraZeneca、Amgen、Roche、Ipsen、Johnson & Johnson 等大型生物制药公司,以及参与 Owkin 联邦数据网络的领先学术医疗中心。 截至 2026 年 6 月,公司的旗舰产品是 K Pro,一款面向生物制药、于 2025 年 10 月发布的 agentic AI 科学家副驾驶。K Pro 建在 Owkin Zero 这一自有生物大语言模型之上,并整合多模态患者数据集(包括 MOSAIC 肿瘤数据集),让研究人员能够大规模查询生物学问题。更早的产品线——主要是 MSIntuit CRC 和 RlapsRisk BC,两款基于 AI、带 CE 标志的结直肠癌和乳腺癌诊断产品——已在 2026 年 3 月拆分到名为 Waiv 的独立公司,使 Owkin 能完全聚焦于药物 R&D 的 agentic AI。Owkin 的联邦学习基础设施已部署在 18 家领先学术研究医院,支撑其专有数据壁垒和保护隐私的 AI 模型训练。 [CO001, CO002, CO003, CO004, CO005, CO032]

Owkin 快照 KPI 表
指标数值 / 状态日期置信度缺口 / 备注
成立年份20162016
总部法国巴黎(另有美国纽约办公室)2026-06-19
阶段私有独角兽,Series B2026-06-192022 年 6 月以来无新的主融资
最近披露估值(USD B)1.02021-11上一轮主融资;二级市场显示最高可达 $1.6B
累计融资额(USD M)304.12026-06-19据 Owkin FAQ
FY2024 收入(USD M)862024-12-31法国监管披露;ARR 估计存在分歧
收入同比增长83%2024-12-31Maddyness / 法国申报;FY2023 基数不清楚
FY2024 净亏损(USD M)5.62024-12-31低于 2023 年 $31.3M 亏损
手头现金(USD M)~2002024-12-31依据法国申报估计;未获独立确认
员工数~450–5002026-062025 年裁员后;未公布官方员工数
毛利率未披露;数据室尽调问题
ARR(USD M)仅披露 2023 年数据(Latka 称 $101M,与 2024 年收入冲突)
客户数未公开披露;尽调问题
CEOThomas Clozel, MD2026-06-19创始人兼 CEO;关键人风险

收入、净亏损和现金数字来自法国监管披露和媒体摘要,而非经审计年报;在获得数据室访问前,所有财务指标都应视为指示性数据。 估值反映上一轮主融资;二级市场数据未经核验。空单元格代表未披露的私有指标。

[CO001, CO002, CO016, CO017, CO027, CO028]
FO002: Owkin 公司快照逻辑

Owkin 的价值链从医院患者数据网络,经由联邦 AI 平台流向药企 R&D 加速;Waiv 分拆后作为独立下游应用承接诊断。

[CO004, CO005, CO033, CO034, CO035, CO036]

1.2 创始人与领导层

Owkin 由 Thomas Clozel, MD 和 Gilles Wainrib, PhD 于 2016 年共同创立。一位临床肿瘤科医生与一位机器学习研究者的组合,体现了公司创立时的判断:生物医学 AI 需要把临床知识与数据科学深度接合。 CEO Thomas Clozel 接受过肿瘤学和血液学训练,截至 2026 年 6 月仍留任。他是公司的公开面孔和主要战略权威,也带来显著关键人依赖。联合创始人兼首席科学官 Gilles Wainrib, PhD 是机器学习教授,继续负责科学战略。 公开披露的更广泛 C-suite 包括:Pascal Weinberger(副 CEO)、Graeme Bell(CFO)、Rodrigo Barnes(CTO)、Caroline Hoffmann(首席医疗官)、Antonia Trower(首席营销官)、Marie-Alice Tantardini(首席人力官)、Jean-Frédéric Petit-Nivard(首席商务官)、Anna Huyghues-Despointes(首席企业发展与运营战略官)以及 Anne-Charlotte Paut(首席法务官)。 关键人风险集中在 Thomas Clozel 身上:他推动投资者关系、大型药企合作谈判,以及面向 BASI 的战略方向。公开信息中没有继任计划迹象。Gilles Wainrib 同时担任联合创始人与学术科学家,也形成中等依赖。2025 年 9 月裁员以及随之出现的对 “huge managerial issues” 的内部批评,提示组织执行缺口,需要持续观察。 [CO006, CO007, CO008, CO009, CO010, CO011]

领导层与创始人表
人员职务背景创始人-市场契合 / 职能覆盖关键人依赖
Thomas Clozel, MD联合创始人兼 CEO临床肿瘤学家和血液学家;联合创立 Owkin 前接受医学训练深厚临床领域专长托住药企合作;唯一具备投资者沟通权威的创始人高——投资者关系、大型药企交易和战略集中在一人身上
Gilles Wainrib, PhD联合创始人兼首席科学官机器学习教授;深度学习研究的学术背景在药企和学术合作中提供技术可信度;领导 AI 研究战略中——科学方向和学术可信度集中在联合创始人角色
Pascal Weinberger副 CEO资深运营高管;支持 CEO 处理日常业务提供管理连续性;降低 CEO 单点运营风险
Graeme BellCFO财务高管,管理 Owkin 资本结构和投资者关系负责财务报告、资金管理和融资流程中——当前寻求资本阶段,融资至关重要
Rodrigo BarnesCTO技术负责人,统筹 Owkin 工程和平台架构领导 K Pro 平台开发和云基础设施中——K Pro 平台是核心商业资产
Caroline Hoffmann首席医疗官医疗和临床事务负责人,为药企合作提供可信度连接临床证据和 AI 工具商业部署低到中

领导层数据汇总自 Owkin 官方 about-us 页面、新闻披露和第三方高管目录(craft.co)。联合创始人之外的职务按公开报道列示, 可能无法反映完整高管团队。关键人依赖评级为作者根据角色关键性作出的判断。

[CO006, CO007, CO008, CO009, CO010, CO011]

1.3 融资、投资者与估值

Owkin 自 2016 年种子轮以来累计股权融资约 $304.1 million,这一点由公司自己的 FAQ 确认。资本结构包括 $2.1 million 种子轮(2016 年 11 月)、累计约 $72 million 的 Series A(2018–2020 年),以及 2021 年 11 月由 Sanofi 领投的 $180 million Series B;后者使公司以 $1 billion 的投后估值进入独角兽行列。2022 年 6 月,Bristol-Myers Squibb 以 $80 million 战略投资延展 Series B,并附带最高 $100 million 里程碑付款——实际上把 BMS 合作总价值最高推至 $180 million。 投资者刻意混合了财务 VC 和战略药企公司投资者。财务支持方包括 GV(Google Ventures)、Bpifrance、Mubadala Investment Company、F-Prime Capital、Cathay Innovation、Eight Roads Ventures、NJF Capital、MACSF(法国医疗从业者养老基金)和 Otium Capital。战略公司投资者是 Sanofi 和 Bristol-Myers Squibb,它们参与 Series B 既是资本背书,也传递了长期采用技术的意图。 最近披露的 $1 billion 投后估值来自 2021 年 11 月。此后没有公开说明的新一轮一级市场估值。PremierAlts 截至 2026 年的二级市场数据提示估值最高可达 $1.6 billion,但该数字来自二级交易聚合平台,无法独立核验。截至 2025 年末,据报道 Owkin 正在寻求 $200–$300 million 新融资,这将意味着估值明显上台阶,但截至运行日期尚未宣布交割。 [CO016, CO017, CO018, CO019, CO020, CO021]

利益相关方或投资者地图
利益相关方角色控制权 / 经济重要性尽调问题
SanofiSeries B 领投方($180M,2021 年 11 月);K Pro 多年期合作伙伴(2026 年 6 月)最大单笔股权支票;战略协同最深;2026 年 6 月扩展协议显示持续投入确认 K Pro 协议的确切持股比例和共同开发 IP 归属条款
Bristol-Myers Squibb (BMS)战略投资者($80M,2022 年 6 月);临床数据科学合作伙伴第二大企业投资者;最高 $100M 里程碑付款;聚焦心血管临床试验确认里程碑进展、IP 归属,以及任何延期谈判的范围
GV (Google Ventures)财务 VC;Series A 领投共同投资方(2018)最早机构轮以来的重要股权;Google 协同提供平台可信度确认当前持股;评估是否考虑 Google Cloud 或 Alphabet 平台集成
Bpifrance Large Ventures法国公共 VC 和创新银行;Series A 投资者法国国家支持的财务锚点;打开法国监管和公共部门网络确认治理权;评估 Bpifrance 对未来融资是否拥有任何批准权
Mubadala Investment Company阿布扎比主权财富基金;Series A 投资者国际主权资本;为中东扩张提供地理可信度确认当前持股规模及任何优先权或共同投资协议
F-Prime Capital聚焦医疗健康的 VC(Fidelity 关联方);Series A 投资者深厚医疗健康投资经验;增强行业可信度和 LP 基础确认按比例跟投权及下一轮预计参与情况
Cathay Innovation中法技术 VC;Series A 投资者提供亚太市场通道和跨境 LP 关系评估其对 Owkin 中国或 APAC 数据战略的相关性
MACSF法国医疗从业者互助养老基金;Series A 投资者拥有直接临床用户社区协同的机构锚点评估 MACSF 临床网络是否提供数据共享或分发优势
OTB Ventures / Alpha Intelligence Capital(投资者)Waiv $33M Series A 领投方(2026 年 3 月);不再是 Owkin 股东这些投资者持有 Waiv 股权;Owkin 保留 Waiv 战略股权确认 Owkin 在 Waiv 的剩余股权,以及剥离后的数据共享条款

多数投资者的持股规模未公开披露;经济重要性反映轮次规模和战略关系,而非经验证的持股比例。Waiv 投资者直接持有 Waiv 股权, 而不是 Owkin 股权。Owkin 未公开披露剥离后自己在 Waiv 的持股。

[CO020, CO021, CO023, CO024, CO025, CO026]

1.4 规模指标与财务表现

Owkin 是私营公司,不公开披露合并财务。可得最细数据来自 Maddyness 引用的法国监管披露和媒体报道。2024 财年收入约 $86 million,较 2023 年同比增长 83%,也是公司首次跨过 $80 million 门槛。尽管增长迅速,Owkin 仍未盈利,2024 年净财务亏损为 $5.6 million——较 2023 年 $31.3 million 亏损显著改善。 公司 2024 年末现金约 $200 million,提供了有意义的跑道,短期内没有迫在眉睫的融资危机。2025 年末裁减约 50 名员工(约占员工总数 15%)后,员工数估计约 450–500 人。确切客户数量未公开;已知企业账户包括 Sanofi、BMS、AstraZeneca、Amgen、Roche、Ipsen 和 Johnson & Johnson,另有 18 家学术研究医院合作伙伴。 Latka 的第三方数据估计 2023 年 ARR 约 $101 million,相比法国来源报道的 2024 年收入 $86 million 显得偏高。这一差异尚未被调和,应视为尽调缺口。Owkin 不披露毛利率、净收入留存率或按细分划分的客户数量。公司目标是在五到六年内盈利(2025 年末表述),这意味着在当前水平下仍会继续消耗现金。 [CO027, CO028, CO029, CO030, CO031, CO040]

FO003: 快照 KPI

Owkin 公开可支撑的 KPI 显示:公司累计融资 $304M、估值 $1B,2024 年收入 $86M 且亏损收窄; 但客户数量、毛利率和 NRR 均未披露。

[CO016, CO017, CO027, CO028, CO029, CO030]

1.5 里程碑、合作关系与负面事件

Owkin 的公司时间线从 2016 年获得种子融资创立开始,到 2021 年成为独角兽,随后完成两项重大药企战略交易,在 CE 认证 AI 诊断上取得重要监管里程碑,并在 2025–2026 年加速转向 agentic AI。 最重要的近期里程碑包括:2025 年 10 月发布 K Pro(首个商业部署的生物制药 agentic AI 副驾驶,建立在 Owkin Zero 之上);2026 年 3 月将诊断部门(Owkin Dx)拆分为 Waiv,并获得 $33 million 新融资;2026 年 6 月与 Sanofi 扩大多年期 K Pro 合作,显示企业 AI 服务关系进一步加深。 负面一侧,2025 年 9 月公司启动集体裁员计划,削减约 50 名员工。Sifted 报道称,离职与在职员工提到 “huge managerial issues”,CEO Thomas Clozel 也公开承认,转型期间 HR 执行并不完美。重组明确与重新聚焦纯 AI 的战略有关,Owkin 同时还计划剥离后来成为 Waiv 的诊断部门。2024 年 $5.6 million 净亏损以及仍需 $200–$300 million 新融资,进一步强化了财务风险画像。截至运行日期,公司本身未发现诉讼、制裁或监管负面事件。Waiv 单独在 2026 年根据 EU IVDR 法规为基于 AI 的乳腺癌和结直肠癌检测取得双 CE 标志,但 Waiv 现在已是独立实体。 [CO018, CO020, CO021, CO032, CO037, CO038]

里程碑表
日期事件类型金额 / 估值 / 状态参与方含义
2016-11Owkin 成立;种子轮完成成立$2.1M 种子轮Thomas Clozel, Gilles Wainrib;天使投资者建立巴黎总部;为肿瘤学联邦 AI 研究提供第一笔资本
2018–2020Series A 轮(多个批次)融资累计 ~$72M投资者:GV、Bpifrance、Mubadala、MACSF、F-Prime、Cathay、Eight Roads、NJF、Otium扩建医院数据网络;推出联邦学习平台
2021-11Sanofi 领投 Series B;达到独角兽状态融资$180M;$1B 投后估值Sanofi(领投)、其他投资者第一个药企战略锚点;估值里程碑;启动肿瘤 AI 合作
2022-06Bristol-Myers Squibb 战略合作和 Series B 扩展轮融资$80M 首付款;最高 $100M 里程碑付款BMS, Owkin心血管临床试验 AI 合作;第二个主要药企锚定投资者
2023MSIntuit CRC 获得 CE-IVD 认证(首个用于 CRC 中 MSI 的 CE 标志 AI 诊断)监管CE-IVD 标志欧盟公告机构;Owkin Dx 部门Owkin 诊断 AI 获得监管验证;支持在欧洲临床部署
2025-05Owkin K(早期 agentic AI 平台)推出产品全面可用Owkin面向市场的首个生物医药研究者 agentic AI 助手
2025-09宣布集体裁员计划;约 50 名员工(15%)将离职反向裁减约 50 名员工Owkin;受影响员工战略重新聚焦纯 AI;内部批评管理处理方式
2025-10K Pro agentic AI 副驾驶推出(基于 Owkin Zero 生物 LLM 构建)产品全面可用Owkin生物医药 agentic AI 的商业旗舰;已获 AstraZeneca、BMS、Sanofi 验证
2026-03Waiv 从 Owkin 剥离,并完成 $33M Series A 融资治理Waiv 融资 $33M投资者:OTB Ventures、Alpha Intelligence Capital、Serena Data Ventures、Karista、SistaFund、CRB Health TechOwkin 退出诊断业务;聚焦药物发现 AI;Waiv 独立运营
2026-06宣布与 Sanofi 扩大的多年期 K Pro 合作合作多年期企业交易(条款未披露)Sanofi, Owkin加深战略协同;agentic AI 嵌入 Sanofi 完整 R&D 管线

只有年份或月份级数据公开时,日期为近似值。2018–2020 年 Series A 合并为一行;实际发生过多个不同批次。2023 年 CE-IVD 日期依据新闻报道近似。财务金额按报道列示;里程碑付款总额为合同上限,不等于已保证收款。

[CO018, CO019, CO020, CO021, CO032, CO037]
FO001: Owkin 公司里程碑时间线

Owkin 的公开记录始于 2016 年 oncology-AI 创立,随后在 2021 年成为独角兽,完成两笔药企战略交易、 CE 认证诊断,2025 年转向 agentic AI,并在 2026 年 6 月扩大 Sanofi 合作。

[CO018, CO020, CO032, CO038, CO041, CO043]

1.6 图表

Chapter 02

02市场分析

2.1 市场边界:Owkin 横跨 AI 药物发现、联邦学习和精准肿瘤诊断

抓取记录显示,理解 Owkin 最好的方式不是把它归入单一类别,而是看作三个相邻市场的交集。Owkin 自有页面描述了一款自主 “AI Scientist”(K Pro),用于自动化生物制药 R&D;训练方式是联邦学习,原始患者数据留在医院服务器,只有模型在站点之间流转。Wikipedia 记录其为一家法国 - 美国公司,2016 年由 Thomas Clozel 和 Gilles Wainrib 创立,产品包括 MSIntuit CRC 和 RlapsRisk BC,员工约 350 人。这把 Owkin 放进 AI 药物发现软件与服务市场;GMInsights 按组件、技术、应用、治疗领域和最终用户切分该市场。 边界纪律很重要,因为不应拿整个精准肿瘤池来给 Owkin 定规模。MarketsandMarkets 将更宽泛的 Drug Discovery Technologies 市场框定为 2025 年 USD 30.58 billion,但其中大部分支出是仪器、试剂和湿实验室流程,与 Owkin 的 AI 层无关。Owkin 替代的现状是人工病理审查、集中式数据汇集和传统高通量筛选;这些是应排除的基线,不是 Owkin 能捕获的收入。可防守的纳入支出是 AI 赋能的生物标志物发现、试验设计、保护隐私的模型训练,以及 MSIntuit 这类产品化诊断。[CM001, CM002, CM003, CM004, CM005, CM006]

Owkin 进入市场的定义与边界
细分 / 类别纳入支出 / 工作流排除支出 / 工作流买方 / 付款方与 Owkin 的相关性
药物发现 AI(软件与服务)靶点识别、先导化合物优化、生物标志物发现、试验设计湿实验室合成、化学试剂、制造制药与生物科技 R&DOwkin AI Scientist 的核心市场
联邦学习基础设施隐私保护的多站点模型训练、无需数据池化的模型交换通用云存储、集中式数据湖医院、学术中心、药企数据科学支撑技术和主要护城河
精准肿瘤诊断从 H&E 切片检测 MSI / 生物标志物、复发风险评分治疗制造、药物定价病理实验室、医疗系统、付款方通过 MSIntuit CRC 和 RlapsRisk BC 产品化
精准医疗 AI患者分层、疗法选择、预后非肿瘤消费健康、健康应用药企、医疗系统相邻扩张视角
现状替代方案手工病理审阅、集中式数据池化、传统高通量筛选不适用(基准)既有实验室和 CROOwkin 取代的排除基准

纳入 / 排除单元格是基于 Owkin 页面、GMInsights 分段和 MarketsandMarkets 范围得出的分析性解释,目的在于避免把 Owkin 放进完整精准肿瘤 TAM 中定规模。

[CM001, CM002, CM005, CM006, CM007, CM028]

2.2 市场规模:多重视角、相互冲突的 AI 估算,以及未披露的 Owkin SOM

没有单一 TAM 能定义 Owkin,因此本章用多个抓取视角交叉校准。AI 药物发现方面,Mordor Intelligence 估算市场在 2025 年为 USD 2.58 billion、2026 年为 USD 3.25 billion,并以 25.94% CAGR 到 2031 年达到 USD 10.29 billion;Fortune Business Insights 则给出 2025 年 USD 4.46 billion、2026 年 USD 5.00 billion,到 2034 年 USD 12.56 billion,CAGR 只有 12.20%。两家可信分析机构对增速的判断相差超过两倍,因此诚实读法是 2026 年约 USD 3.25-5.0 billion 的区间,而不是一个点估计。MarketsandMarkets 又给出 AI 精准医疗视角:2024 年 USD 0.78 billion,以 30.7% 增长至 2030 年。 使能技术层面,Precedence Research 将联邦学习市场估为 2026 年 USD 1.59 billion,CAGR 30.5%,医疗和生命科学约占四分之一。需求侧池子——精准肿瘤——大得多:Mordor 对 2026 年的估算是 USD 127.68 billion,Precedence 是 USD 146.24 billion,Fortune 是 USD 133.06 billion,集中在 USD 127-146 billion。Owkin 可服务的切片仍更窄,大致是聚焦肿瘤的药企 R&D 合作加欧洲 MSI 诊断;其真实 SOM 未公开,因此规模金字塔底层刻意留空。[CM008, CM009, CM010, CM011, CM012, CM013]

横跨 AI、联邦学习与精准肿瘤的 TAM、SAM 和测算视角
发布方 / 视角年份地区数值(USD bn)CAGR / 趋势方法置信度局限
Mordor 药物发现 AI2026全球3.2525.94% 至 2031 年分析师市场模型软件与服务口径;与 Fortune CAGR 冲突
Fortune 药物发现 AI2026全球512.20% 至 2034 年分析师市场模型同一市场的 CAGR 不到 Mordor 的一半
MarketsandMarkets 精准医疗 AI2024全球0.7830.7% 至 2030 年分析师市场模型范围比 Owkin 更宽的精准医疗口径
Precedence 联邦学习2026全球1.5930.5% 至 2035 年分析师市场模型跨行业;医疗健康约占 25%
Mordor 精准肿瘤2026全球127.689.53% 至 2031 年分析师市场模型治疗业务占比重的 TAM,不是 Owkin SAM
Precedence 精准肿瘤2026全球146.2411.33% 至 2035 年分析师市场模型宽口径肿瘤 TAM
MarketsandMarkets 药物发现技术2025全球30.5811.0% 至 2030 年分析师市场模型宽相邻口径,含仪器 / 试剂
Owkin 专属 SOM2026全球n/a推导 / 无抓取到的公开证据未披露 Owkin 专属可获得市场

数值列有意混合不同市场视角;底部行为空,是因为抓取到的来源没有单独隔离 Owkin 专属可获得市场。药物发现 AI 应使用区间, 不应依赖单个点估计。

[CM006, CM008, CM009, CM010, CM012, CM013]
FM001: 从精准肿瘤 TAM 到 Owkin 未披露 SOM 的市场规模视角

每向下收窄一层,都需要更多假设;Owkin-specific 可获得市场未公开披露,因此底部留作 null。

联邦学习层是分析转换(医疗约占 Precedence USD 1.59bn 市场的 25%); 底层为 null,因为没有已抓取来源单独切出 Owkin 的可获得市场。

[CM008, CM012, CM013, CM016, CM020]
FM002: 不同分析机构和年份下的 AI 药物发现市场区间

两家可信分析机构分歧很大,因此 AI 药物发现市场最好按年份呈现低 / 基准 / 高区间,单位均为 USD billions。

低值采用 Mordor,高值采用 Fortune Business Insights;基准值取中点。2031-2034 行跨越不同终点年份, 因为两家分析机构使用不同预测期,这正是需要保留的矛盾。

[CM008, CM009, CM011]

2.3 买方、用户与付款方地图:核心由药企 R&D 预算买单,诊断实验室是相邻通道

主要经济买方是制药 R&D 和业务发展部门。最清晰的证明是 Sanofi 2021 年 12 月投入 USD 180 million 使 Owkin 成为独角兽,同时配套一项价值 USD 90 million、覆盖四种癌症的三年合作和研究里程碑。这个买方背后是很大的资金池:Evaluate 数据显示 2024 年全球药企 R&D 接近 USD 288 billion,EFPIA 报告欧洲行业 2023 年 R&D 约 EUR 50 billion,Nature Reviews Drug Discovery 的一项分析将 2021 年全球生物制药 R&D 定位在 USD 276 billion。Owkin 累计融资超过 USD 255 million。买方内部的用户,是需要访问孤岛化、多机构数据的数据科学家、生物学家或临床试验设计者。 第二条通道经过病理实验室和医疗系统,服务于产品化诊断。MSIntuit CRC 是唯一用 H&E 切片做 MSI 检测、带 CE 标志的 AI 测试,瞄准病理医生短缺和免疫疗法匹配;预算所有者从 R&D 转向实验室,并最终转向付款方报销。联邦学习的最终用户集中在医疗服务提供方和药企,约占该市场四分之一,北美占主导。Roche、Guardant、Tempus、Caris 和 Illumina 等精准肿瘤既有玩家,勾勒出 Owkin 最终必须进入的面向付款方竞争格局。[CM021, CM022, CM023, CM024, CM025, CM026]

Owkin 市场的细分、买方、用户与付款方地图
细分买方用户付款方解决的工作流预算所有者采用触发因素
药企肿瘤 R&D药企 R&D / 业务发展负责人数据科学家和生物学家药企 R&D 预算生物标志物发现、靶点验证、试验设计R&D 和公司发展需要孤岛化的多站点患者数据
联邦数据网络医院或学术医疗中心医院数据科学家研究经费 / 合作资金隐私保护模型训练研究办公室GDPR/HIPAA 下的数据主权
精准诊断病理实验室或医院病理医生支付方 / 医疗体系从 H&E 切片检测 MSI实验室预算病理医生短缺与免疫治疗匹配
生物医药联合开发药企高管发起人(如 Sanofi)临床与转化团队股权加里程碑付款跨癌种组合提速公司发展管线缺口与 AI 差异化
医疗体系(未来)医疗服务网络主治临床医生支付方报销诊疗现场部署诊断临床运营AI 检测报销路径清晰度

这张图把经济买方、用户和支付方拆开看,因为 Owkin 已跑通的收入路径是药企研发合作;诊断渠道仍卡在实验室预算和未经验证的报销上。

[CM021, CM022, CM024, CM027, CM028, CM029]
FM003: Owkin 两条收入渠道的买方、用户和付款方流向

药企 R&D 为核心联邦学习渠道买单,病理实验室和支付方支撑诊断渠道;两条路径最终都汇向患者结果。

该流程压缩了两条不同的商业化路线;Owkin 已公开证明的节点是由药企资助的联邦渠道,诊断渠道仍取决于实验室预算和报销。

[CM021, CM027, CM028, CM029, CM031]

2.4 增长驱动与采用约束:隐私顺风对上验证、监管和执行风险

多个抓取到的驱动因素支撑长期需求。GDPR 和 HIPAA 等隐私监管把医院和药企推向联邦学习,因为它能在不移动原始数据的情况下训练模型;Precedence 将其列为市场首要驱动。癌症负担是结构性的:WHO 记录 2022 年癌症死亡接近 1000 万,单是结直肠癌在 2020 年约 200 万病例中就造成约 100 万死亡。AI 和生物学模型也在成熟,Nature Biotechnology 对大规模扰动筛选的报道就是例证;FDA 持续批准 AI/ML 赋能医疗器械,也在降低商业诊断路径风险。 约束同样具体。临床验证很慢:MSIntuit 在获批前需要一项盲法 Nature Communications 研究证明 96% 敏感性,而报销仍未验证。EU AI Act 和 EMA 反思文件施加基于风险的制度,临床 AI 被视为高风险并需要 CE 标志,即便仅用于药物 R&D 的 AI 可豁免,这带来双重合规复杂度。联邦学习自身也有技术摩擦,来自 non-IID 数据异质性和单站点高基础设施成本。公司特定风险也不小:Owkin 在 2023 年裁员,并转向 AI agents 模式,释放出战略和执行不确定性,单靠市场规模无法消化这一点。[CM030, CM032, CM033, CM034, CM035, CM036]

增长驱动因素与采用约束
驱动因素 / 约束方向时间影响尽调问题
隐私监管(GDPR/HIPAA)正向当前让联邦学习比中心化数据交易更占优相比直接数据授权,联邦学习护城河有多能守?
癌症负担上升正向当前肿瘤生物标志物和诊断需求长期存在哪些瘤种支撑 Owkin 管线价值?
AI 与生物学模型成熟正向2024-2026提高模型准确性和可信度Owkin 持有哪些专有多模态数据优势?
活跃的 FDA AI/ML 器械路径正向当前降低诊断商业化风险Owkin 的 MSIntuit FDA 申报时间表是什么?
EU AI Act / EMA 监管混合2026-2027临床 AI 属高风险 / CE;仅用于研发的 AI 获豁免Owkin 哪些产品落入高风险制度?
漫长的临床验证周期负向持续拖慢从模型到检测的收入转化从模型到可报销诊断要多久?
公司转向与 2023 年裁员负向2023-2025战略和执行风险当前烧钱速度、现金跑道和员工数是多少?
联邦数据异质性与基础设施成本负向持续推高单个站点接入成本和时间每家医院接入要花多少钱、多久?

方向指对品类采用的影响,而不是对 Owkin 份额的影响;顺风因素和卡点放在一起,因为两者共同决定现实的商业化节奏。

[CM030, CM032, CM033, CM034, CM035, CM036]
FM004: 从数据访问到可报销临床使用的采用漏斗

Owkin 的采用路径从广泛的数据访问需求收窄到少数能通过监管和报销关口的产品。

阶段是顺序关系,不是实测转化率;它们概括了已抓取来源描述的从验证到报销路径。

[CM030, CM033, CM034, CM035, CM028]

2.5 规模与采用尽调缺口:相互冲突的分析师估算和未披露的 Owkin SOM

最重要的市场结论,是公开记录无法支撑什么。抓取证据没有披露 Owkin 自身收入、run-rate、Sanofi 标题之外的合作经济性,以及任何可服务可获取市场数字;因此本章所有美元视角都是自上而下的 TAM,而不是 Owkin 特定的 SOM。正因为有这个缺口,规模金字塔才把底层留空,后续估值工作也必须要求管理层提供内部切分。 两个结构性矛盾被保留下来,而不是被平均掉。第一,可信分析机构对 AI 药物发现增长分歧很大,Mordor 的 25.94% CAGR 是 Fortune 12.20% 的两倍多,仅这一项就能让 2031-2034 年市场规模相差数十亿美元。第二,诊断商业化路径取决于报销和付款方覆盖,但没有抓取来源确认 MSIntuit 已获得这些支持,第二条收入通道仍未证明。尽调路径是取得 Owkin 按渠道划分的收入、已签约里程碑时间表、监管申报时间线,以及任何独立检测量或报销数据,再与这里的外部 TAM 视角对账。在那之前,市场最好被描述为庞大且增长中,但只与 Owkin 已证明的触达松散相连。[CM011, CM016, CM020, CM041, CM042, CM043]

2.6 图表

Chapter 03

03竞争对手

3.1 竞争格局:药物发现 AI、病理既有玩家与现状替代方案

截至 2026 年中,Owkin 位于三个相邻竞争场域的交叉点。第一是 AI 原生药物发现平台,Insilico Medicine、合并后的 Recursion-Exscientia 实体、Isomorphic Labs(DeepMind 拆分公司)、BenevolentAI 和 Schrödinger 都试图通过生成式 AI、phenomics、基于物理的模拟或知识图谱方法压缩制药 R&D 周期。第二是数字病理 AI,Owkin 过去的诊断能力(现归入拆分后的 Waiv)曾与 PathAI 和 Paige AI 竞争;前者正被 Roche 以 $750 million 首付款收购,后者于 2025 年 8 月被 Tempus 以 $81.2 million 收购。第三是临床数据基础设施与联邦分析,Owkin 的医院网络在这里与 Tempus 的集中式临床数据平台、Flatiron Health(Roche 子公司)以及 Foundation Medicine 的基因组伴随诊断网络竞争。 除这些具名对手外,最常见替代方案是药企内部自建:Pfizer、Roche、Novartis、AstraZeneca 和 Sanofi 每年都在内部 AI R&D 基础设施上投入数亿美元,也维持着与 Schrödinger 等计算化学平台供应商的既有关系。大型药企结构上很容易多栖:公司会并行评估供应商,且常常同时授权多个 AI 平台,从而限制任何单一供应商的定价权。2026 年要观察的结构性信号是整合——PathAI 进入 Roche、Paige 进入 Tempus,都符合既有玩家收购专业 AI 团队而非自建的模式,压缩了独立病理 AI 供应商的窗口期。AI 药物发现市场 2025 年估值约 $1.94 billion,预计 2026 年达到 $2.6–2.8 billion;全球目前有超过 173 个 AI 设计药物项目处于临床试验。[CP010, CP012, CP029, CP030, CP031, CP034]

FP001: 竞争定位图

Owkin 位于高联邦化、中等临床验证深度象限,是该领域主要的联邦学习平台;PathAI/Roche 和 Insilico 的临床与监管锚点更强,Isomorphic Labs 融资最高但仍处临床前。

X 轴(联邦 / 隐私架构)和 Y 轴(临床 / 监管验证深度)是截至 2026 年 6 月、基于公开证据得出的 0–100 顺序评分,不是数值基准。临床验证深度反映监管里程碑状态(FDA 批准=高,III 期=中高,II 期=中,临床前=低)。

[CP010, CP012, CP021, CP027, CP029, CP030]

3.2 竞争对手画像:规模、管线与战略方向

截至 2026 年中,Insilico Medicine 拥有该领域最重要的临床证明点。其 rentosertib 候选药物——完全由 AI 为特发性肺纤维化设计——在 2025 年 6 月发表于 Nature Medicine 的 Phase IIa 结果为阳性,十二周内用药组用力肺活量改善 98.4 mL,而安慰剂组下降 20.3 mL。公司 2025 年完成 $110 million Series E,2025 年 12 月在港交所上市,FY2025 报告收入 $56.2 million、现金 $393 million,并在 2026 年 3 月与 Eli Lilly 签署 $2.75 billion R&D 合作,其中首付款 $115 million。其 Pharma.AI 平台覆盖 Biology42(靶点发现)、Chemistry42(生成式化学)和 Medicine42(临床设计)模块。Insilico 在靶点识别上与 Owkin 争夺药企合作预算;Owkin 的多模态患者数据和联邦访问,与 Insilico 生成式 AI 优先、集中式数据的模型形成对比。 合并后的 Recursion-Exscientia 实体(RXRX)代表该领域最大的 AI 药物开发操作系统,把 Recursion 由 phenomics 驱动的高通量生物学与 Exscientia 的精准分子设计整合在一起。这笔 $688 million 全股票合并于 2025 年初完成。与 NVIDIA 共建的 BioHive-2 超级计算机支持跨 60+ PB 数据训练,数据覆盖 phenomics、转录组学、蛋白质组学和 ADME。合并后实体 2026 年报告现金 $754–850 million,跑道可延至 2028 年初,并持有来自 Roche/Genentech、Sanofi 和 Bayer 的超过 $20 billion 潜在里程碑付款。Recursion 主要在生物学和化学方向与 Owkin 争夺药企 R&D 预算,而不是在病理 AI 上竞争。 Isomorphic Labs 于 2026 年 5 月完成由 Thrive Capital 领投的 $2.1 billion Series B,Alphabet、GV、MGX、Temasek、CapitalG 和 UK Sovereign AI Fund 参投,使累计融资超过 $2.6 billion。其 IsoDDE 平台源自基于 AlphaFold 的 AI,推进肿瘤学、免疫学和心血管疾病药物设计。公司与 Novartis、Eli Lilly 和 Johnson & Johnson 有合作;首批临床候选药物预计 2026 年底前出现。按累计融资计,Isomorphic 相对 Owkin 的资金优势约为 7 倍,在算力和基础模型训练投资上形成结构性差距。 BenevolentAI(Amsterdam: BAI)自 2022 年上市以来反复重组。第一轮裁员 180 人后,又进一步削减 30% 员工、关闭美国办公室,并搁置 SaaS 知识工具。其股价较 2022 年 IPO 下跌超过 90%,市值降至约 £85 million。公司保留 Phase II 的 BEN-8744(一款用于溃疡性结肠炎的 PDE10 抑制剂)以及 $594 million Merck KGaA 合作。BenevolentAI 是本章主要负面证据锚点:当发现时间线超过资本市场耐心时,AI 平台公司会面临严峻烧钱挑战。[CP014, CP015, CP016, CP017, CP018, CP019]

竞争对手画像表
竞争对手类别规模 / 融资(USD)目标细分市场关键差异化主要限制
Insilico MedicineAI 药物发现平台(上市;香港交易所,2025 年 12 月)$393M 现金;估值 >$1B;Lilly 交易 $2.75B(2026 年 3 月)大药企研发;CRO;肿瘤 / 纤维化首个 AI 设计、Phase IIa 读出阳性的药物(rentosertib);Pharma.AI 模块化平台(靶点→分子→临床)中心化数据架构;无联邦访问;无病理 AI 能力
Recursion + ExscientiaAI 药物开发 OS(上市;RXRX)$754–850M 现金;管线里程碑 >$20B(Roche、Sanofi、Bayer)罕见病;肿瘤;大药企研发端到端 OS(表型组学 + 精准化学 + BioHive-2 超算);预计 10 个临床读出并购后整合风险;三个项目搁置;股价远低于 2021 年高点
Isomorphic LabsAI 药物设计 / AlphaFold 衍生平台(私有;Alphabet 支持)累计融资 $2.6B+,含 $2.1B Series B(2026 年 5 月)大药企(Lilly、Novartis、J&J);肿瘤 / 免疫学AlphaFold 衍生 IsoDDE;可用 Alphabet 算力;17 个内部项目;史上最大生物科技 Series B,$2.1B截至 2026 年中仍处临床前;未披露收入;预计 2026 年底首批进入临床
BenevolentAIAI 药物发现(上市;Amsterdam BAI)约 £85M 市值;与 Merck KGaA 的 $594M 合作(结构上有风险)以合作伙伴为中心的早期发现用于靶点发现的知识图谱;BEN-8744 Phase II(UC);已建立科学 IP2022 年 IPO 后股价跌幅超过 90%;裁员 180+ 后再砍 30%;美国办公室关闭;SaaS 搁置
Schrödinger基于物理的计算化学(上市;SDGR)广泛药企合作;累计 $500M+药物设计(肿瘤;炎症性疾病;材料)物理优先的 FEP+ 模拟;zasocitinib(Nimbus/Takeda)处于 Phase III;平台历史最久不是联邦或病理 AI 平台;只在分子设计上竞争;无专有临床管线
PathAI → Roche数字病理 AI(收购待 2026 年下半年完成)首付款 $750M + 最高 $300M 里程碑(Roche;2026 年 5 月宣布)药企临床试验;医院病理实验室;伴随诊断AISight IMS,带 AI 分析和工作流;伴随诊断联合开发;BMS 与 Merck 为锚定伙伴收购后路线图将由 Roche 主导;独立产品战略取消
Paige AI → Tempus数字病理 AI(2025 年 8 月被收购)收购价 $81.2M;累计融资 $241M医院病理;肿瘤诊断;数字工作流FDA 批准的癌症 AI 应用(前列腺、乳腺、GI);多癌种基础模型;已整合进 Tempus 平台并入更大的 Tempus 实体;独立路线图退出;产品走向不确定
Proscia数字病理工作流 + AI(私有)私有;融资未披露药企药物发现;医院病理实验室面向组织病理学、覆盖发现到病例复核的 Concentriq Gen 5 AI 原生架构未披露大型药企锚定交易;公开临床验证证据有限;规模小于 PathAI

融资和估值数据反映截至 2026 年 6 月的最新公开披露。Roche-PathAI 交易已宣布,但仍待 2026 年下半年监管批准完成。BenevolentAI 市值基于阿姆斯特丹上市价格,不是私募估值。Schrödinger 和 Recursion 为上市公司;私有公司数据(Isomorphic、Proscia)来自已披露轮次或分析师估算。

[CP009, CP010, CP012, CP013, CP014, CP018]

3.3 功能与能力对比:Owkin 领先在哪里,落后在哪里

Owkin 最强的差异点是联邦学习架构。原始患者数据从不离开成员医院系统;模型在本地训练,只有梯度更新被集中聚合。这让 Owkin 能访问 GDPR 下欧洲医院和学术医疗中心的真实世界临床数据,而 Insilico、Recursion、Isomorphic Labs 等集中式竞争对手很难这样运作。MOSAIC 空间多组学数据集进一步强化这一点:它覆盖多种癌症类型,整合六类数据模态——空间转录组学、单细胞 RNA-seq、bulk RNA-seq、全外显子组测序、临床数据和 H&E 病理图像——按公司披露,是全球最大的空间多组学肿瘤数据集,规模比任何可比替代方案大 100 倍。没有竞争对手公开披露过同等规模、可比的联邦、多机构空间多组学计划。 Owkin 落后的地方是纯化学设计。K Pro 目前不提供可与 Insilico 的 Chemistry42、Recursion-Exscientia 的整合精准化学或 Schrödinger 基于物理的 FEP+ 套件相比的从头分子生成能力。Owkin 的主张在上游——靶点识别、患者分层、agentic 研究辅助——而不是下游分子设计和 ADME 预测。这使其在分子设计上与 Recursion 和 Insilico 更互补,在靶点发现和临床试验优化上重叠更直接。PathAI(被 Roche 收购前)和 Proscia 在临床病理工作流自动化上竞争,而 Owkin 已通过 Waiv 拆分基本退出该类别。Foundation Medicine 和 Flatiron Health(均属 Roche)是基因组诊断和临床数据既有玩家,拥有深厚医院关系和 FDA 批准产品锚点;Owkin 的 K Pro 在监管就绪度上尚未达到同等水平。[CP002, CP006, CP007, CP011, CP028, CP034]

功能 / 能力矩阵
能力Owkin K ProRecursion + ExscientiaInsilico Pharma.AIPathAI → RocheIsomorphic IsoDDE
联邦 / 隐私保护学习是 — 核心架构;数据留在医院否 — 中心化专有数据集(60+ PB)否 — 中心化否 — 中心化否 — 中心化
药物靶点识别(AI 辅助)是 — K Pro / Owkin Zero 生物 LLM;声称提速 70%是 — 表型组学驱动的靶点发现是 — Biology42 模块部分 — 聚焦生物标志物;不是主要靶点识别是 — 基于 AlphaFold 结构的靶点发现
从头分子生成 / 设计否 — 不是 K Pro 功能是 — 已整合 Exscientia 精准化学是 — Chemistry42 生成式化学否 — 不提供是 — IsoDDE 药物设计引擎
数字病理 / 组织学 AI部分 — 已拆分至 Waiv 子公司;K Pro 不包含是 — AISight IMS 核心产品;FDA 伴随诊断算法
临床试验优化 / 患者分层是 — 已证明试验周期缩短 35%;患者分群识别是 — 通过 Recursion OS 支持临床试验入组是 — Medicine42 临床模块部分 — 为试验入组提供伴随诊断支持未知 — 截至 2026 年中聚焦临床前
空间 / 多模态组学数据访问是 — MOSAIC(7,000+ 名患者;6 种模态;公司称全球最大)部分 — 包含转录组学和表型组学;未披露空间组学部分 — 面向组学数据的生成模型;空间能力有限否 — 未披露Unknown
智能体 / 自主研究工作流是 — K Pro 作为 AI Scientist;自然语言查询;Owkin Zero LLM部分 — 自动化实验室机器人;不是对话式智能体推理部分 — Pharma.AI 模块;未明确具备智能体能力Unknown
FDA 批准或监管批准诊断部分 — BRCAura 为 RUO 状态(未批准);诊断业务在 Waiv是 — 多个 FDA 伴随诊断算法;FDA 批准 AISight
已披露战略药企合作(2026)AstraZeneca(3 年 K Pro);Sanofi(5 年 K Pro);BMS(投资方 / 研究)合作方:Roche/Genentech、Sanofi、Bayer、NVIDIAEli Lilly($2.75B);前 20 大药企中 13 家(软件授权)Roche(收购);BMS(2026 年 6 月扩大合作);MerckLilly;Novartis;Johnson & Johnson

基于截至 2026 年 6 月的官方产品页面、新闻稿和已发表论文。标为未知的单元格缺乏足够公开证据;标为部分的单元格表示有暗示能力,但尚无完整独立验证。Owkin 的数字病理能力已转入 Waiv 分拆实体,Waiv 是独立实体,不在本表覆盖范围内。

[CP002, CP003, CP011, CP017, CP034, CP037]
FP002: 功能广度 / 能力图

Owkin 在联邦架构、空间多组学和智能体 AI 工作流上领先,但在从头分子设计和 FDA 批准诊断上落后;PathAI/Roche 领先于受监管病理 AI;Insilico 和 Recursion 领先于小分子化学。

基于截至 2026 年 6 月的官方产品页面和新闻稿。Unicode 对勾(已确认存在)、叉号(未提供)、部分(公开证据显示但不完整)、未知(无公开披露)。Owkin 的数字病理能力位于 Waiv 分拆公司。

[CP006, CP029, CP034, CP045, CP046]

3.4 定价、GTM 与分销力量

Owkin K Pro 通过多年期企业协议授权。AstraZeneca 交易(三年 K Pro 许可加嵌入式 agent 开发)和 Sanofi 交易(五年 K Pro 许可加共同开发定制化生物制药 agents)确立了经常性多年合同结构,但确切财务条款仍未披露。根据官方公告描述的嵌入式 agent 交付范围、决策智能访问和 Owkin Zero 生物 LLM,这些合同很可能是每年中高个位数百万美元规模的安排。公开可得的可比临床 AI 平台许可分析师估算显示,单站点部署每年 $150,000–$250,000;带定制 agent 开发的大型药企企业协议明显更高。 Insilico Medicine 已将 Pharma.AI 授权给全球前二十大药企中的十三家,显示广泛分销渗透。Recursion 和 Exscientia 主要通过基于里程碑的合作分销,整个行业首付款平均约为标题交易价值的 2%——这让它们的收入比 Owkin 经常性许可模式更依赖后续事项、更受里程碑约束。PathAI(收购前)通过直销进入医院实验室和药企临床试验运营,BMS 和 Merck 是锚定合作伙伴。Schrödinger 以订阅方式授权 FEP+,披露的大型药企合同年额超过 $50 million。 Owkin 的 GTM 优势来自学术医院网络,它创造了双边飞轮:医院贡献联邦参与,换取研究洞察和数据访问;这加深 Owkin 的数据壁垒,也增强面向药企的商业提案。关键风险是收入集中——AstraZeneca 和 Sanofi 合计很可能是 Owkin 最大的两个 K Pro 客户之一,若两三家锚定药企关系过度集中,一旦任何一方不续约,公司会暴露脆弱性。[CP001, CP003, CP004, CP005, CP009, CP038]

定价 / 包装对比
供应商合同模式指示性年价值已知期限主要包含内容 / 备注置信度
Owkin K Pro多年企业授权 + 内嵌 AI 智能体开发中高七位数(估算);未披露3–5 年(AZ 3 年;Sanofi 披露 5 年)K Pro 平台访问;定制智能体开发;Owkin Zero LLM;MOSAIC 数据访问低 — 未公开披露定价;按交易范围估算
Insilico Pharma.AI软件授权 + 基于里程碑的合作条款前付款未知;按里程碑释放;Lilly 交易披露 $115M 前付款多年(Lilly 交易于 2026 年 3 月签署)Biology42 / Chemistry42 / Medicine42 模块访问;合作资产联合开发权低 — 精确授权价格未披露;仅有一个前付款参考
Recursion OS 合作基于里程碑的研发合作;药企出资获取生物学和化学能力标称交易价值约 2% 作为前付款(行业平均);Roche 披露首笔付款 $30M多年(Roche、Bayer、Sanofi 交易仍有效)表型组学和化学平台访问;共同生成数据;临床项目联合开发中 — 2% 前付款平均值已有公开;单笔交易披露不完整
PathAI AISight(Roche 前)SaaS IMS 订阅 + 药企服务合同未知;Roche 以 $750M 首付款收购,交割后模式改变已被收购(预计 2026 年下半年交割)图像管理系统;AI 分析;伴随诊断算法联合开发低 — 收购后商业模式透明度消失;历史定价未披露
Schrödinger FEP+(平台)年度订阅 + 项目专项合作费主要药企客户年费 $50M+(分析师估算)年度续约 + 项目专项基于物理的 FEP+ 分子模拟;药物设计工作流;合作伙伴联合开发中 — 区间来自股权披露和已发布分析师覆盖推断

所有 AI 药物发现平台的定价在结构上都不透明;多数披露值是包含里程碑的总潜在交易价值,不是已实现年收入。约 2% 前付款平均值来自分析师对 2025 年交易流的行业估算,不代表任何单家公司条款。标为低置信度的估算不应用于财务建模。

[CP004, CP005, CP009, CP010, CP020, CP044]

3.5 壁垒耐久性、转换成本与竞争风险

Owkin 最耐久的竞争资产,是联邦学习基础设施与 MOSAIC 数据集的组合。复制这套资产,竞争对手需要跨多个司法辖区与数百家医院签约,在每家机构内部搭建合规的联邦计算基础设施,大规模整理空间组学数据,并等待多年,让所得数据集达到可与 MOSAIC 相比的临床多样性。Insilico、Recursion 和 Isomorphic Labs 都没有公开披露过可比的联邦医院网络计划;三者都使用集中式专有数据集。Foundation Medicine 和 Flatiron Health 拥有大型医院网络,但分别聚焦基因组测序和电子健康记录,而不是空间多组学与联邦 AI 训练的结合。 转换成本不低。把 K Pro 嵌入药物开发工作流的药企伙伴,会面临数据重新集成成本、临床试验场景中 AI 辅助决策的监管再验证,以及科学团队和高管团队再培训。这带来估计 6–18 个月迁移摩擦和中等锁定。多栖——药企并行运行多个 AI 平台——比直接替换更威胁定价权,因为大型药企通常会同时在 3–5 个 AI 平台关系中分散供应商风险。 最大的结构性风险在行业层面:截至 2026 年 6 月,没有 AI 设计药物获得监管批准。BenevolentAI 的经历——股价暴跌 90% 以上、反复裁员、SaaS 业务搁置、美国办公室关闭——说明在没有监管里程碑的漫长时间线上,资本市场耐心有限。AI 支持候选药物的 Phase II 成功率平均约 40%,与传统成功率相当,削弱了临床溢价叙事。Recursion 合并后搁置了三个项目。如果 AI 药物发现进入持续可信度缺口,药企合作预算可能转向更小的首付款承诺,挤压 Owkin 的经常性许可收入模式。[CP006, CP007, CP024, CP025, CP029, CP032]

护城河耐久性 / 竞争风险登记表
护城河主张主要威胁严重程度缓释措施 / 尽调问题
联邦学习在主要 AI 药物发现竞争者中独一份Isomorphic Labs 或 Recursion 可借助 Alphabet/NVIDIA 资源,通过学术合作建立联邦医院访问核实 AstraZeneca 和 Sanofi 合同是否包含数据访问排他性;评估 Owkin 联邦网络是否有 IP 保护或合同壁垒,阻止竞争对手复制
MOSAIC 是全球最大的空间多组学肿瘤数据集(公司主张)Roche 收购 PathAI 后,可借助其全球医院网络和 Foundation Medicine 关系,出资搭建竞争性空间组学联盟确认 MOSAIC 医院贡献方是否有排他条款;评估数据刷新节奏能否顶住 Roche 资源;要求独立验证规模
K Pro Owkin Zero LLM 在生物学推理上优于通用 LLMRecursion 和 Isomorphic Labs 正在构建领域专用生物基础模型,算力预算大 5–7×;通用 LLM 也在快速进步要求提供 Owkin Zero(arXiv 2508.16315)除自报指标外的独立基准结果;评估 12 个月内预期的复现或挑战研究
多年药企合同带来收入稳定性和切换成本药企预算削减、管线失败或并购可能在续约时触发提前终止或重新谈判;多平台并用会降低切换紧迫性获取客户集中度数据;识别 K Pro 收入中来自前两大客户(AstraZeneca、Sanofi)的占比;评估续约条款和通知期
医院网络带来专有双边飞轮,并随使用复利增长大型医院体系自建 AI 数据湖(Epic、NHS、Intermountain)后,加入第三方联邦网络的动力可能下降;政府数据计划可复制覆盖面评估医院网络流失率和合同续约条款;识别医院参与联邦网络可获得什么回报
尚无 AI 设计药物获监管批准;全行业可信度风险可能压缩 AI 药企预算行业内 Phase II 持续失败或遭遇监管挫折,可能降低药企为发现 AI 平台付费的意愿;BenevolentAI 崩盘就是先例跟踪 rentosertib Phase IIb 这一行业风向标;跟踪 AstraZeneca 和 Sanofi 公开披露中把发现归因于 K Pro 的情况;评估 Owkin 合同结构是按里程碑闸门付款,还是纯授权

严重程度按截至 2026 年 6 月的公开竞争证据评估。高严重程度表示短期内会实质威胁 Owkin 竞争位置,且没有清晰披露的缓释因素。中等表示风险可通过主动尽调管理,或周期较长。标为公司陈述的护城河主张尚未获得独立验证。

[CP006, CP007, CP024, CP025, CP033, CP034]
FP003: 护城河 / 竞争就绪 KPI

Owkin 的联邦数据网络和 K Pro 生产力主张,是最主要的可量化护城河信号;相对 Isomorphic Labs 的 7 倍融资差距,以及行业零获批的背景,是两项头部风险指标。

K Pro 加速和 MOSAIC 规模比较来自公司披露,未经独立验证。融资差距计算使用累计披露总额;Owkin 2026 年 3 月的 $33M 延展融资已计入约 $354M 估算。II 期成功率是分析师报道中的行业均值,不是 Owkin 特定数据。

[CP003, CP006, CP009, CP010, CP016, CP021]

3.6 图表

Chapter 04

04财务

4.1 收入架构与商业模式

Owkin 采用混合型药企合作模式,建立在三个相互咬合的收入杠杆上:(1) 签约时收到的预付股权或合作费用,(2) K Pro agentic AI 平台的多年期经常性 SaaS 许可,以及 (3) 与特定药物发现研究成果挂钩、由里程碑触发的付款。基础性 Sanofi 交易(2021 年 11 月)把 $180 million 股权投资与同期 $90 million、三年期肿瘤合作付款安排结合在一起——Pharmaphorum 通过独立报道确认了这一结构。Bristol-Myers Squibb 随后在 2022 年 6 月投入 $80 million Series B 股权 tranche,并附带最高 $100 million 潜在里程碑付款,用于心血管试验 AI 应用。2026 年,Sanofi 关系通过一份五年期 K Pro 企业许可进一步加深;AstraZeneca 也在 2026 年 5 月签署独立的三年期 K Pro 许可协议,显示头部十强药企对平台的采用扩大。Owkin 的医院网络——现已覆盖 180+ 家机构——是联邦数据获取资产,而不是直接收入来源;价值在于提供保护隐私的多组学数据,支撑 AI 平台的临床可信度。2026 年 3 月,Owkin 将诊断部门(原 Owkin Dx)拆分为 Waiv,这是一家单独资本化的实体并融资 $33 million;Owkin 保留 Waiv 少数股权,但诊断收入流不再并入 Owkin 的 P&L。[CI001, CI002, CI003, CI004, CI005, CI006]

Owkin 收入流
收入流机制单位 / 定价当前状态收入质量尽调问题
药企合作授权交易签署时的前置股权投资 + 同步合作费用定制企业交易;已披露最大金额:$180M Sanofi(2021)、$80M BMS(2022)活跃 — Sanofi 和 BMS 为锚定伙伴价值高但非经常性;前置确认确认总收入中前付款与经常性组成部分的比例
里程碑触发的研发付款基于事件的付款,由可验证研究成果触发(生物标志物验证、试验设计获接受)单笔交易时间表未披露;BMS 披露单笔交易最高潜在 $100M活跃 — 多个合作中里程碑付款持续发生波动大、非经常性;长药物发现周期带来时间风险要求完整里程碑时间表、触发条件和历史达成率
K Pro 智能体 AI 平台(SaaS 授权)多年企业 SaaS 订阅,将 AI 药物发现智能体嵌入药企研发工作流未披露;Sanofi:5 年授权;AstraZeneca:3 年授权活跃且增长中 — 2026 年 5 月新增第三家大型药企客户(AstraZeneca)最高收入质量 — 经常性、多年;规模未披露按合作伙伴和合同年份披露 K Pro 平台授权 ARR
诊断(Waiv / Owkin Dx)原先并表的 Owkin Dx 诊断收入;2026 年 3 月拆分为 WaivN/A — Owkin 仅保留少数股权;不再并表已剥离 — 保留 Waiv 少数股权收入流从 Owkin P&L 移除;降低多元化确认分拆对 Owkin P&L 的净财务影响,以及是否仍有持续特许权使用费或授权安排

K Pro 授权和里程碑付款的收入数据未公开披露;Sanofi($180M 股权、$90M 合作)和 BMS($80M 股权、最高 $100M 里程碑)的前付款金额已由新闻稿确认。Waiv 分拆状态由 2026 年 3 月 BusinessWire 新闻稿确认。

[CI001, CI003, CI004, CI005, CI006, CI025]
FI001: 收入模型桥接——从客户活动到 Owkin 收入

展示药企企业客户如何通过多年合作合同结构,为 Owkin 贡献三类不同收入流。

收入总额(约 $125M)是 Growjo 和 Compworth 的分析师估算;Owkin 未确认。毛利润节点反映未披露的信息缺口。流程呈现机制,不代表各收入流的比例拆分。

[CI002, CI008, CI009]

4.2 定价与合同结构

Owkin 不公布任何产品价目表。K Pro 许可采用定制化多年期企业协议,与每家药企客户的 C-suite 和 VP-R&D 对手方直接谈判。Sanofi 的 K Pro 交易为五年;AstraZeneca 为三年。合同经济性混合了预付合作费用、年度经常性许可付款,以及由生物标志物验证或临床试验设计接受等可验证研究里程碑触发的里程碑 tranche。收入确认遵循 IFRS 15 / ASC 606 原则:归属于不同履约义务的预付款在交付时确认,经常性许可收入在合同期内按比例确认,里程碑付款在底层成果被验证且收款权确立时确认。确切交易经济性——年度许可费、里程碑时间表、里程碑付款金额——在所有已披露合作中均为保密。$90 million Sanofi 肿瘤合作(2021–2024 年)和 $80 million BMS 首付款加最高 $100 million 潜在 BMS 里程碑,是唯一公开量化的合同条款;2026 年 K Pro 许可没有披露财务数字。[CI007, CI008, CI009, CI010, CI012, CI013]

定价与商业化摘要
收入类型价格 / 单位 / 合同结构标价 vs. 实现价格折扣 / 未知项来源
药企合作(Sanofi 2021)$180M 股权投资 + $90M 合作付款,分 3 年支付按交易条款计股权;合作费用在披露期间经常性确认里程碑条款和续约经济性未披露Owkin 官方新闻稿;BusinessWire 文件
药企合作(BMS 2022)$80M 前期款(股权 / 费用)+ 最高 $100M 潜在里程碑付款前期款已确认;里程碑取决于 R&D 进展已触发的实际里程碑及收到金额未披露Pharmaphorum;BusinessWire Series B 文件
K Pro SaaS 许可(Sanofi,2026)K Pro agentic AI 平台 5 年企业许可未公开披露全部财务条款保密;未披露年费或 TCVOwkin 官方;RD World Online;AllSci
K Pro SaaS 许可(AstraZeneca,2026)K Pro agentic AI 平台 3 年企业许可未公开披露财务条款保密;交易于 2026 年 5 月宣布FierceBiotech;mHealthSpot
里程碑付款(组合)覆盖所有药企合作,按事件触发;逐单约定总额未披露;BMS 单笔交易披露最高 $100M迄今已触发的累计里程碑未获确认;时间不确定Pharmaphorum;分析师估算

K Pro SaaS 许可全部采用定制谈判,价格保密。Sanofi 合作的财务条款是公开记录里最完整的一组;BMS 条款只得到部分确认。Owkin 任何产品都没有自助定价层级,也没有公开价目表。

[CI007, CI008, CI009, CI010, CI013]

4.3 单位经济模型与成本结构

Owkin 没有披露财务报表,也没有披露单位经济指标。Growjo 和 Compworth 基于员工规模和交易规模估算其年收入约 $125 million——这是分析师估算,Owkin 未确认也未否认。按约 470 名员工(分析师劳动力情报,2025 年 12 月)计算,隐含人均年收入约 $217K,符合一家处于验证到规模化拐点、尚未盈利的 AI 服务公司。毛利率完全未披露;联邦学习架构将计算分散到医院节点,而不是集中处理,相比集中式 AI 竞争对手,可能在结构上抬高基础设施成本。成本项包括 AI 研究和模型开发、分布式云计算、180+ 站点的医院网络集成与维护、多司法辖区患者数据框架的监管与法律合规,以及面向大型药企的企业销售和业务发展。CAC、销售周期长度、净收入留存率和回本期均未确认。Owkin 的药物发现合作模式内生收入时点风险:药企 R&D 周期从靶点识别到监管批准通常需要五到十年,意味着里程碑付款天然不平滑,难以单独建模。[CI014, CI015, CI016, CI017, CI018, CI019]

单位经济模型——已知、估算与缺失
指标数值 / 估算置信度重要性尽调要求
年收入(估算)~$125M(Growjo / Compworth 分析师估算,2025)低——仅为分析师估算;Owkin 未确认确认商业化规模与增长轨迹索取 FY2024 和 FY2025 经审计 P&L 或管理账
人均收入~$217K/年(推导:~$125M 收入 ÷ ~470 名员工)低——由未经确认的输入推导衡量规模化效率的生产率代理指标确认实际员工数和收入,以验证该比率
毛利率未披露None决定盈利路径,以及相对于成本基础的定价权DD 中要求按收入流拆分的毛利率明细
CAC / 回本周期未披露None衡量多年期药企企业交易的 GTM 效率索取销售运营数据、平均交易周期和单 logo 获取成本指标
净收入留存(NRR)未披露None衡量经常性 SaaS 许可中的账户扩张与流失要求 K Pro 许可续约、扩容的 cohort 分析
月度烧钱率未披露;方向性估算 ~$3–9M/月(按员工数代理)无(估算)判断现金 runway 与下一次融资的紧迫性要求过去 12 个月现金流量表和当前现金余额

除收入估算外,其他单位经济指标都没有进入公开记录。收入估算来自分析师使用的 workforce-intelligence 平台,Owkin 未确认。烧钱率估算按员工数(~470 人)和 biotech-AI 行业薪酬基准推导;实际烧钱率可能因 R&D 支出、基础设施成本和合作收入抵扣而大幅不同。

[CI014, CI017, CI018, CI019, CI020]
FI002: 单位经济桥接——成本结构与毛利润缺口

定性拆解 Owkin 的服务成本构成,说明为何无法用公开数据计算毛利率;所有成本节点均未量化。

所有成本节点均为定性且未量化。收入节点采用 $125M 分析师估算。由于分布式节点计算,联邦学习基础设施成本预计在结构上高于集中式 AI 竞争者。Owkin 未披露任何毛利率数据。

[CI015, CI016, CI011]
FI003: 关键财务估算区间——Owkin 2026

对公开数据或分析师估算覆盖的关键财务变量,给出有来源支撑的低 / 高边界;所有区间都只是方向性,若无管理层确认,不应直接用于承销。

所有区间均由公开可得的分析师数据和行业基准推导。Owkin 未确认其中任何数字。区间只能视作粗略方向性边界;实际值可能大幅落在区间之外。

[CI017, CI023, CI028]

4.4 资本充足性与融资

Owkin 自 2016 年创立以来累计融资约 $321 million。融资历史细节——Seed($2.1M,2016 年)、多轮 Series A(2018–2020 年,合计约 $59M)、$180M Series B(Sanofi,2021 年 11 月)以及 $80M Series B 延展轮(BMS,2022 年 6 月)——在公司概览章节已有记录。截至 2026 年中,关键资本充足性问题是跑道:Owkin 母公司层面最近一次一级市场股权融资在 2022 年 6 月完成,间隔约四年。Premier Alts 估计 2026 年隐含二级市场估值约 ~$1.6 billion,高于 2021 年 Series B 的 $1 billion 投后估值,说明二级市场乐观,但没有确认的新股权融资。Waiv 诊断拆分($33M,2026 年 3 月)把现金加到 Waiv 资产负债表,而不是 Owkin。Owkin 没有披露现金头寸、月烧钱率或跑道数字。可以从员工数方向性近似烧钱:按 ~470 名员工和 biotech-AI 混合薪酬基准,月现金成本(工资、福利、基础设施、管理费用)可能为每月 $3–9 million。合作收入是否完全抵消这一成本,或 Owkin 是否在消耗 2022 年融资现金,均未披露。没有公开债务融资或项目融资义务公告。[CI021, CI022, CI023, CI024, CI028, CI029]

资本充足性评估
项目数值 / 估算来源 / 置信度注释
已募集总资本(Owkin 母公司)~$321M(从种子轮到 Series B,含 grants)中——经 Tracxn、Clay、Signalbase 汇总已披露融资确认不含 Waiv $33M spinout 融资(属于 Waiv 股权结构,不属于 Owkin)
账面现金未披露无——私营公司,无公开申报Owkin 未发布资产负债表;现金头寸未知
月度烧钱率未披露;代理估算 $3–9M/月无——仅为按员工数推导的方向性代理代理假设约 470 名员工 × biotech-AI 混合成本;实际烧钱率未知
2022 年 6 月融资后的估算 runway2022 年 6 月交割后 12–48 个月(取决于烧钱率与合作收入抵扣)低——仅为推导估算假设 $80M BMS 融资是最后一次 primary liquidity event;合作收入抵扣烧钱
计划资金用途2026 年未正式披露None上一次披露用途:AI R&D、医院网络扩张、agentic AI 商业化
债务 / 项目融资义务未披露;未宣布公开债务融资工具None小额 EU / 政府 grants(累计约 $800K–$1M)为非稀释资金,但金额不重大

对 Owkin 来说,资本充足性是最重要、但未披露的财务维度。Runway 估算只是方向性代理;如果合作收入显著抵扣烧钱,实际情况可能差异很大。所有标为「未披露」的数字,均确认在已审阅公开来源中缺失。完整逐轮融资时间线见 Company Overview 章节;本章本地 claims 使用 Financials 章节 sourceRefs 独立生成。

[CI021, CI022, CI023, CI028]
FI004: 资本强度与现金流图

展示 Owkin 来自股权轮次和合作收入的资本流入来源,以及构成现金消耗基础的主要运营成本桶。

资本流入反映公开来源确认的股权轮次。成本流出节点为定性描述;Owkin 未披露支出分配。最后一次股权流入(2022 年 6 月)到报告运行日期(2026 年 6 月)之间相隔四年,是关键资本充足性信号。

[CI021, CI024, CI029, CI015]

4.5 财务数据缺口与尽调要求

作为私营公司,Owkin 无需提交公开财务报表,也没有自愿披露。由此产生的缺口是结构性且广泛的:毛利率、营业利润率、EBITDA、烧钱率、现金余额、按合同划分的 ARR、NRR、CAC 和回本期都不在公开记录中。客户收入集中度——Sanofi、BMS、AstraZeneca 与其他客户分别贡献总收入的比例——未被量化。Sifted 2025 年报道称,Owkin 员工在公司转向 agentic AI 期间公开批评内部管理,提到裁员、组织透明度不足和战略变化沟通不佳——这是关于转型执行质量的负面信号,并直接影响财务。历史合作价值绝大部分依赖两家锚定伙伴(Sanofi 和 BMS),形成结构性客户集中风险;若任一伙伴缩小范围或把 AI 能力收归内部,影响可能重大。没有管理账和独立验证财务报表,就无法承销回报资本模型。[CI030, CI031, CI032, CI034, CI038]

公开财务数据缺口
缺失指标对投资分析的影响精确尽调路径
毛利率 %无法评估成本结构、毛利可持续性或 SaaS benchmark索取 FY2024–FY2025 按收入流拆分的经审计利润表
月度烧钱率和现金余额无法计算 runway、下一轮融资紧迫性或稀释风险索取过去 12 个月现金流量表和最近月末资产负债表
按合作方和合同年度拆分的 ARR无法区分经常性与非经常性收入质量,也无法评估模型续约风险索取 ARR 明细,包含合同起止日期、续约选项和自动续约状态
CAC 和企业销售周期长度无法评估 GTM 效率、销售杠杆或增长投入的生产率索取销售运营数据:pipeline 指标、平均交易规模、成交率和单 logo 成本
客户收入集中度(Sanofi vs. BMS vs. 其他的占比)无法量化单一合作方依赖,或合作方流失对模型收入的影响索取 FY2023–FY2025 按客户拆分收入;确认是否存在最低购买保障
净收入留存和 cohort 数据无法建模账户扩张轨迹,也无法将 K Pro 与 SaaS 可比公司 benchmark索取 K Pro 迄今续约、扩容和流失合同的 cohort 分析

截至 2026-06-19,上述缺口均确认在所有已审阅公开来源中缺失。对 Owkin 这一阶段的私营公司来说,这些缺口是结构性的,只能通过签署 NDA 后进入投资者 data room 解决。影响评级反映投资承销的重要性,不代表普通商业尽调评估。

[CI030, CI031, CI038]

4.6 财务结论

Owkin 的财务画像是:合作收入潜力高,资本强度高,单位经济模型结构性不透明。Sanofi 五年 K Pro 许可和新的 AstraZeneca 交易提供多年收入可见性,这是耐久的正面信号。但距上一轮股权融资已有四年,披露交易价值集中在两家伙伴,烧钱和利润率未披露,Sifted 报道中的管理文化信号,以及 Waiv 拆分导致诊断收入流流失,合在一起形成重大信息赤字。分析师估算收入 $125 million 若正确,对这一阶段公司而言不错;但没有毛利率数据,投资者无法建模单位层面盈利能力,也无法判断下一平台周期所需 R&D 资本。收入质量混合:里程碑付款天然不平滑且非经常;来自 K Pro 的经常性 SaaS 许可是前瞻收入质量最好的组成部分,但规模完全未披露。核心尽调要求是获取过去三年经审计财务报表、当前资产负债表,以及一份带有合作收入短缺敏感性的 24 个月现金流预测。[CI035, CI036, CI037, CI040]

4.7 图表

Chapter 05

05产品与技术

5.1 产品组合与客户工作流

Owkin 的商业界面围绕「Owkin K」展开。这是面向生物制药的可互操作智能体式基础设施,旗舰产品 K Pro 则被 Owkin 描述为「由多模态患者数据驱动、帮助生物制药做出更聪明决策的生物学 AI 智能体」。K Pro 是一个用自然语言查询的企业平台:单一智能编排器会动态选择并组合专用 AI 技能,回答从靶点发现到组合策略的药物开发管线问题,并返回基于多模态患者数据、可用于发表的答案。面向科学家的免费层 K Pro FREE(此前品牌为 K Navigator)帮助生物学家写作、查文献、验证假设和可视化数据,并确保结果能追溯到 PubMed 等可信来源。底层资产包括数据和基础设施:MOSAIC 空间与多组学数据集(11 种肿瘤类型、2,700 个癌症样本)、K Pro 页面称为「全球最大」的患者数据网络(104 家医院),以及 Owkin 的联邦学习工具。曾经支撑受监管产品线的 CE 标志诊断产品 MSIntuit CRC 和 RlapsRisk BC 已在 2026 年 3 月剥离至 Waiv,因此当前 Owkin 组合集中在 K 平台、OwkinZero 模型和底层数据网络,而不是已交付的体外诊断设备。客户工作流从药企 R&D 问题开始,进入自然语言 K Pro 查询,再由系统在患者数据上编排生物学推理,最后给出带证据链接、排好优先级的答案,可能改变项目轨迹。[CE001, CE002, CE003, CE004, CE005, CE006]

产品模块与资产矩阵
模块 / 资产主要用户状态 / 成熟度差异化尽调缺口
K Pro(agentic AI co-pilot,智能体 AI 副驾驶)药企 / biotech R&D 科学家和高管2025 年 10 月发布;已商业授权(AZ、Sanofi)一个 orchestrator 调度专门 AI skills,并以多模态患者数据 grounding准确性 / 性能主张没有独立审计;无医疗器械许可
K Pro FREE(原 K Navigator)个人生物学家 / 生物医学研究者免费层已上线;定位科学助手科学家优先;答案可追溯到 PubMed 和生物医学数据库使用规模、留存和向企业版转化未披露
OwkinZero(生物推理模型)支撑 Owkin K / K Pro 的内部模型层已发表(arXiv,2025 年 8 月);正借助 NVIDIA 扩展领域专用 biological LLM,用可验证奖励做 RL 训练生产模型大小、延迟和 guardrails 未公开说明
MOSAIC 数据集 + 医院网络用于模型训练和验证的内部数据层已投产;2,700 个样本 / 11 种肿瘤;104 家医院网络通过 federated learning 拼出的专有多模态肿瘤数据数据权利持续性和单院 consent 条款未公开
Substra / FLamby(FL 基础设施)Owkin 工程师 + 开源 FL 社区成熟;Substra 由 Linux Foundation 托管;FLamby benchmark开源、可审计的 federated-learning 栈,已在医院部署商业上依赖社区维护的 OSS;活跃度一般
MSIntuit CRC / RlapsRisk BC(诊断)病理医生 / 肿瘤临床医生CE 认证;2026 年 3 月 spin out 至 Waiv面向结直肠癌和乳腺癌、已获监管 clearance 的 AI IVD不再属于 Owkin 组合;仅保留 Waiv 少数股权

状态综合 Owkin 官方产品页、OwkinZero 论文和 2026 年合作报道。诊断业务行反映 2026 年 3 月 Waiv spin-out。网络规模(104 家医院)来自截至 2026 年 6 月 K Pro 页面披露的数字,与前文引用的员工数不同;应视为公司披露、未经审计的数字。

[CE001, CE002, CE005, CE007, CE013, CE020]
工作流与用例地图
用户任务传统工作流Owkin 方案声称收益限制
识别新的药物靶点在孤立数据集之间手工做多组学分析,耗时数月用自然语言向 K Pro 提问,编排多模态推理靶点识别从 12+ 个月缩短到约 3 个月(公司报告)收益来自单个内部案例,未独立验证
为 IND 申报定位资产传统竞品和生物学分析需数月K Pro 生成 IND-ready 资产定位策略用数小时而非数月完成(公司报告)未披露方法论或外部 benchmark
从病理预测治疗反应在条件允许时使用基于测序的 biomarker assaysENLIGHT–DeepPT 从 H&E slides 推断 transcriptomicsOdds ratio 2.28;预测应答者的应答率 +39.5%回顾性 cohort;不是已部署临床产品
文献综述和科学写作手工 PubMed 检索和起草K Pro FREE 提供可追溯、附来源链接的答案带回链 PubMed 的引用,综述更快免费层范围;未声称企业级验证
药物再定位 / biomarker 发现中心化汇集数据,会遇到隐私 / consent 摩擦在 MOSAIC + 医院网络上做 federated training不移动患者记录,也能访问更大数据池取决于可续约的医院数据协议

收益数字来自 Owkin 的 K Pro 发布公告和 ENLIGHT–DeepPT Nature Cancer 论文;生产率数字是单个内部案例主张,并未在客户群体中审计。

[CE004, CE006, CE018, CE021, CE022, CE026]
FE002: 使用 K Pro 的客户药物发现工作流

药企 R&D 问题如何通过 K Pro 流向带证据链接的优先级决策。

流程综合了 Owkin 描述的 K Pro 机制;中间步骤用于说明 Owkin 营销的“编排器 + 验证”模式。

[CE002, CE021, CE033]

5.2 技术架构与运营模式

Owkin 的技术栈是在联邦数据底座之上叠加生物学基础模型。模型层是 OwkinZero,一个面向生物学领域的大型推理模型,训练数据超过 300,000 对已验证问答,覆盖 8 类生物医学任务(包括靶点可成药性、模态适配性、药物扰动效应),并利用专有患者数据、空间转录组和 bulk 转录组学。配套 arXiv 论文显示,OwkinZero 模型通过对开源 8B–32B LLM 做 Reinforcement Learning from Verifiable Rewards 后训练得到;这些专用模型在生物学基准上超过更大的商业 LLM,并展现跨任务泛化能力。2026 年 1 月与 NVIDIA 的合作使用 Nemotron 开放模型家族和 NVIDIA NeMo RL 框架,扩展驱动 Owkin K 的下一代 OwkinZero。数据层由 MOSAIC 多模态肿瘤数据集和联邦医院网络组成:Owkin 不集中病历,而是用联邦学习让数据留在医院服务器,站点之间只传递模型更新。Owkin 将这一设计开源为 Substra(现由 Linux Foundation for AI & Data 托管,打包在 Kubernetes 中,可审计),并通过 FLamby 做基准测试。病理组织学的计算机视觉基础模型 Phikon(85.8M 参数 ViT-Base,用 iBOT 在 40M TCGA 图块上训练)和 Phikon-v2(ViT-Large,用 DINOv2 在 450M 图像 PANCAN-XL 语料上训练)提供切片级表征,输入 ENLIGHT–DeepPT 框架等生物标志物发现和治疗反应模型。这套多模态、联邦式、模型加数据的架构才是核心运营模式,而不是一个单体应用。[CE007, CE008, CE009, CE010, CE011, CE012]

技术与运营架构
层级 / 组件角色关键依赖风险
Agent 层——K Pro orchestrator将自然语言查询路由到专门 AI skills;返回附带证据链接的答案OwkinZero 模型 + 经整理的多模态数据编排准确性未经审计;科学输出存在幻觉风险
模型层——OwkinZero围绕可成药性、modality、perturbation 任务做生物推理开源基础 LLM(8B–32B);NVIDIA Nemotron / NeMo RL依赖第三方开放模型和 NVIDIA 算力
视觉基础模型——Phikon / Phikon-v2将整张 histopathology slide 编码为下游任务特征公开病理语料(TCGA、CPTAC、GTEx / PANCAN-XL)已发布 weights 为非商业许可;训练数据是公开数据而非专有数据
数据层——MOSAIC + 医院网络多模态肿瘤数据(临床、病理、基因组、空间)医院数据共享协议;federated nodes数据访问取决于可续约的机构合作
联邦学习底座——Substra / SubstraFL不移动原始数据,在多站点训练模型;日志可审计Linux Foundation OSS;Kubernetes;单站点基础设施分布式计算在运营上比集中式训练更重
验证层——patient validation hub用真实世界患者数据校验模型输出持续临床数据流入 + 用户反馈验证方法和覆盖范围未对外记录

架构由 Owkin K Pro / OwkinZero 页面、OwkinZero arXiv 论文、Substra 文档和 Phikon model cards 重构。层级边界是分析口径;Owkin 未发布正式 reference architecture diagram。

[CE003, CE007, CE008, CE011, CE013, CE015]
FE001: Owkin K 平台架构栈

分层呈现 Owkin K 平台:从自然语言智能体界面,到联邦医院数据底座。

分层是分析归类;Owkin 没有发布正式参考架构。NVIDIA 组件反映 2026 年 1 月宣布的合作。

[CE003, CE008, CE015, CE029]

5.3 部署、集成与差异化

K Pro 以安全企业平台交付:Owkin 将其定位为「安全、可解释、可互操作」,允许生物制药和学术客户安全上传专有数据集,与 Owkin 自有数据一起用于推理,并集成「患者验证枢纽」,让模型输出能用真实世界患者数据校验。企业合作验证了集成和采用:AstraZeneca、Bristol Myers Squibb 和 Sanofi 均参与其中——AstraZeneca 在 2026 年 5 月签下 3 年 K Pro 授权,Sanofi 则签署多年期 / 5 年协议,构建定制药物开发智能体。Owkin 的差异化来自竞争者难以轻易复制的资产:联邦医院网络在不集中患者数据的情况下产出专有 MOSAIC 多模态数据集;不是通用 LLM,而是生物学专用推理模型 OwkinZero;Nature Medicine 和 Nature Cancer 的同行评审验证;以及开源基础设施(Substra、FLamby)和公开基础模型(Phikon、Phikon-v2),共同支撑科学可信度和开发者足迹。K Pro 发布时披露的自报部署结果包括:内部药物靶点识别加速 70%(从 12+ 个月缩短到 3 个月)、数小时内做出可用于 IND 的资产定位策略、一天内产出后来发表的专家审查。这些结果均由公司披露,未经独立审计;平台还高度依赖第三方:医院数据共享协议、NVIDIA 算力、开放基础模型,以及核心药企合同的耐久性。[CE019, CE022, CE023, CE024, CE025, CE026]

FE003: 关键依赖图

K 平台依赖的外部供应商、数据权利、监管机构和合作伙伴。

依赖边反映公开描述的关系;合同排他性和数据权利持久性未披露。

[CE023, CE027, CE030]

5.4 信任、安全、隐私与合规

Owkin 的信任基础主要建立在保护隐私的联邦学习上:原始患者数据留在医院服务器,站点间只移动模型参数。这一设计意在尊重多机构网络中的数据所有权、患者隐私和保密性,对其欧洲医院基础下的 GDPR 合规至关重要。开源 Substra 框架进一步强化了这一点:所有机器学习操作都有可审计、只读日志,并以「定期审计」的 Kubernetes 包形式交付,让 Owkin 拥有纯集中式竞争者缺少的可追溯叙事。在受监管设备侧,CE 标志诊断产品 MSIntuit CRC(结直肠癌)和 RlapsRisk BC(乳腺癌复发)曾是 Owkin 最清晰的正式监管批准,但二者均在 2026 年 3 月转入 Waiv 剥离实体,因此 Owkin 自有组合已不再交付带 CE 标志的 IVD。K Pro 本身定位为研究和决策支持副驾驶——没有披露 FDA clearance 或作为医疗器械的 CE mark。作为研究工具,这一姿态合理,但也限制了其输出在没有进一步验证前能推动受监管临床决策的程度。Owkin 强调 K Pro「安全、可解释、可互操作」,并称洞见会用真实患者验证;但公开材料没有披露 SOC 2 / ISO 27001 认证、模型风险管理框架或独立审计结果,安全和模型治理控制仍是开放尽调项。[CE013, CE016, CE029, CE030, CE031, CE032]

信任、质量与合规控制
控制 / 认证状态范围缺口
Federated privacy(数据留在本地)已实施(核心架构)所有医院网络训练;支持 GDPR 对齐正式 DP / secure-aggregation 参数未公开说明
可审计 ML 操作日志(Substra)OSS framework 中可用Substra 上的 federated training runs企业版 K Pro 是否继承同一审计日志未披露
CE 认证(IVD)由 MSIntuit CRC / RlapsRisk BC 持有——现属 Waiv结直肠癌和乳腺癌诊断2026 年 3 月 spin-out 后,不再属于 Owkin 自有产品组合
K Pro 医疗器械 clearance未披露(研究 / 决策支持工具)Biopharma R&D 决策支持未获进一步批准前,不能驱动受监管临床决策
安全认证(SOC 2 / ISO 27001)公开页面未披露处理客户专有数据的企业 SaaS无公开 attestations;是企业采购尽调关键项

合规状态来自 Owkin federated-learning 和 K Pro 页面、Substra docs、Waiv spin-out 公告以及 FDA SaMD 指南。此处缺少某项认证,意思是公开页面未找到,不代表已确认不存在。

[CE013, CE016, CE029, CE030, CE031, CE032]

5.5 路线图、成熟度与产品风险

Owkin 宣称的方向是迈向「Biological Artificial Super Intelligence」(BASI):借助 NVIDIA 扩展 OwkinZero,把 K Pro 从肿瘤学延伸到相邻治疗领域(Sanofi 关系明确扩展到免疫学药物定位),并构建内部「AI Scientist」,自动发现新疗法和诊断。前瞻路线图由多年期药企承诺锚定——Sanofi 5 年定制智能体项目和 AstraZeneca 3 年授权——二者同时提供收入可见度和产品方向。组合成熟度并不均衡:已发表科学、开源 FL 基础设施和病理基础模型较成熟,并获得外部验证;智能体式 K Pro 产品仍早期(2025 年 10 月发布),其核心性能主张由公司自报。主要产品和技术风险包括:K Pro 没有 FDA clearance 或 CE mark,限制其用于受监管临床决策;模型可解释性缺口让监管申报更复杂;Waiv 剥离 将带 CE 标志的诊断线移出 Owkin 自身护城河;开发者生态(GitHub、Hugging Face)真实存在但相较公司野心仍偏小;Recursion、Insilico、Isomorphic 等全管线竞争者追逐相邻智能体式 AI 策略,带来复制风险。Owkin 在 2023 年(17% 员工)和 2025 年(约 15% 员工)向智能体式 AI 转型期间裁员,并伴随员工对内部管理的批评,也给路线图增加执行风险。[CE007, CE024, CE025, CE034, CE037, CE038]

路线图与开发阶段地图
时间 / 阶段里程碑状态含义来源
2025 年 10 月K Pro 作为 biopharma agentic AI co-pilot 发布已发布推动 Owkin 从服务转向可授权平台Owkin 发布公告;FierceBiotech
2025 年 8 月OwkinZero 生物推理模型发布(arXiv)已发布 preprintK Pro 推理的技术基础arXiv 2508.16315;Owkin OwkinZero 页面
2026 年 1 月与 NVIDIA 合作扩展 OwkinZero(Nemotron / NeMo)已宣布借助算力 + 开放模型向 BASI 扩展BusinessWire(NVIDIA 合作)
2026 年 5 月AstraZeneca 三年期 K Pro 许可已签署第三个 top-pharma 平台客户FierceBiotech;mHealthSpot
2026 →将 K Pro 扩展至免疫学;构建内部 AI Scientist进行中 / 计划中治疗领域扩张和自主发现野心Owkin(Sanofi 免疫学);BusinessWire

路线图条目按 Owkin 官方公告和 2026 年合作报道定日期;前瞻条目(「AI Scientist」、免疫学扩张)是已表述的目标,不是已上线能力。

[CE001, CE007, CE009, CE024, CE025, CE037]
FE004: 产品成熟度与能力图

各项能力在验证、部署和监管维度上的相对成熟度。

定性评级综合了发表状态、合作披露和监管记录;单元格概括证据质量,而不是数值评分。

[CE024, CE034, CE038]

5.6 展示材料

Chapter 06

06客户

6.1 药企合作伙伴基础与客户分层

Owkin 的付费客户基础分为两层。核心商业层是大型生物制药公司,它们签署多年期企业合作和 SaaS 授权协议,并通过预付合作费、经常性 K Pro 授权付款和里程碑触发的 R&D 付款组合付费。截至 2026 年 6 月,该层已确认的具名伙伴包括 Sanofi、Bristol-Myers Squibb、AstraZeneca、Servier、MSD/Merck(通过剥离后的 Waiv)、Amgen,以及 Johnson & Johnson。这 7 个名字覆盖全球药企头部梯队,也代表了 Owkin 公开量化交易价值的全部。 次级层由参与 Owkin 联邦学习网络的学术研究医院和癌症中心组成。18+ 个已配备学术站点——更广的数据访问网络覆盖 800+ 家机构——不是创收客户。它们的贡献是提供保护隐私的多模态患者数据集,支撑 K Pro 生物学推理可信度;Owkin 为这些中心配备 FL 基础设施,换取受管理研究协议下的数据访问。因此,医院网络参与方更适合被归为战略性数据生成伙伴,而不是商业账户。 第三类新兴客群是早期生物科技公司,可从 Micregen 证言以及其他 K Pro 被生物科技研究人员用于投资人推介材料准备和科学文献分析的表述中看到。该客群没有公开交易规模数据,体量也太小,暂无法评估为收入贡献者。地域结构以欧盟(法国、英国、德国)和美国为主,与 Owkin 巴黎总部、纽约办公室以及欧洲医院网络集中度一致。自 2026 年 3 月 Waiv 剥离后,AstraZeneca 和 MSD 成为独立 Waiv 实体的客户,其诊断产品收入不再并入 Owkin 账户。[CU001, CU002, CU003, CU004, CU005, CU006]

客户细分表
细分买方 / 用户 / 付款方主要用例规模(已知)收入 / 战略价值关键缺口
大型药企(企业)CDO / VP R&D(买方);研究科学家(用户);药企 P&L(付款方)用于发现、转化研究、临床试验设计的 K Pro agentic AI7 个具名合作方(Sanofi、BMS、AZ、Servier、MSD、Amgen、J&J)主要收入来源;多年期许可 + 里程碑付款;已披露交易价值 $270M+单客户 ARR 未披露;NRR/GRR 未披露
学术医院研究网络研究科学家 / 病理医生(用户);医院 / grant 预算(付款方)Federated learning 数据生成;贡献 MOSAIC 数据集18+ 个活跃 FL nodes;更广泛数据访问网络覆盖 800+ 家机构非收入;战略数据 moat,支撑 K Pro 差异化医院收入模型未披露;数据访问条款未披露
Biotech(新兴)BD / 科学负责人(买方兼用户)K Pro 用于科学文献综合、投资人 pitch、靶点发现1 个具名用户(Micregen);pipeline 未披露未披露;交易规模可能小于头部药企未发布企业许可条款或价格;细分规模未量化
前 Owkin Dx / Waiv 客户Pharma R&D / 诊断(买方);病理医生 / 临床实验室(用户)AI 诊断(MSIntuit CRC、RlapsRisk BC、gBRCA 预筛)AstraZeneca、MSD(经 Waiv);Medipath(法国最大病理网络)2026 年 3 月 spinout 后,收入不再并入 Owkin P&LWaiv 独立后,Owkin 失去该细分增长的可见度和经济利益

大型药企规模数字只反映公开具名合作方;活跃客户总数未披露。医院网络的 18+ 个活跃 FL nodes 来自 Owkin federated learning 页面;800+ 家机构反映 AZ K Pro 公告中引用的更广泛数据访问网络。Spinout 后的 Waiv 客户不并入 Owkin 账户。

[CU001, CU002, CU003, CU004, CU006, CU007]
FU001: 客户旅程图

Owkin 客户从发现到多年续约分五个阶段推进;Sanofi 已到达最深的续约阶段,大多数其他具名账户仍处于许可或扩展阶段,反映 K Pro(2025 年 10 月推出)的商业成熟度仍早。

阶段分配是基于公开公告的定性推断。AstraZeneca 从评估快速推进到许可,反映此前已有诊断 AI 关系。Micregen 的阶段来自单篇已发布证言,未确认企业许可。

[CU009, CU011, CU016, CU019, CU020, CU022]

6.2 具名客户部署与生产证据

Sanofi 是 Owkin 的旗舰客户,也是最早的商业伙伴,关系跨越连续三个阶段。2021 年 11 月,Sanofi 投入 $180 million 股权投资,并配套 $90 million 肿瘤合作,覆盖非小细胞肺癌、三阴性乳腺癌、间皮瘤和多发性骨髓瘤。到 2022 年 10 月,范围扩展到 Sanofi 免疫学管线中的药物定位。2026 年 6 月,合作再次深化为 5 年 K Pro 企业授权,Owkin 将构建定制智能体式 AI 系统,运行在 Sanofi 自有 IT 栈内。Sanofi 首席数字官 Emmanuel Frenehard 公开背书该安排,称专门构建的 智能体式系统将让 Sanofi 团队「以更高速度、更深洞察和更强信心运转」。 Bristol-Myers Squibb 于 2022 年 6 月进入多年期战略合作,覆盖由 AI 驱动的心血管疾病临床试验设计,包括协变量调整和外部对照组。交易包含 $80 million 预付款(股权 + 合作费),以及最高 $100 million 的成功里程碑付款。BMS 高级副总裁 Venkat Sethuraman 公开背书该合作,称其用「来自临床和真实世界来源的高质量、多样化数据」增强 BMS 项目。截至 2026 年 6 月,BMS 尚未正式宣布 K Pro 授权。 AstraZeneca 从 2024 年诊断 AI 合作(BRCAura gBRCA 突变预筛查,在 ESMO 展示结果中达到 93% 敏感性)推进到 2026 年 5 月签署 3 年 K Pro 企业授权。MSD 于 2023 年 12 月签署合作,为欧盟市场开发跨 4 种癌症类型的 MSI-H AI 诊断;这些资产在 2026 年 3 月转入 Waiv。Servier 于 2023 年 10 月进入多年期合作,方向是肿瘤转化医学。 Micregen 的 MD 提供了具名证言,确认使用 K Pro 准备投资人推介材料,并称用传统方法几乎「不可能」。Institut Gustave Roussy 研究部负责人称 K Pro 是快速多模态患者数据分析的「改变规则的工具」,这是除药企层以外,平台获得的第一份有独立归属的医院领域背书。[CU009, CU010, CU011, CU012, CU013, CU014]

具名客户证明表
客户细分部署 / 用例生产环境 vs 试点结果证据关键限制
Sanofi大型药企肿瘤学生物标志物识别 / 靶点发现(2021–2024);免疫学药物定位(2022+);K Pro agentic AI agents(2026–)生产化 — 2026 年 6 月签署 5 年期 K Pro 企业许可;此前合作已完全执行CDO Frenehard 公开背书 K Pro;$90M 肿瘤学合作已完成;Sanofi 投入 $180M 股权投资尚未发布 K Pro 成果指标;2026 年 K Pro 部署仍处最早期;Sanofi 同时与 CytoReason 合作
Bristol-Myers Squibb大型药企用 AI 设计心血管疾病临床试验;协变量调整;外部对照臂生产化 — 多年期合作自 2022 年 6 月起持续;已承诺 $80M 首付款SVP Sethuraman 公开背书该方法;公司称这是大规模使用 AI 设计试验尚未发布量化试验结果改善;BMS 也与 Anthropic 签署 AI 协议(2026 年 5 月);未宣布 K Pro 许可
AstraZeneca大型药企gBRCA 突变 AI 预筛(2024,BRCAura);用于竞争情报与药物发现的 K Pro 企业许可(2026)生产化 — gBRCA 工具处于研究使用;2026 年 5 月签署 3 年期 K Pro 企业许可BRCAura 达到 93% 敏感性,排除了 40% 非携带者患者(ESMO 结果)gBRCA Dx 资产转至 Waiv;K Pro 财务条款未披露;AZ 正在并行管理多个 AI 合作
MSD (Merck & Co)大型药企面向子宫内膜癌、胃癌、小肠癌和胆道癌(EU)的 MSI-H AI 诊断生产化合作 — 2023 年 12 月建立伙伴关系;资产于 2026 年 3 月转至 Waiv已正式宣布具名合作;MSI-H 预筛工具覆盖四类癌症截至 2026 年 3 月,全部诊断资产已归入 Waiv(独立实体);Owkin 不再直接确认收入
Micregen生物技术公司K Pro 用于准备投资人路演和综合生物医学文献试点 / 早期使用 — 公司 MD 具名证言MD Ben Mellows 公开表示,K Pro 在几个工作会话中处理的规模,原本几乎不可能完成单一生物技术公司证言;未披露企业许可或交易规模;不能确认持续关系
Servier中型药企肿瘤学转化医学和数字病理;多年期合作自 2023 年 10 月起保持活跃合作多家独立新闻来源报道;确认多年期范围未披露财务条款;无成果指标;未宣布 K Pro 许可

覆盖范围并不完整 — 代表截至 2026 年 6 月已公开宣布或确认具名的药企与生物技术客户。未披露的早期关系(如 Amgen、J&J 早期试点)已纳入 TU001,但证据不足,未列入本表。Waiv 分拆后的客户也一并列入,因为它们源自 Owkin。

[CU009, CU010, CU011, CU012, CU013, CU014]
FU003: 客户证据评分矩阵

Sanofi 和 BMS 提供了最强的综合证据质量(高管引述加披露交易价值);AstraZeneca 的独特点在于,它是唯一发布结局数据的客户(BRCAura ESMO 结果);多数其他客户缺少披露交易价值或量化结局,反映 K Pro 部署仍处早期。

数值为二元(1=存在,0=不存在),基于截至 2026 年 6 月的公开证据。AstraZeneca 的“已发布结局数据”指 BRCAura ESMO 结果(2024 年 gBRCA 研究);所有客户仍未发布 K Pro 特定结局。Micregen 的“已发布结局数据”来自投资者推介证言,其中描述了可衡量的任务完成。

[CU024, CU028, CU031, CU035, CU037, CU039]

6.3 K Pro 商业采用与采用轨迹

K Pro 于 2025 年 10 月商业发布,面向药企、生物科技公司和投资人,通过自然语言界面支持生物医学研究。平台结合 Owkin Zero(微调后的生物学 LLM)、来自 MOSAIC 联盟和更广医院网络的多模态患者数据,以及智能体式 AI 架构。发布时,Owkin 声称内部性能基准显示药物靶点识别加速 70%(从 12+ 个月缩短至约 3 个月),且至少一个未具名药企合作通过改进患者分群识别将试验周期缩短 35%。这些均为公司主张,未经独立验证。 2025 年 10 月发布后的 7 个月内,Owkin 拿下两项已宣布企业授权交易:Sanofi 5 年协议(2026 年 6 月)和 AstraZeneca 3 年协议(2026 年 5 月)。二者都涉及 Owkin 构建专用 AI 智能体,并部署在客户自有 IT 基础设施内——这种深度嵌入的部署模式会形成机构切换成本。K Pro 发布前,Owkin 与 STAT Brand Studio 对 200 多名药企高管做了一项调查,近半数受访者认为克服数据挑战是智能体式 AI 的主要机会。该调查由 Owkin 委托,不能视为独立市场验证。Owkin 2024 年约 $86 million 收入、同比增长 83%,与药企合作收入增长相符,但无法按产品或客户拆分。[CU023, CU024, CU025, CU026, CU027, CU028]

客户增长与采用轨迹表
指标数值日期来源依据置信度含义缺失分母 / 注意事项
具名药企企业合作伙伴7 家(Sanofi、BMS、AZ、MSD、Servier、Amgen、J&J)截至 2026 年 6 月Owkin 新闻稿 + Wikipedia + 独立媒体代表可核验的完整具名管线;未披露客户总数可能更高未披露企业账户总数;管线转化率未知
活跃 K Pro 企业许可(已确认)2(Sanofi 5 年、AstraZeneca 3 年)2026 年 6 月Owkin 官方公告;FierceBiotech;R&D WorldK Pro 企业渗透仍早,但上线 7 个月内已覆盖两家全球前十药企交易财务条款未披露;可能还有未宣布许可
K Pro 上线至首个披露许可< 7 个月(2025 年 10 月上线 → 2026 年 5 月 AZ 交易)2026 年 5 月FierceBiotech;mHealthSpot;Owkin 官方(AZ 公告)从上线到许可的企业转化很快;与既有 Owkin–AZ 关系一致K Pro 销售周期和管线规模未公开披露
医院数据网络规模800+ 家机构(更广泛数据访问);18+ 个活跃 FL 节点2026 年 6 月FierceBiotech AZ 交易新闻稿(800+);Owkin FL 页面大型机构数据网络支撑 K Pro 差异化FL 节点数来自 Owkin 网站;更广泛的 800+ 数字未获独立确认
收入(FY 2024)~$86 millionFY 2024Maddyness(法国监管披露)较 2023 年同比增长 83%;与药企合作加深相符FY 2025 收入尚未披露;无法按客户或产品拆分
药企高管 AI 优先级调研200+ 名受访药企高管中约 50%2025 年 10 月(上线前)Owkin + STAT Brand Studio(公司委托)支撑 K Pro 产品市场契合叙事调研由 Owkin 委托并设计;不是独立市场验证

收入数字来自 Maddyness 引用的法国监管披露(2024 年数据)。企业 K Pro 许可数量仅反映已公开宣布的协议;公司可能还有未披露的早期访问或试点关系。

[CU002, CU023, CU024, CU027, CU028, CU029]
FU002: 采用部署漏斗

K Pro 的漏斗顶部是全球前 50 大药企;截至 2026 年 6 月,已有七项具名合作确认,两家公司签署 K Pro 企业许可,一家(Sanofi)在 K Pro 推出后 7 个月内进入扩展 / 续约阶段。

漏斗顶部(50)是按 R&D 支出排名的前 50 大药企示意范围;Owkin 的实际管线可能包括更小的生物技术公司和其他未披露药企账户。具名合作(7)和活跃合作(5)仅基于公开公告。

[CU002, CU023, CU024]

6.4 留存、耐久性与扩张信号

Owkin 不披露净收入留存、总收入留存,也不披露任何结构化队列数据。缺少这些指标是核心尽调缺口:没有 NRR 或流失 数据,就无法独立验证商业牵引力是否耐久。 间接留存信号偏正面。Sanofi 合作在 5 年里持续加深(2021 股权 + 肿瘤 → 2022 免疫学扩展 → 2026 K Pro),同一旗舰客户连续进入三轮价值创造阶段。AstraZeneca 从诊断 AI 试点(2024 BRCAura)推进到 K Pro 企业授权(2026),显示清晰的交叉销售模式。截至 2026 年 6 月,没有药企客户公开宣布退出或拒绝续约 Owkin 合作,尽管 MSD/AstraZeneca 诊断关系转移给 Waiv,而不是自然终止。 已披露 K Pro 交易的合同期限(Sanofi 5 年、AstraZeneca 3 年)显示承诺周期结构性较长,符合深度嵌入智能体式 系统带来的高切换成本。Owkin 的联邦学习基础设施在数据层形成持久的机构切换成本:一个中心一旦集成 Owkin 的 FL 工具,替换就需要大量重新布设。药企企业客户面对不同的切换动态——Sanofi 同时与 CytoReason 合作,BMS 同时与 Anthropic 合作,说明客户粘性并非排他,头部药企账户通常会并行评估多个供应商。[CU030, CU031, CU032, CU033, CU034, CU035]

留存、重复使用与满意度表
指标数值 / 状态分部置信度尽调请求
净收入留存(NRR)null — 未披露全部企业客户尽调中要求按 cohort 年份(2021、2022、2023、2024)提供 NRR;区分药企平台 NRR 与诊断(Waiv) NRR
总收入留存(GRR)null — 未披露全部企业客户要求同时提供 GRR 和流失事件日志;确认不存在未公开的合同不续约
Sanofi 合作持续时间连续并加深合作 5+ 年(2021–2026)旗舰药企验证 Sanofi 后续每个阶段是否都经历竞争性重新选择,以确认真实留存质量
AstraZeneca 扩张信号交叉销售已确认:Dx 项目(2024)→ K Pro 3 年期许可(2026)大型药企确认 AZ 内部采用 K Pro agents 的广度;询问正在活跃使用 K Pro 的研究团队数量
BMS 合作续约状态未宣布 K Pro 续约或升级;合作自 2022 年起活跃大型药企确认 BMS 合作范围、交易到期时间,以及 K Pro 追加销售是否在管线中;BMS 的 Anthropic 交易构成竞争信号
客户满意度分数(CSAT / NPS)null — 未披露全部分部要求提供 K Pro 企业试点用户匿名 CSAT 或 NPS 分数

公开记录缺失所有正式留存指标(NRR、GRR、CSAT、NPS)。定性信号(合作加深、交叉销售)只能作为代理证据。置信度为 null 表示该指标完全未经验证。

[CU030, CU031, CU032, CU033]
合同期限与续约信号表
客户初始交易日期交易类型期限 / 合同期扩张事件截至 2026 年 6 月的续约状态
Sanofi2021 年 11 月股权($180M)+ 3 年期肿瘤学合作($90M)3 年(肿瘤学),随后续约并扩张2022:免疫学扩张(价值未披露);2026:5 年期 K Pro 许可活跃并加深 — 2026 年 6 月签署 5 年期 K Pro 合同
Bristol-Myers Squibb2022 年 6 月股权部分($80M)+ 基于里程碑的合作(最高 $100M)多年期(具体期限未披露)未公开 K Pro 追加销售;BMS 新增 Anthropic AI 合作(2026 年 5 月)不明确 — 未宣布 K Pro 许可;原合作仍活跃,但未公开续约
AstraZeneca2024(Dx 合作);2026 年 5 月(K Pro 许可)诊断 AI 合作 → K Pro 3 年期企业许可3 年(K Pro);诊断合作未披露2024 → 2026:从诊断交叉销售至 K Pro 平台活跃并扩张 — 2026 年 5 月签署 3 年期 K Pro 许可
MSD (Merck)2023 年 12 月诊断 AI 合作(EU 市场)未披露2026 年 3 月:资产在分拆时转至 Waiv已转至 Waiv — 与 Owkin 无直接 K Pro 关系
Servier2023 年 10 月多年期肿瘤学 + 数字病理研究合作多年期(具体期限未披露)未宣布扩张或 K Pro 追加销售活跃 — 早期合作,自 2023 年以来无公开更新
Micregen2025(试点使用)K Pro 试点 / 早期访问Unknown未宣布扩张已由公开证言确认早期使用;未宣布企业许可

本表替代原计划的留存 / 重复 cohort 图,因为 Owkin 企业合作尚无 NRR 百分比 cohort 数据(按时间桶 0–100% 留存)。所有合同条款均来自公开公告和独立媒体报道;单个年度费用、里程碑金额和续约选项均未公开披露。

[CU031, CU032, CU033, CU034]

6.5 客户集中度、反向信号与风险评估

Sanofi 在 Owkin 公开交易记录中的主导地位非常清楚。单一 Sanofi 关系贡献了 $180 million 股权投资、$90 million 肿瘤合作付款、免疫学扩展(金额未披露),以及 2026 年 6 月 K Pro 5 年授权(金额未披露)——合计构成任何 Owkin 交易对手中最大且最持续的商业承诺。没有单个客户 ARR 披露,精确收入集中度无法量化;但单一客户横跨每一条收入渠道的定性主导地位,对一家尚未盈利且尚未完成下一轮融资的公司构成重大投资风险。 反向信号方面,Sifted 2025 年 9 月报道称,离开 Owkin 的员工提到「严重管理问题」,显示内部交付能力可能跟不上多年期企业 AI 部署的商业野心。约 50 个岗位(约法国员工 15%)的重组被包装为 AI 智能体转型,但同时出现的内部批评说明,风险不只是人员规模管理。BMS 2026 年 5 月另行与 Anthropic 建立 AI 合作,Sanofi 也与 CytoReason 并行合作,确认大型药企客户管理的是多元 AI 供应商组合,而不是独家押注 Owkin。2026 年 3 月 Waiv 剥离改善了 Owkin 的战略聚焦,但因两个具名客户从合并账户移出,也减少了可见客户多样性。截至 2026 年 6 月,Owkin 地域结构仍集中在欧盟和美国,没有证据显示其在亚太有实质商业渗透。[CU037, CU038, CU039, CU040, CU041, CU042]

扩张与集中度风险表
扩张驱动因素 / 风险因素集中度风险若成真影响尽调路径
Sanofi 收入集中关键 — Sanofi 很可能占累计交易价值的大多数;$180M 股权 + $90M 肿瘤学 + K Pro 许可若 Sanofi 合同流失或支出减少,收入和对新潜在客户的信号质量都会严重受损获取管理账,显示 Sanofi 在 2024 年收入中的占比;确认 2026 年 K Pro 年费和里程碑时间表
K Pro 在 Sanofi 内部 land-and-expand正面 — 5 年期嵌入式 agent 部署带来多年收入可见度和追加销售机会K Pro agent 开发成功会加深锁定;交付失败则触发声誉和收入风险跟踪 Sanofi 在 ESMO / 投资者活动中关于 K Pro 成果的公开表述;要求提供里程碑付款时间表
AstraZeneca 交叉销售已验证中等 — 3 年期 K Pro 许可确认从 Dx 到平台的交叉销售,但 Waiv 分拆移除了 Dx 收入AZ 内部 K Pro 扩张是正面扩张信号;Dx 收入流失部分抵消该信号确认 AZ 的 K Pro 许可是否包含里程碑付款,还是纯 SaaS;量化年度许可费
药企多供应商 AI 策略重要 — BMS + Anthropic(2026 年 5 月)、Sanofi + CytoReason(并行)— 药企客户在分散 AI 供应商如果竞争 AI 平台更快交付结果,Owkin 可能在预算分配中被降级跟踪 Owkin 具名客户中的竞争 AI 平台胜单;评估 K Pro 相对 frontier LLM APIs 的差异化
医院数据网络作为护城河正面 — 800+ 家机构数据网络带来差异化 AI 训练数据;机构惯性让流失率偏低关键学术数据伙伴流失会削弱数据集质量和 K Pro 可信度审查数据访问协议的排他性、期限和退出条款
Waiv 分拆降低客户多样性中等 — AZ 和 MSD 移出 Owkin P&L;剩余具名账户更集中当前收入基础更集中在少数更大的药企关系上获取 Waiv 股权价值,以及 Owkin 与 Waiv 之间任何持续商业数据共享协议

集中度严重性评估为定性判断,来自已公开交易价值和结构。单个客户 ARR 未披露。「关键」Sanofi 集中度基于可观察到的披露交易经济性:相对其他具名伙伴,Sanofi 占据主导。

[CU035, CU036, CU038, CU039, CU040, CU041]

6.6 展示材料

Chapter 07

07风险

7.1 监管与法律风险

Owkin 的监管暴露横跨三个不同框架:EU AI Act、GDPR Article 9 健康数据义务,以及美国 FDA 对 AI 医疗器械的监管。EU AI Act 分阶段全面生效:2025 年 2 月适用禁止类 AI 规则,2025 年 8 月适用通用 AI(GPAI)模型义务,2026 年 8 月 2 日法案全面适用。OwkinZero——K Pro 核心的 Owkin 微调生物学大语言模型——属于 GPAI 模型,因此自 2025 年 8 月起承担透明度、版权合规和系统性风险义务。具体而言,Owkin 必须发布 OwkinZero 训练数据内容摘要,维持版权政策,并评估 OwkinZero 是否构成系统性风险(高能力阈值)。未找到 OwkinZero 的公开 GPAI Code of Practice 备案或训练数据透明度摘要。European AI Industry Alliance 的 GPAI Code of Practice 于 2025 年 7 月发布,是 Owkin 应遵循的自愿合规工具;不采用它并不构成违法,但会失去符合性推定利益。 对高风险 AI 分类而言,有一个重要缓释因素:EFPIA 立场文件确认,仅用于药物 R&D 的 AI 工具很可能根据 Articles 2.6 和 2.8(科学研究豁免)免于 EU AI Act 高风险分类。K Pro 若仅部署在药企 R&D 工作流中,而不进入临床诊断,可能因此避开 CE 标志要求。但该豁免存在争议,且需逐案判断;任何从研究跨入临床决策支持的部署,都可能触发高风险分类。根据 2026 年 5 月 AI Omnibus 政治协议,通用领域 AI 系统(生物识别、就业等)的高风险规则将自 2027 年 12 月起适用。 GDPR 方面,Owkin 根据 Article 9 处理特殊类别健康数据。该条禁止处理,除非适用特定例外(Art.9(2)(j) 下的科学研究,或 Art.9(2)(a) 下的明确同意)。Owkin 隐私政策最近更新于 2026 年 6 月 8 日,确认 GDPR 合规承诺,以及欧洲数据使用 HDS 认证云基础设施。跨境传输到 Owkin US 使用标准合同条款。法国 CNIL 对 Owkin 法国实体具有监管权;任何医院节点的数据泄露或同意失败,都可能触发 CNIL 调查,罚款最高为 €20 million 或全球年营业额 4% 二者中较高者。本研究未发现针对 Owkin 的执法行动。 美国方面,Owkin 的 K Pro 平台定位为药企科学家的研究和决策支持工具,而不是 FDA 监管医疗器械。FDA 的 AI-Enabled Medical Devices 清单确认了临床 AI 诊断的独立路径(例如 Waiv 现持有的前 MSIntuit CRC 和 RlapsRisk BC 产品)。2026 年 3 月 Waiv 剥离将这些 CE 标志诊断产品移出 Owkin 监管边界。K Pro 嵌入的开源权重模型组件(Qwen、其他微调模型)带来潜在但尚未量化的 IP 风险,投资前需要完整 IP 审计。[CR001, CR002, CR003, CR004, CR005, CR006]

监管 / 法律风险登记表
风险 / 规则司法辖区状态(2026)可能性(1-5)严重性(1-5)主要缓解措施剩余暴露尽调路径
EU AI Act GPAI 义务 — OwkinZero 训练数据透明度、版权合规、系统性风险评估EU自 2025 年 8 月起生效44采纳 GPAI Code of Practice(自愿);OwkinZero 模型卡披露高 — 未找到公开 GPAI 备案或训练数据摘要要求提供 GPAI 合规路线图;确认 OwkinZero 系统性风险自评
GDPR Art. 9 — 180+ 个医院节点上的特殊类别健康数据EU / France持续35HDS 认证云、假名化、跨境传输 SCCs中高 — 去中心化同意框架在医院层面执行;受 CNIL 监管审查主要医院节点的数据处理协议和伦理委员会批准
EU AI Act 高风险分类 — K Pro 进入临床决策支持工作流EU监测中(规则 2027 年 12 月适用)24EFPIA 豁免主张(Art. 2.6 科学研究);K Pro 定位为药企 R&D 工具,而非医疗器械中 — 如果 K Pro 输出影响实际临床决策,边界不清获取关于临床使用边界的法律意见;确认 CE 标记诊断风险由 Waiv 持有
US FDA — 面向临床使用 AI 的 AI 医疗器械监管US监测中13K Pro 未被归类为 FDA 监管器械;Waiv 持有原 CE 标记诊断低 — 只要维持研究工具定位确认 K Pro 功能没有待审 FDA pre-submission 或 510(k) 申报
IP / 开放权重模型许可 — K Pro 中的 Qwen 和其他 LLM 组件全球潜在23商业许可审查;Owkin 微调开放权重模型中 — 上游许可变化或模型弃用可能要求重新训练对嵌入 K Pro 和 OwkinZero 的全部第三方模型组件进行完整 IP 审计
法国劳动法 — 2025 年 rupture conventionnelle collective 的剩余索赔France基本解决(2025 年 9 月)12程序按法国劳动法完成;CSE 已获告知低 — 集体协议结构降低个人索赔风险确认 2025 年重组没有待决劳动法庭程序或工会挑战

可能性和严重性按 1(最低)到 5(最高)评分,由分析师结合 EU AI Act 文本、EFPIA 立场文件和 Owkin 公开披露判断。状态日期来自 EU AI Act 时间线(digital-strategy.ec.europa.eu)以及 2026 年 5 月 AI Omnibus 政治协议。本研究未发现针对 Owkin 的正式监管执法行动。

[CR001, CR002, CR003, CR004, CR005, CR006]
FR001: 风险热力图——可能性 vs. 影响(Owkin,2026 年 6 月)

八类已识别风险按可能性(1 = 罕见,5 = 可能)与影响(1 = 最小,5 = 严重)绘制,剩余严重性标签由综合得分推导。

可能性和影响是分析师的定性估计;没有量化概率模型支撑这些评分。剩余严重性是叙述性标签,不是数学乘积。

[CR001, CR005, CR019, CR020, CR027, CR028]

7.2 技术与运营风险

Owkin 的核心技术风险聚集在三条轴线:模型泛化、智能体式 AI 可靠性、分布式基础设施韧性。 模型泛化是联邦医疗 AI 中结构性最难的风险。跨 180+ 医院节点训练,会引入来自不同患者群体、影像系统和病理流程的非 IID(non-independently-and-identically distributed)数据。联邦学习市场文献确认,数据异质性是限制联邦模型准确率的首要因素,需要专门聚合算法控制漂移。Owkin 已发表联邦学习方法论文,但未找到 K Pro 或 OwkinZero 在分布外医院数据上表现的独立外部审计。K Pro 声称药物靶点识别加速 70%,该数字由公司披露并出现在自发新闻稿中;截至 2026 年 6 月,尚未被独立复现。 智能体式 AI 幻觉是第二个关键风险。K Pro 部署自主智能体,围绕生物学文献和多模态患者数据推理,合成假设,并在数小时内生成可用于 IND 的 资产定位策略。若错误生物学推断以高置信输出呈现,可能误导耗资数亿美元的药企 R&D 项目。Owkin 所称缓释措施——人工审查和「科学家优先」副驾驶 设计——可以降低但不能消除该风险。公司没有公开披露系统性失败率或错误率。 跨 180+ 医院节点的基础设施可靠性会带来分布式故障风险。重要枢纽流失——无论来自医院数据协议终止、IT 系统不兼容,还是数据使用面临法律挑战——都可能削弱训练数据多样性和模型质量。Owkin 患者信息页面确认 HDS 认证云存储和假名化,但未披露 uptime SLA 或冗余架构。联邦学习理论上也容易受到模型反演和成员推断攻击;联邦架构虽减少原始数据暴露,但梯度共享在对抗条件下可能泄露训练记录信息。Owkin 未披露防御此类攻击的架构。 此外,Owkin 依赖专有 Qwen(以及其他开源权重 LLM)作为底层推理引擎,会受上游模型提供方决策影响。如果 Qwen 授权条款变化或模型弃用,Owkin 必须在替代基础模型上重新微调,带来成本和时间风险。[CR027, CR028, CR029, CR030, CR033, CR034]

运营、质量与安全风险登记表
失效模式可能性(1-5)严重性(1-5)缓解成熟度剩余暴露未解决缺口
模型泛化失败 — 非 IID 医院数据人群之间出现分布偏移44中(联邦验证协议;多站点训练)未发布独立外部性能审计;只有公司自有基准
Agentic AI 幻觉 — K Pro agent 给出错误生物学推断,影响药企项目决策35低(声称 human-in-loop 审查;未披露系统性错误率)未披露失败率;没有正式不良输出事件日志
医院数据访问终止 — 主要节点退出网络或撤销数据协议24中(长期伙伴关系;联邦架构降低单节点关键性)单个医院合同期限和终止条款未公开
医院节点健康数据泄露 — 假名化数据被重新识别或泄露25中(HDS 认证、加密、假名化)中高未披露渗透测试结果或第三方安全审计
AI 模型偏差 — 在代表性不足患者群体上系统性表现较差33低(声明 Privacy by Design;未披露偏差审计)未发布 OwkinZero 或 K Pro 的独立偏差 / 公平性评估
上游 LLM 依赖 — Qwen 或其他开放权重模型被弃用或重新许可23低(公开表述显示 Owkin 监控 Kimi 和其他替代项)未披露正式应急计划或模型可移植架构

评分为分析师估算,基于联邦学习文献(Precedence Research)和 Owkin 产品披露。严重性分数反映最坏情境下对 Owkin 药企客户关系和收入的影响。缓解成熟度:低 = 仅有声明;中 = 有部分落地证据;高 = 经独立验证(本类别无)。

[CR027, CR028, CR029, CR030, CR033, CR034]

7.3 商业集中度与财务风险

商业集中度风险是最直接的投资担忧。Owkin 已披露药企合作——Sanofi、Bristol-Myers Squibb 和 AstraZeneca——几乎肯定贡献了主要收入。Sanofi 尤其集中:它同时是 Owkin 的锚定股权投资人(2021 年 $180M Series B)、最大已披露合作伙伴(2021–2024 年 $90M 肿瘤合作,2026 年起转为 5 年 K Pro 企业授权),也是发布当天公开验证 Owkin 技术的公司 CDO 所在方。若 Sanofi 退出或实质降低参与度——无论因为 Sanofi AI 战略变化、并购、管线优先级重排,还是竞争替代——都会同时冲击收入、移除关键信誉锚,并让依赖 Sanofi 隐性背书的未来融资更复杂。Sanofi 也没有把 Owkin 当作独家 AI 伙伴:2026 年 1 月,Sanofi 扩大与 CytoReason 的并行合作(价值最高 $16M),显示其愿意把 AI 预算分散给多个供应商。 BMS 和 AstraZeneca 增加了关键规模,但并未从结构上分散集中度。二者均签有多年期协议,财务条款未公开披露。AstraZeneca 交易为 3 年期(2026–2029);到期后,Owkin 将面对与竞争者重新采购的周期。 财务风险集中在资本充足性和现金续航。Owkin 母体上一次一级股权融资在 2022 年 6 月完成——已相隔 4 年。Maddyness 2025 年 9 月报道称,Owkin 持有 $200M 现金,2024 年净亏损 $5.6M(较 2023 年 $31.3M 改善),2024 年收入 $86M。公司截至 2025 年底寻求新一轮 $200–300M 融资,并预计从 2025 年末起 5–6 年内盈利。按约 470 名员工规模估算,Owkin 每月现金流出为 $3–9M,取决于基础设施、R&D 强度和可变薪酬。若收入里程碑延后,或新一轮以较低估值完成,资本配置选项会收窄。2026 年 3 月 Waiv 剥离给 Waiv 资产负债表增加了现金(Waiv 融资 $33M),并未给 Owkin 增加现金,同时移除了一条收入流(来自 AstraZeneca、MSD 等的临床诊断按次收费)。来自 BMS 和未来伙伴的里程碑付款在结构上不平滑:药企 R&D 周期长达 5–10 年,里程碑触发取决于 Owkin 不控制的外部验证事件。 二级市场隐含估值约 $1.6B(Premier Alts,2026),高于 2021 年 $1B 的 Series B,但在没有确认 2026 年股权融资的情况下,缺少一级市场佐证,只能说明二级市场乐观。[CR012, CR013, CR014, CR015, CR016, CR020]

合作伙伴与依赖风险登记表
依赖对手方角色集中度关键失效情境严重性(1-5)缓解措施剩余暴露
锚定药企投资方 + 最大客户Sanofi SA股权锚定方($180M)、收入来源(5 年期 K Pro 许可)、声誉验证方关键Sanofi 不续签 K Pro 许可或退出 AI 战略;Owkin 同时失去大部分收入和投资者可信度55 年期合同锁定至 2031 年;K Pro agents 深度嵌入 Sanofi 工作流非常高 — 在 Sanofi 规模上没有多元化兜底
战略股权投资方 + 客户Bristol-Myers Squibb股权投资方($80M)、客户(心血管 + 肿瘤 AI)BMS 将 AI 支出重新优先分配给内部能力或竞争平台4$80M 股权绑定;2022 年最高 $100M 的里程碑结构形成持续财务激励高 — 合同续约条款不清;BMS 也与 Exscientia 合作
客户(2026 年新增)AstraZeneca plc3 年期 K Pro 许可(竞争情报 agents)中等3 年合同 2029 年到期;AstraZeneca 选择不续约3MCP 架构深度嵌入工作流;此前在 Waiv 有 gBRCA 合作中等 — 2029 年续约敞口
患者训练数据网络180+ 家学术医院支撑 AI 模型训练的多模态患者数据来源监管挑战、同意争议,或医院 IT 变更导致主要节点终止数据访问4联邦架构降低单一节点关键性;长期学术合作关系中等 — 未公开合同期限或终止条款
受监管云基础设施HDS 认证云服务商(AWS、Azure 或同等服务商)欧洲患者数据存储与处理基础设施中等HDS 认证取消、云故障,或数据主权法规变化2按法国法律使用 HDS 认证服务商;假定采用多区域架构

交易对手角色来自公开披露(Owkin newsroom、Pharmaphorum、BusinessWire、Maddyness)。 各客户收入分成未披露;严重性评分反映关系终止后可能冲击 Owkin 总收入和估值的程度。 Sanofi 被评为关键级别,因为它同时是股权投资方和主要客户,且未披露合同是否可切分。

[CR020, CR021, CR022, CR023, CR024, CR036]

7.4 人员、治理与执行风险

Thomas Clozel 的关键人集中度,是最严重的治理风险。Clozel 是公司创始人、CEO、与大型药企 CDO 和 CEO 达成交易的主要推动者、投资者关系负责人,也是 Owkin Biological Artificial Superintelligence(BASI)愿景的公众面孔。他自 2016 年公司创立以来一直任职。没有公开文件显示继任计划或副 CEO 路径。若 Clozel 因疾病、离职或被竞争对手招募而离开,由个人信任锚定的 Sanofi 和 BMS 关系将面临风险。 Gilles Wainrib(CSO)是第二位关键人物,从学术背景出发负责科学战略。他同时担任教授和初创公司 CSO,这种双重角色带来学术创始人公司常见的部分可用性风险。未发现 Wainrib 离职时已有指定科学继任者的证据。 2025 年 9 月裁员约 50 人(15%),按法国法下 rupture conventionnelle collective 设计,聚焦新智能体式 AI 核心之外的岗位。Maddyness 确认了裁员及其财务理由。Sifted 引用的一名消息人士称,当时 Owkin 存在「严重管理问题」, 与公司从诊断转向纯智能体式 AI 所需的组织重塑相符。重组移除了诊断人才(转入 Waiv),并削减了被视为非 K Pro 核心的职能人数。重组后的工程留存风险尚未验证。同样,2024 年 2 月 Bioptimus 剥离(获 €35M 资金支持的基础模型初创公司)也让一批 AI 研究人员离开 Owkin 直接员工队伍。 治理不透明是另一项执行风险。Owkin 是法国私营 SAS,不公开董事会构成,没有发布审计委员会或薪酬委员会程序,也没有自愿 ESG 或 DEI 报告。缺少独立董事会监督信号,使外部难以验证财务控制、数据治理政策和管理决策是否接受独立审查。Sanofi 保留董事会代表席位(作为投资人)提供了一定外部约束,但 Sanofi 同时作为客户和投资人的利益,可能与少数财务投资人并不完全一致。[CR017, CR018, CR019, CR040, CR041, CR042]

人员与执行风险登记表
角色 / 职能依赖或缺口可能性(1-5)严重性(1-5)缓释措施尽调路径
CEO — Thomas Clozel MD单一关键人:主导药企合作谈判、投资者关系、BASI 愿景沟通;未披露继任计划35联合创始人履历扎实;Sanofi 关系部分已通过合同制度化确认继任计划是否存在;核验副 CEO Pascal Weinberger 能否完全替代;关键人物寿险 / 伤残险
CSO — Gilles Wainrib PhDAI 研究战略负责人;学术与产业双重角色带来可用时间和承诺冲突风险24技术联合创始人根基深;学术网络提供研究管线评估 AI 研究团队梯队深度;确认其对 Owkin 是全职或接近全职投入
工程 / 平台团队(2025 年重组后)2025 年 9 月裁员 15% 可能削减了平台工程师;Bioptimus 分拆(2024 年)带走 AI 研究团队33按既定战略聚焦招聘 agentic AI 岗位索取重组后的组织架构图和按职能拆分的人数;评估 K Pro 工程团队留存
商业 / BD 负责人 — Jean-Frédéric Petit-Nivard(CCO)Sanofi、BMS、AstraZeneca 交易的关键客户管理;未来药企关系管线23已披露兼具商业和科学背景的 C-suite向药企对口方做 reference check;评估 CRM 管线能否带来新客户多元化
医院网络关系经理180+ 家学术医院的数据访问协议要延续,需要主动维护关系23学术合作关系多年;联邦架构降低单站点关键性核验前 10 大医院节点的合同管理职能和续约日历

严重性评分反映相关岗位空缺且无人替代后,对 Owkin 商业和技术能力的冲击。 可能性评分反映行业薪酬惯例和 Owkin 近期重组背景下的感知流失风险。没有内部人数数据; 缺口仅依据公开披露识别。

[CR017, CR018, CR019, CR040, CR042]

7.5 否决标准、缓释框架与尽调要求

Owkin 的投资论点,是对三项相互连接主张的集中押注:(1)OwkinZero/K Pro 相比前沿通用 LLM 能产出真正更优的生物学推理;(2)Sanofi 和 BMS 会深化而不是退出其平台承诺;(3)Owkin 能在现金吃紧前完成新一轮股权融资,并在约 2030–2031 年实现盈利。每一项都有可衡量的否决标准。 论点破裂触发器 1(模型差异化失败):若独立同行评审研究复现 OwkinZero 的基准任务,并显示其表现等于或低于 GPT-4o 或 Claude 3.7,Owkin 的技术护城河将受到挑战。该风险目前无法从公开数据验证。监控:跟踪 OwkinZero ArXiv 引用、独立生物医学 AI 基准论文,以及 Owkin 2026 年发表管线。 论点破裂触发器 2(客户集中):若 Sanofi 在 5 年窗口(2026–2031)内发出不续约或实质降低 K Pro 参与度的通知,或 BMS 宣布战略性 AI 转向,离开 Owkin 平台,收入将非对称坍塌。监控:跟踪 Sanofi 和 BMS AI 合作公告中的竞争性承诺。 论点破裂触发器 3(资本):若 Owkin 以低于 $1B 估值完成融资,或到 2027 年 Q1 仍未宣布新融资(大致对应 $200M 现金加 $86M/年收入、扣除烧钱后接近关键水平的时间点),应重新评估入场价格和投资论点。监控:跟踪 Owkin 一级资本融资公告。 论点破裂触发器 4(监管):若 CNIL 或欧盟监管机构对 Owkin 健康数据处理启动正式调查,或 Owkin 未履行 GPAI 义务并收到 AI Office 执法通知,监管悬置将带来重大风险。监控:CNIL 执法登记;EU AI Office GPAI 合规数据库(可用后)。 关键开放尽调要求均需要管理层访问:(a)在公司赞助研究之外,对 OwkinZero 做独立基准验证;(b)按集中度拆分 K Pro 客户收入——尤其是 Sanofi 和 BMS 占总收入比例;(c)截至 2026 年 Q2 的当前现金头寸和烧钱速度;(d)GPAI Code of Practice 采用状态和 OwkinZero 训练数据透明度摘要;(e)董事会构成和继任框架;(f)最大医院节点的数据网络合同条款(期限、终止权)。[CR001, CR002, CR013, CR016, CR019, CR029]

缓释与否决标准表
风险可监控触发项阈值 / 事件行动含义
客户集中 — SanofiSanofi 公开 AI 战略或产品组合公告Sanofi 合作降级、竞争对手 AI 授权,或 2031 年前出现不续约信号立即复盘投资论点;测算收入影响;大概率维持观望持有
资本充足性主轮股权融资公告,或迟迟没有公告到 2027 年 Q1 仍无新一轮且现金低于 $120M,或 down-round 估值低于 $1B重定入场价格;修正盈利时间线;评估过桥融资风险
GPAI / GDPR 监管执法CNIL 或 EU AI Office 正式通知对 Owkin 发起正式调查或下达执法令结果明确前暂停投放资本;评估罚款相对资产负债表的敞口
模型差异化失败独立同行评审复现 OwkinZero 基准发表研究显示 OwkinZero 在生物推理任务上表现不高于 GPT-class LLM重新评估技术护城河;评估竞争平台证据基础
CEO 离职LinkedIn / 公开公告 CEO 变更Thomas Clozel 离任且未指定继任者暂停 60 天;评估药企合作连续性;评估新任管理层资历
医院数据访问丧失新闻稿或学术机构公告主要枢纽(>10% 训练数据)终止数据访问协议评估模型质量影响;要求提供剩余网络多样性指标

阈值是分析师建议的监控触发项,不是 Owkin 的合同契约或披露。所有阈值都具方向性; 实际投资决策需要用非公开财务数据做定量影响建模。

[CR001, CR013, CR014, CR015, CR016, CR019]
FR002: 风险传导图——Owkin 风险如何流向收入与估值

有向无环图展示主要风险事件如何沿 Owkin 的商业、财务和监管系统传导,并影响收入、投资者情绪和最终估值。

因果边由分析师根据风险因素之间的关系推断;未分配正式概率权重。边标签只是对传导机制的方向性描述。

[CR016, CR020, CR021, CR029, CR036, CR038]
FR003: 依赖关系图 —— 关键交易对手与基础设施

该有向图展示 Owkin 的关键上游依赖:投资者、客户、数据提供方、基础设施和监管方如何输入 K Pro 平台,以及价值交付如何向客户流出。

[CR006, CR008, CR020, CR021, CR022, CR023]

7.6 展示材料

Chapter 08

08估值

8.1 估值背景:最近已知估值、二级市场与资本头寸

Owkin 唯一公开确认的投后估值是 $1B,来自 2021 年 11 月 Sanofi 领投的 $180M Series B。2022 年 6 月 $80M Bristol-Myers Squibb 战略投资延展了 Series B,但没有披露修订后的投后估值。截至 2026 年 6 月——间隔约 4 年——Owkin 没有宣布新的一级股权融资,因此 $1B 是唯一锚定估值。二级市场聚合器 PremierAlts 引用的 2026 年估值约 $1.6B,但它并非基于已披露交易价格,应被视为方向性信号,而不是独立交易标记。 Owkin 报告 FY2024 收入约 $86M,同比增长 83%,净财务亏损 $5.6M(较 FY2023 的 $31.3M 显著改善)。公司 2024 年底持有约 $200M 现金,在方向性烧钱假设下提供约 2-3 年现金续航(470 名员工规模下每月 $3-9M)。截至 2025 年底,Owkin 目标一级融资 $200-300M;截至运行日期,尚未确认交割。2026 年 3 月 Waiv(前 Owkin Dx)剥离 单独融资 $33M,将诊断收入流从 Owkin 合并 P&L 中移出。Owkin FAQ 确认,所有投资人累计融资 $304.1M。$200M 现金头寸叠加 Sanofi 和 AstraZeneca 多年期 K Pro 合同,说明现金续航并非迫在眉睫;但没有确认的新一轮融资,估值仍锚定在 5 年前的价格,造成新一轮相对该锚点如何定价的根本不确定性。[CV002, CV003, CV004, CV005, CV006, CV007]

FV003: 按情景划分的估值与回报区间

截至 2026 年中,Owkin 在低位(熊市)、基准和高位(牛市)情景下的隐含企业价值区间,并列明各情景假设。

所有数字均为百万美元。区间是方向性估计,来自行业倍数分析、上市可比公司交易水平、M&A 参考(PathAI)和二级市场数据。不是正式 DCF,也不是风险调整后 NPV。

[CV005, CV009, CV022, CV023, CV029, CV030]

8.2 可比公司组:公开交易可比、M&A 退出与行业倍数

Owkin 的可比位置并不常见:其收入和模型更像 AI 原生药企合作,而不是传统 SaaS。因此,若不调整收入质量(里程碑波动 vs. 经常性 SaaS),单纯收入倍数分析并不充分。 最好的公开交易可比是 Recursion Pharmaceuticals(RXRX)和 Insilico Medicine(HKEX: 3696)。Recursion 2026 年 6 月市值 $1.71B,FY2025 收入 $74.7M(隐含 P/S 约 22.9x),但其现金运营支出约 $400M,同时持有 $753.9M 现金——资本结构与 Owkin 根本不同,且公开流通盘吸收了流动性折价。Insilico Medicine 2025 年 12 月完成香港 IPO,市值约 $2.1B,到 2026 年 6 月增至约 $2.52B,并报告 FY2024 收入 $85.83M(隐含 P/S 约 29x)——但其收入波动大且由里程碑驱动,2025 年 H1 下降 54%。两家公开可比公司都定价了大量临床管线可选性。 最相关的 M&A 可比是 PathAI(Roche,$750M 预付款 / $1.05B 总价,2026 年 5 月)和 Paige AI(Tempus,$81.25M,2025 年 8 月)。PathAI 累计融资 $355-660M,估计收入 $100-250M,隐含收购倍数为收入的 3-7x。Paige AI 以 $81.25M 退出,而此前累计融资 $220M+,构成显著折价退出,是任何与数字病理相邻平台的警示性可比。Finerva 2026 年 Q1 生物科技中位 EV/Revenue 4.4x 代表行业地板;有临床证明的 AI 平台可达到 6-10x。按 $1B 估值计算,Owkin 隐含倍数为 11.6x;按 $1.6B 二级估值计算为 18.6x,二者均高于行业中位数,但依赖前瞻叙事(K Pro SaaS 规模化),尚未由披露的收入质量指标支撑。[CV015, CV016, CV017, CV018, CV019, CV020]

可比估值表
可比对象类型收入 / 关键指标估值 / 退出价值隐含倍数与 Owkin 的相关性关键限制
Recursion Pharmaceuticals(RXRX)上市交易可比FY2025 收入 $74.7M;现金 $753.9M市值 $1.71B(2026 年 6 月)约 22.9x P/S trailing revenue最接近的上市可比:AI-native 药物发现平台,拥有药企合作(Sanofi、Roche);合作—里程碑收入模式相似上市公司已计入管线选择权;每年 burn $400M;Owkin 缺少临床管线,难以拿到完整临床溢价
Insilico Medicine(HKEX: 3696)上市交易可比FY2024 收入 $85.83M;2025 年 H1 收入 $27.5M(−54% YoY)市值约 $2.52B(2026 年 6 月);2025 年 12 月 IPO 估值约 $2.1BFY2024 收入约 29x P/S规模相近($86M vs $86M);同为 AI 药物发现;在同等收入水平都依赖药企合作拥有临床阶段药物(rentosertib Phase IIa);里程碑收入波动大;季节性不均;上市公司 vs. 私营公司折价
PathAI → RocheM&A 退出可比估计收入 $100-250M;总融资 $355-660M$750M upfront + $300M milestones = $1.05B total(2026 年 5 月)3-7x trailing revenue(收入估计不确定,区间较宽)数字病理 AI;与 Owkin 原 Dx 板块(现 Waiv)重叠;监管级算法;战略收购方(Roche)诊断收入模式不同于 Owkin 的药物发现重点;Waiv 已分拆;Owkin K Pro 的定位不同于 AISight
Paige AI → TempusM&A 退出可比已融资 $220M+;细节有限$81.25M 收购(2025 年 8 月)低于此前融资成本;深度 down-exit数字病理 AI 先行者(病理学首个 FDA-cleared AI);给 diagnostics-AI 估值提供警示性底部战略背景很不一样(Tempus AI 基础模型布局);down-exit 反映 2024-2025 年病理 AI 市场错位,不是 K Pro 的通用可比
BioTech & Genomics 行业中位数行业倍数基准41 家 GNOM ETF 成分公司集合的 EV/Revenue 中位数2026 年 Q1 EV/Revenue 中位数 4.4x(低于 2020 年 Q4 峰值 15.19x)4.4x 行业底部提供底部倍数锚点;Owkin 在 $1B 估值下为 11.6x,较行业中位数溢价 2.6x——AI 叙事和 K Pro SaaS 成分可以解释溢价,但披露的单位经济尚不足以支撑以收入为基础的倍数对盈利前 biotech 并不可靠;行业中位数包含许多亏损公司

收入数据来自已抓取的公司披露:Recursion FY2025 来自 ir.recursion.com Q4 earnings;Insilico 来自 Forbes IPO coverage(2025 年 12 月)和 companiesmarketcap.com;PathAI 来自 meddeviceguide.com 与 lw.com 交易条款;Paige 来自 hlth.com Tempus 公告。行业中位数来自 Finerva Q1 2026 报告。Owkin 收入来自 Maddyness(法国监管数据)。倍数计算使用最近披露或估算收入和最近已知估值标记。PathAI 收入区间较宽, 因其为私营公司;走向收购的路径可能包含 Roche 此前合作带来的未披露收入。

[CV015, CV016, CV017, CV018, CV019, CV020]
FV002: 估值敏感性:不同收入假设和倍数下的 EV

Owkin 企业价值对收入假设(FY2024 实际 $86M,对比分析师估计 $125M)和行业倍数情景(行业底部 4.4x、高溢价平台 11.6x、上市可比公司 22.9x)的敏感性。

所有 EV 数字均为百万美元。收入假设:$86M 来自 Maddyness 引述的法国监管数据(FY2024);$125M 来自 Growjo / Compworth 分析师估计(未确认)。倍数来自 Finerva 行业中位数(4.4x)、Owkin $1B 估值 / $86M 收入(11.6x)、PremierAlts 二级市场估值(18.6x)和 Recursion 可比公司(22.9x)。这些只是示意性的敏感性点;实际企业价值还取决于现金、优先权和少数股权等这里未纳入的因素。

[CV006, CV013, CV015, CV016, CV026, CV029]

8.3 牛、基准与熊情景

牛市情形($1.8B-$2.5B):2026 年下半年或 2027 年上半年,公司完成一轮新的 $200-300M 主融资,投后估值上调至 $2-2.5B。驱动因素包括 K Pro 经常性收入指标得到确认、AstraZeneca 交易扩容,以及大型药企战略买家按 PathAI/Roche 模板发起的主动上门 M&A 兴趣。到 2027 年,收入随 K Pro SaaS 放量和里程碑节奏增至 $130-160M。联邦数据护城河(180+ 家医院站点)让 Owkin 区别于集中式竞争对手。毛利率披露为 50%+, 支撑更高估值倍数。 基准情形($1.0B-$1.6B):2026-2027 年,公司在 $1B 附近完成新一轮融资(平轮),但财务披露不透明、 四年融资空窗压住估值。K Pro 带来温和收入增长;但 NRR 和毛利率未公开,按披露收入计,倍数压缩至 10-14x。Sanofi 和 BMS 托住收入,但客户集中风险限制新客户赢单带来的上行。现金头寸足够,无需困境融资。 熊市情形($600M-$900M):2027 年,公司以 $800-900M 水平下调估值融资,触发因素可能是 AI 生物科技 整体倍数压缩、Sanofi/AstraZeneca 的 K Pro 合同未能规模化落地,或药企合作伙伴预算意外收缩。里程碑节奏推迟, 收入停滞在 $100M 以下。Sifted 报道的内部执行问题加速关键人才流失。基本面转弱、交易枯竭后,PremierAlts $1.6B 的二级市场估值不再可得。[CV006, CV009, CV011, CV029, CV030, CV031]

牛 / 基准 / 熊情景摘要
情景关键假设隐含估值关键风险 / 概率信号
2026 年 H2 或 2027 年 H1 完成新一轮 $200-300M 融资,post-money $2-2.5B;K Pro ARR 披露为 $80-100M 经常性收入;毛利率 55%+;AstraZeneca 交易扩展$1.8B-$2.5B需要财务披露,也需要 K Pro 企业采用速度超出 Sanofi 和 AstraZeneca;概率信号:低(12 个月)、中(24 个月)
基准以 $1-1.5B 完成持平 / 小幅上行融资;K Pro 温和增长;不披露收入拆分;现金不需要紧急融资$1.0B-$1.5B最符合现有证据;不需要重大负面消息;概率信号:中等
融资价格低于 $1B 或拖到 2027 年之后;行业倍数压缩;合作伙伴集中风险兑现;管理层流失加速$600M-$900M需要宏观转向或 Owkin 具体执行失败;概率信号:低,但 Sifted 对执行文化的负面报道抬高了风险

估值具方向性,基于 Maddyness 报道 FY2024 收入约 $86M 的 EV/Revenue 分析、Finerva 行业中位数倍数、 Recursion 和 Insilico Market Cap 数据的上市可比公司倍数,以及 PathAI M&A 参考。概率信号为定性判断, 反映证据权重,不是统计估计。

[CV005, CV006, CV007, CV008, CV009, CV026]

8.4 投资论点、反论点与建议

投资论点靠四根支柱撑住:(1)Owkin 的联邦学习架构接入全球 180+ 家研究型医院,数据访问优势结构性难复制; (2)K Pro 与 Sanofi、AstraZeneca 签下多年期企业合同,引入经常性收入成分,收入质量明显高于纯里程碑付款; (3)公司在收入同比增长 83% 的同时,一年内把亏损从 $31.3M 收窄至 $5.6M,显示可持续性在改善; (4)二级市场估值约 ~$1.6B,且公司正在寻求 $200-300M 新一轮融资,使 Owkin 接近 BASI 叙事拐点: 生物医药 agentic AI 的叙事价值可向战略投资者拿到溢价。 反论点同样有力:(1)$1B 的主融资估值已是五年前的标记,且尚未被新资金检验——自该估值以来,行业倍数已大幅压缩 (峰谷从 15x 到 4.4x);(2)毛利率、NRR、详细合同经济性均未披露,投资者无法建模单元经济或回报曲线; (3)Thomas Clozel 关键人集中度高,Sifted 报道的 2025 年内部管理批评构成真实但未量化的执行风险; (4)Waiv 剥离拿走了一条收入流,也让公司进一步集中在 Sanofi 和 BMS 两个锚定客户的合作上,而这两家客户对 AI 投资的持续意愿可能受自身战略调整影响;(5)Recursion 即便现金储备充足,市值仍同比下跌 22.6%,说明当价值证明时间线令人失望时, 市场会向下重估整个板块。 建议为跟踪:K Pro 的商业牵引确实令人鼓舞,但财务不透明使买入结论站不住。若公司完成新一轮主融资并披露条款, 或投资者获得管理账目,才具备上调至买入的门槛条件。[CV004, CV008, CV009, CV011, CV012, CV014]

建议摘要
维度评估理由
建议跟踪收入在增长,但财务不透明,尚不足以下买入结论;需要确认新一轮融资或管理账
信心无审计财务、无披露毛利率或 NRR;依赖分析师估计和二手数据
风险评级合作伙伴集中、4 年融资空窗、关键人依赖、无公开财务
估值立场偏高$1B 上轮估值对应 $86M FY2024 收入的 11.6x EV/Rev;行业中位数 4.4x;二级市场 $1.6B 对应 18.6x
决策含义跟踪 → 买入门槛监控 K Pro 合同管线;索取管理账;等待新的主轮融资以评估重定价

估值倍数基于 FY2024 收入 $86M(Maddyness / 法国监管数据);行业中位数来自 Finerva Q1 2026 BioTech & Genomics 报告。建议对价格敏感;若确认以合理价格完成新一轮融资、披露毛利率 ≥50%, 且新增两个或更多具名 K Pro 企业客户,建议会转为买入。

[CV004, CV026, CV029, CV030]
投资论点 vs. 反论点
支柱正方论点反论点 / 什么会改变判断
数据护城河横跨 180+ 家研究医院的联邦学习,形成保护隐私的数据访问优势,中心化竞争者难以快速复制药企内部 AI 建设(Sanofi、AZ、Roche 每年各投入 $300-500M+ 到 AI)只要时间和资本足够,也可能复制联邦访问
收入质量K Pro 企业授权(Sanofi 5 年、AstraZeneca 3 年)引入经常性 SaaS 成分,优于纯里程碑收入K Pro 合同未披露财务条款;经常性收入与里程碑收入的完整拆分未知
单位经济轨迹净亏损同比改善 82%($31.3M → $5.6M),收入增长 83%;指向经营杠杆毛利率未披露;没有毛利率,亏损改善可能只是里程碑确认时点变化,而非结构性改善
平台估值叙事在 2026 年 AI 环境中,BASI / agentic AI 叙事能从战略投资者处拿到溢价倍数AI biotech 倍数已从 15x 峰值(2020 年 Q4)压缩到 4.4x 行业中位数(2026 年 Q1);叙事溢价可能迅速蒸发
资本位置2024 年末 $200M 现金限制了短期 distressed-raise 风险距上次主轮融资已有 4 年;burn rate 未披露;若市场情绪转向,新一轮价格可能持平或低于上一轮估值

反论点来自 Finerva 倍数数据、Sifted 负面报道(2025 年)和跨章节财务数据。 正反论点都有证据支撑;这不是泛泛的质量评价。

[CV006, CV007, CV008, CV012, CV014, CV026]
FV001: 建议逻辑流

从证据支柱到风险评估、估值纪律,再到 TRACK 建议的端到端链条。

该流程用于示意逻辑链条;节点标签是各章节证据的缩写摘要。

[CV004, CV006, CV012, CV026, CV029]
FV004: 投资 KPI 评分卡

面向 IC 的七维评分:市场、证明、护城河、经济性、风险、估值和证据质量。评分 1–10,10 为最佳。

评分是分析师基于本次报告收集证据作出的定性判断,并非专有评分模型。市场评分使用 Finerva 和 Axis Intelligence 数据。证明评分因缺少 K Pro 已发表结果而扣分。经济性评分因完全没有毛利率 / NRR 披露而大幅扣分。

[CV004, CV006, CV012, CV014, CV026, CV037]

8.5 退出准备度、止损标准与最终尽调问题

Owkin 的退出准备度最多只能算中等。三条退出路径有一定可行性:(1)IPO——2025-2026 年生物科技 IPO 市场回暖 (Eikon $381M、Insilico 港股 IPO)说明,具备临床证明和 50%+ 收入可见度的公司窗口已开;但 Owkin 缺少公开财务、 审计账目,以及独立于合作伙伴决策之外的临床阶段管线验证。除非加速财务披露,否则未来 12 个月 IPO 概率不高。 (2)战略 M&A——PathAI/Roche 模板(约 $100-250M 收入对应 $750M 首付款)是 Owkin 近期最可能的退出路径, 估值也与当前标记相符;潜在买家包括 Sanofi(已深度整合)、Roche(通过 PathAI 在病理 AI 建立布局)和 AstraZeneca;$1B 入场价格为低于上一轮标记的交易设置了障碍。(3)二级交易——PremierAlts 等平台曾在约 $1.6B 水平提供有限流动性。 止损标准包括:低于 $800M 的下调估值融资(显示基本面估值重置)、24 个月内未能完成待定的 $200-300M 融资 (资本充足性风险)、Thomas Clozel 离职且没有可信接班方案(关键人风险),或 Sanofi 或 BMS 合作范围实质收缩。 Waiv 少数股权在诊断业务上提供一点期权缓冲,但对 Owkin 母公司的估值论点并不关键。[CV003, CV005, CV009, CV022, CV023, CV034]

论点击穿与否决标准
触发项阈值 / 事件传导至投资论点行动含义
估值下调融资新主轮股权融资以低于 $800M post-money 完成证实市场重估 Owkin 的商业模式和 K Pro 牵引力;侵蚀现有投资者回报退出或降低敞口;全面重估投资论点
融资失败(未完成)尽管声明目标为 $200-300M,到 2027 年 Q2 仍未完成主轮融资信号要么是市场拒绝其估值,要么是未披露业务恶化;现金 runway 向 2027 年收缩要求管理账和月度财务;推动过桥或结构化债务
关键人离职Thomas Clozel 退出 CEO 角色且未宣布继任安排他是 Sanofi、BMS 和 AstraZeneca 合作关系的主要驱动者,也是 BASI 叙事所有者;估值折价会立即出现密切监控;将继任计划作为继续投资条件
合作伙伴集中事件Sanofi 或 BMS 实质性削减或终止 K Pro 合作范围锚定收入被移除,而该部分估计占已披露交易价值的大多数;收入坍缩到 K Pro 授权经常性收入复核资本充足性;建模剔除锚定合作伙伴后的收入,评估存活能力
行业倍数压缩BioTech / genomics EV/Revenue 中位数连续两个季度低于 3xOwkin 新一轮会较当前估值标记大幅折价;二级市场流动性枯竭将基准情景下调至 $700-900M;重看回报模型

阈值尽可能量化,但所有触发项都需要结合语境判断。行业倍数数据来自 Finerva Q1 2026 基准。 合作伙伴集中风险参考 Sifted 负面报道和前文章节客户分析。否决标准用于投资监控建议; 不能替代法律 / 合同权利。

[CV009, CV014, CV026, CV031, CV036]
最终尽调问题清单
主题缺失证据重要性负责人 / 尽调路径
审计财务过去 3 年审计 P&L、资产负债表和现金流量表没有审计数据,毛利率、NRR 和 burn rate 无法核验;无法承保回报模型Owkin 管理层 / CFO Graeme Bell;正式 data room 请求
K Pro 合同经济性Sanofi 和 AstraZeneca K Pro 授权的年合同价值(ACV)、合同期限拆分、续约选择权、里程碑安排K Pro 是核心估值驱动项;没有定价清晰度,SaaS 经常性收入和一次性收入混合比例未知Owkin CFO;任何潜在投资者都应收到 Sanofi / AZ 脱敏 term sheet
按产品线划分的毛利率K Pro SaaS 授权、里程碑付款和传统合作费之间的毛利率拆分毛利率决定业务是否具备可扩展的单位经济;目前未披露CFO data room;管理账中 IFRS 15 收入确认表
新一轮条款和优先权结构已确认的新主轮融资价格、优先权条款、反稀释条款、participating preferred vs. common已有 4 年的 cap table 中,清算优先权层级未定义;新一轮可能显著稀释现有持有人Owkin legal / Chief Legal Officer Anne-Charlotte Paut;现有投资者财团提供 cap table
Waiv 股权价值Owkin 在 Waiv 的持股比例、Waiv 完成 $33M 融资后的估值,以及 Owkin 持股的预期流动性路径为 Owkin 母公司估值提供选择权缓冲;Waiv 是独立资本化载体Waiv CEO Meriem Sefta / Owkin 企业发展团队
关键人和继任计划Thomas Clozel 与 Gilles Wainrib 的成文 CEO 继任计划和有约束力雇佣协议关键人集中是最大的非财务风险;Sifted 关于管理文化的负面报道放大了这一担忧Owkin Board;索取治理文件和关键人物保险

尽调优先级按重要性排序:财务披露(没有则阻塞)、K Pro 经济性(重要)、毛利率(重要)、 优先权结构(重要)、Waiv 持股(次要)、继任计划(重要)。所有事项都是 Series B+ 机构投资人的标准请求; 若 Owkin 推进其所称 $200-300M 融资,应能在 data room 提供。

[CV009, CV012, CV014, CV033]

8.6 附录

免责声明

本报告是基于公开证据的尽调快照,不构成投资建议。重要的财务、法律、技术和合同事实仍未公开;任何投资决策前,都应直接向管理层和一手文件核验。

证据索引

结论
编号陈述可信度来源
CO001 Owkin was founded in 2016 in Paris, France, by Thomas Clozel and Gilles Wainrib. SO001, SO002
CO002 Owkin is headquartered in Paris, France, with a significant US presence in New York and operates as a private French-American AI biotech company. SO001, SO002
CO003 Owkin is a private company (SAS, Société par Actions Simplifiée) that has not undergone an IPO. SO003
CO004 Owkin's stated mission is to achieve Biological Artificial Superintelligence (BASI) — AI capable of modelling and engineering biology beyond human cognition to deliver transformative therapies faster. SO001, SO005
CO005 Owkin's business model is multi-year enterprise R&D collaboration agreements with large pharmaceutical companies, generating revenue through AI platform embedding rather than pure transactional licensing. SO009, SO004
CO006 Thomas Clozel, MD, is Owkin's co-founder and CEO; he trained as a clinical oncologist and haematologist before co-founding Owkin. SO002, SO005
CO007 Gilles Wainrib, PhD, is Owkin's co-founder and Chief Scientific Officer; he is a professor of machine learning who contributes deep learning expertise to Owkin's scientific strategy. SO002, SO013
CO008 Thomas Clozel continues to serve as Owkin's CEO as of June 2026. SO004, SO005
CO009 Pascal Weinberger serves as Owkin's Deputy CEO. SO022
CO010 Graeme Bell serves as Owkin's CFO. SO022
CO011 Rodrigo Barnes serves as Owkin's CTO. SO022
CO012 Caroline Hoffmann serves as Owkin's Chief Medical Officer. SO022
CO013 Owkin's executive team additionally includes a Chief Marketing Officer, Chief People Officer, Chief Legal Officer, Chief Commercial Officer, and a Chief Corporate Development & Operations Strategy Officer. SO002, SO022
CO014 Thomas Clozel acknowledged in 2025 that Owkin had not handled HR matters perfectly during its AI pivot restructuring. SO010, SO011
CO015 The co-founding pairing of an oncologist (Clozel) and a machine-learning professor (Wainrib) reflects Owkin's thesis that biomedical AI requires integrated clinical and data science expertise. SO001, SO002
CO016 Owkin has raised approximately $304.1 million in total equity capital across its seed, Series A, and Series B rounds as of mid-2026, per the company's own FAQ. SO003, SO012
CO017 Owkin's post-money valuation was $1 billion following the November 2021 Series B round led by Sanofi, achieving unicorn status. SO012, SO019
CO018 Owkin raised a $2.1 million seed round in November 2016 to fund its initial federated AI research. SO003, SO024
CO019 Owkin raised approximately $72 million across multiple Series A tranches from 2018 through 2020, with GV, Bpifrance, Mubadala, MACSF, F-Prime, Cathay Innovation, Eight Roads, NJF Capital, and Otium Capital participating. SO003, SO024
CO020 Owkin raised $180 million in Series B funding in November 2021, led by Sanofi; the round also included an oncology AI discovery collaboration agreement with Sanofi. SO007, SO012
CO021 In June 2022, Bristol-Myers Squibb made an $80 million strategic investment in Owkin as part of a multi-year clinical data science collaboration, with up to $100 million in additional milestone payments. SO007, SO026
CO022 The total potential value of the BMS-Owkin collaboration, including upfront investment and milestone payments, is up to $180 million. SO012, SO026
CO023 Owkin's investors include GV (Google Ventures), Bpifrance Large Ventures, MACSF, Mubadala, F-Prime Capital, Cathay Innovation, Eight Roads Ventures, NJF Capital, Otium Capital, Plug and Play, Sanofi, and Bristol-Myers Squibb. SO003, SO019
CO024 Sanofi is Owkin's largest institutional investor, having led the $180 million Series B, and is simultaneously the company's deepest strategic R&D collaboration partner. SO012, SO004
CO025 Bristol-Myers Squibb invested $80 million in Owkin in June 2022 through a multi-year clinical data science collaboration focused initially on cardiovascular disease. SO007, SO012
CO026 GV (Google Ventures) participated in Owkin's Series A rounds beginning in 2018, providing an early-stage financial anchor. SO003, SO019
CO027 Owkin reported revenue of approximately $86 million in fiscal year 2024, representing 83% year-over-year growth from fiscal year 2023, as disclosed in French regulatory filings cited by Maddyness. SO011, SO020
CO028 Owkin posted a net financial loss of approximately $5.6 million in fiscal year 2024, a significant improvement from a $31.3 million loss in fiscal year 2023. SO011
CO029 Owkin held approximately $200 million in cash reserves as of end of fiscal year 2024, providing multi-year runway absent new funding. SO011
CO030 Owkin employs approximately 450–500 people as of June 2026, following a reduction of approximately 50 employees in the second half of 2025. SO010, SO011
CO031 Owkin was seeking a new funding round of $200–$300 million as of late 2025, with a stated goal of reaching profitability within five to six years. SO011
CO032 Owkin launched K Pro — described as the first agentic AI co-pilot for biopharma — in October 2025, built on Owkin Zero, a proprietary biological large language model. SO005, SO013
CO033 Owkin Zero is a fine-tuned biological LLM that performs biological reasoning; it underpins K Pro's advanced AI capabilities and integrates the MOSAIC multimodal oncology dataset. SO005, SO013
CO034 K Pro has accelerated Owkin's internal drug target identification by 70%, reducing the process from over 12 months to approximately 3 months. SO005, SO016
CO035 Owkin's federated learning technology enables AI model training across hospital patient datasets without centralizing sensitive patient data, preserving privacy while generating biomedical insights. SO009, SO001
CO036 Owkin has equipped approximately 18 leading academic research hospitals with its federated learning technology and participates in approximately 8 biopharma and research consortia. SO009
CO037 In March 2026, Owkin spun off its diagnostics division (Owkin Dx) into a new independent company named Waiv, which received $33 million in initial financing led by OTB Ventures and Alpha Intelligence Capital. SO006, SO014
CO038 In June 2026, Owkin and Sanofi announced an expanded multi-year collaboration to develop purpose-built agentic AI agents using the K Pro platform, embedded across Sanofi's full drug development pipeline. SO004, SO017
CO039 The BMS-Owkin collaboration, initiated in June 2022, focuses on using AI for clinical trial design, endpoint optimization, and covariate adjustment, initially in cardiovascular disease with potential to extend to oncology. SO007, SO012
CO040 Owkin has research and commercial partnerships with AstraZeneca, Amgen, Roche, Ipsen, Johnson & Johnson, and additional biopharma and academic institutions beyond the flagship Sanofi and BMS deals. SO009, SO005
CO041 Owkin's MSIntuit CRC received CE-IVD certification, making it the first CE-marked AI diagnostic for detecting microsatellite instability (MSI) in colorectal cancer from routine digital pathology slides. SO008, SO015
CO042 Waiv (the Owkin-spun-off diagnostics company) achieved dual CE marking under the EU IVDR regulation in 2026 for its AI precision testing tools covering breast cancer (RlapsRisk BC) and colorectal cancer. SO018, SO014
CO043 In September 2025, Owkin announced a plan to cut approximately 50 employees — representing roughly 15% of its workforce — as part of a strategic refocus on pure agentic AI activities. SO010, SO011
CO044 Owkin employees publicly criticized "huge managerial issues" in the context of the 2025 restructuring, as reported by Sifted; CEO Thomas Clozel acknowledged the company had not handled HR execution perfectly. SO010
CO045 Owkin's strategic rationale for spinning off Waiv in March 2026 was to simplify its product portfolio, reduce costs, and concentrate entirely on agentic AI for drug discovery and development. SO006, SO011
CM001 Owkin operates at the intersection of AI in drug discovery, federated learning infrastructure, and precision-oncology diagnostics rather than in a single market. SM001, SM002
CM002 Owkin's federated learning keeps raw data locked on hospital servers and moves only predictive models between sites, preserving privacy. SM002
CM003 Owkin is building an autonomous AI Scientist (K Pro) intended to automate biopharma R&D and progress toward a "Biological Artificial SuperIntelligence." SM001
CM004 Owkin is a French-American company founded in 2016 by Thomas Clozel and Gilles Wainrib, with products MSIntuit CRC and RlapsRisk BC and around 350 employees as of 2023. SM006
CM005 Analysts segment AI in drug discovery by component (software, services), technology (machine learning), application (target identification, de novo design, optimization), therapeutic area, and end user. SM007
CM006 MarketsandMarkets sizes the broad Drug Discovery Technologies market at USD 30.58 billion in 2025 growing to USD 51.51 billion by 2030 at an 11.0% CAGR, most of which is instruments and reagents outside Owkin's AI layer. SM020
CM007 The status-quo substitutes Owkin displaces are manual pathology review, centralized data pooling, and traditional high-throughput screening, which form the excluded baseline rather than captured revenue. SM002, SM022
CM008 Mordor values the AI-in-drug-discovery market at USD 2.58 billion in 2025, USD 3.25 billion in 2026, and USD 10.29 billion by 2031 at a 25.94% CAGR. SM008
CM009 Fortune Business Insights values the AI-in-drug-discovery market at USD 4.46 billion in 2025, USD 5.00 billion in 2026, and USD 12.56 billion by 2034 at only a 12.20% CAGR, contradicting Mordor's growth rate. SM009
CM010 MarketsandMarkets sizes the adjacent AI-in-precision-medicine market at USD 0.78 billion in 2024 growing 30.7% to USD 3.92 billion by 2030. SM020
CM011 Because Mordor and Fortune disagree, the credible AI-in-drug-discovery market for 2026 is a range of roughly USD 3.25-5.0 billion rather than a single point. SM008, SM009
CM012 Precedence sizes the federated-learning market at USD 1.59 billion in 2026 on a 30.5% CAGR, with healthcare and life sciences about a quarter of it and North America dominant. SM025
CM013 Mordor sizes the precision-oncology market at USD 127.68 billion in 2026 at a 9.53% CAGR, with diagnostics growing faster than therapeutics at 10.06%. SM011
CM014 Precedence sizes the precision-oncology market at USD 146.24 billion in 2026 rising to USD 338.89 billion by 2035 at an 11.33% CAGR. SM012
CM015 Fortune sizes the precision-oncology market at USD 133.06 billion in 2026 rising to USD 317.16 billion by 2034 at an 11.47% CAGR, with North America near 39%. SM013
CM016 Three independent analysts cluster the 2026 precision-oncology market at roughly USD 127-146 billion, a far larger pool than Owkin's serviceable slice. SM011, SM012, SM013
CM017 Mordor cites a roughly USD 2.6 billion average cost to commercialize a single molecule as a structural driver steering budget toward AI discovery platforms. SM008
CM018 Mordor reports that software led AI in drug discovery with a 62.43% revenue share in 2025 and machine learning held a 46.54% technology share. SM008
CM019 MarketsandMarkets sizes the drug-discovery-informatics adjacency at USD 2.2 billion in 2020 growing to USD 3.5 billion by 2025. SM020
CM020 Owkin's serviceable market is best framed as oncology-focused pharma R&D partnerships plus European MSI diagnostics, a narrow slice of the precision-oncology TAM that the fetched record cannot quantify. SM015, SM011
CM021 The primary economic buyer is pharmaceutical R&D, evidenced by Sanofi's USD 180 million investment plus a three-year USD 90 million collaboration across four cancers. SM003, SM004
CM022 Sanofi's December 2021 investment of USD 180 million made Owkin a unicorn valued above USD 1 billion. SM003, SM006
CM023 Owkin has raised over USD 255 million in total funding. SM006
CM024 Evaluate data put global pharmaceutical R&D spending near USD 288 billion in 2024, potentially reaching USD 340 billion by 2030, defining the buyer budget pool. SM016
CM025 EFPIA reports European pharmaceutical R&D investment of roughly EUR 50 billion in 2023 with about 900,000 direct employees. SM015
CM026 A Nature Reviews Drug Discovery analysis pegged global biopharma R&D investment at USD 276 billion in 2021, more than double commonly cited estimates. SM017
CM027 Healthcare providers and pharmaceutical companies represent about a quarter of the federated-learning market by end user, with North America holding about 40% in 2025. SM025
CM028 A secondary buyer channel runs through pathology labs and health systems for productized diagnostics, with adoption triggered by pathologist shortages and immunotherapy matching. SM014, SM002
CM029 Precision-oncology incumbents such as Roche, Guardant Health, Tempus, Caris, and Illumina frame the competitive payer-facing landscape Owkin must eventually navigate. SM013
CM030 Privacy regulation such as GDPR and HIPAA is cited as the leading driver of federated learning because it lets models train without transferring raw data. SM025
CM031 Owkin's MSIntuit CRC is the only CE-marked AI diagnostic for MSI testing in colorectal cancer from H&E slides, validated at 96% sensitivity in a blind Nature Communications study. SM014
CM032 AI and biology models are maturing, illustrated by large-scale pooled perturbation screens with single-cell and spatial readouts, strengthening the technical case for AI discovery. SM022
CM033 The FDA continues to clear AI/ML-enabled medical devices through early 2026, evidencing an active regulatory pathway that de-risks AI diagnostics commercialization. SM021
CM034 The EU AI Act and the EMA reflection paper impose a risk-based regime in which clinical AI is high-risk and CE-marked, while AI used solely for medicines R&D is exempt, creating dual-compliance complexity. SM018, SM019
CM035 Clinical validation is slow and capital-intensive; MSIntuit required a blind Nature Communications study before clearance and its reimbursement path remains unproven. SM014
CM036 Owkin cut staff and pivoted toward an AI-agents business model, signaling strategy and execution uncertainty. SM005
CM037 Federated learning faces technical constraints from non-IID data heterogeneity and high per-site infrastructure cost that raise deployment economics. SM025
CM038 Evaluate's World Preview 2026 highlights patent expirations, GLP-1 dominance, pricing pressure, and rising Chinese competition as forces reshaping the pharma buyer backdrop. SM024
CM039 Global pharmaceutical R&D spending is corroborated across sources at roughly USD 288 billion in 2024 (Evaluate) and EUR 50 billion in Europe in 2023 (EFPIA), confirming a large and growing buyer budget. SM016, SM015
CM040 The cancer burden is structural, with the WHO recording nearly 10 million cancer deaths in 2022 and colorectal cancer alone causing about one million deaths from roughly two million cases in 2020. SM023, SM014
CM041 No fetched public source discloses Owkin's revenue, run-rate, customer count, or serviceable obtainable market, so all sizing here is top-down TAM. SM006, SM003
CM042 Reputable analysts conflict materially on AI-in-drug-discovery growth, with Mordor's 25.94% CAGR more than double Fortune's 12.20%, swinging the out-year market by billions. SM008, SM009
CM043 Reimbursement and payer coverage for MSIntuit and Owkin diagnostics are not confirmed in any fetched source, leaving the diagnostics revenue channel unproven. SM014
CP001 Owkin launched K Pro, its agentic AI platform for biopharma decision-making, described as the first agentic AI co-pilot for biopharma powered by biological reasoning models. SP001, SP002
CP002 K Pro integrates Owkin Zero, a fine-tuned biological LLM that the company states performs biological reasoning with higher performance than other leading LLMs, as described in an arXiv paper (2508.16315). SP001, SP003
CP003 K Pro has accelerated Owkin's internal drug target identification by 70%, reducing the timeline from over 12 months to approximately 3 months, per Owkin's official product announcement. SP001, SP002
CP004 AstraZeneca signed a three-year license of Owkin's K Pro platform and commissioned custom AI agents for competitive intelligence, building on prior work developing the AI-based gBRCA pre-screen solution for breast cancer. SP002, SP001
CP005 Owkin and Sanofi announced a multi-year expansion of their partnership in June 2026 to co-develop biopharma agents backed by a five-year K Pro license; Sanofi's chief digital officer confirmed the intent to deploy purpose-built agentic systems. SP003, SP023
CP006 MOSAIC is described by Owkin as the world's largest spatial multiomics oncology dataset, stated to be 100 times larger than any comparable alternative dataset. SP004, SP005
CP007 MOSAIC covers cancer types including NSCLC, ovarian, bladder, mesothelioma, glioblastoma, breast cancer, and DLBCL, with six integrated data modalities including spatial transcriptomics, single-cell RNA-seq, bulk RNA-seq, whole exome sequencing, clinical data, and H&E pathology images. SP004, SP005
CP008 Owkin was co-founded in 2016 by Thomas Clozel MD (current CEO) and Gilles Wainrib, with Clozel bringing a background as a former assistant professor in clinical onco-hematology. SP004, SP001
CP009 Owkin has raised over $354 million in total funding, including from Sanofi, BMS, Fidelity, GV, and BPI, reaching unicorn status through a $180 million Series B in 2021 that included Sanofi as a lead investor. SP012, SP013
CP010 Roche announced a definitive merger agreement to acquire PathAI on 7 May 2026 for a purchase price of $750 million upfront and additional milestone payments of up to $300 million, with the transaction expected to close in the second half of 2026. SP006, SP012
CP011 PathAI's AISight IMS is a cloud-native, open platform enterprise workflow solution used by leading laboratories and research centers globally for digital pathology workflows and AI applications in pharma clinical trial support. SP018, SP006
CP012 Paige AI was acquired by Tempus in August 2025 for approximately $81.2 million; prior to acquisition Paige had raised $241 million over eight funding rounds and was the leading FDA-cleared pathology AI developer for cancer diagnostics. SP012, SP014
CP013 Paige AI developed FDA-cleared AI applications for prostate, breast, and gastrointestinal cancer pathology, as well as a multi-cancer detection foundation model, prior to its acquisition by Tempus. SP017, SP012
CP014 Insilico Medicine raised $110 million in a Series E round in 2025, completed an IPO on the Hong Kong exchange in December 2025, and reported $393 million in cash and $56.2 million in revenue for FY2025. SP012, SP013
CP015 Insilico Medicine's rentosertib demonstrated positive Phase IIa results for idiopathic pulmonary fibrosis—a 98.4 mL forced vital capacity improvement at the 60 mg dose versus a 20.3 mL decline in the placebo group—published in Nature Medicine in June 2025, marking the first AI-designed drug to show clinical efficacy. SP012, SP013
CP016 Insilico Medicine signed a $2.75 billion R&D partnership with Eli Lilly in March 2026, including $115 million upfront and up to $2.63 billion in potential milestone payments, covering cardiometabolic and other therapeutic areas. SP012, SP013
CP017 Insilico Medicine's Pharma.AI platform consists of three modules: Biology42 for AI-driven target discovery, Chemistry42 for generative chemistry and molecular design, and Medicine42 for clinical program design and trial optimization. SP010, SP011
CP018 Recursion Pharmaceuticals and Exscientia completed a $688 million all-stock merger in early 2025, creating an end-to-end AI drug development platform combining Recursion's phenomics-driven biology with Exscientia's precision molecular design. SP016, SP024
CP019 The combined Recursion entity reported approximately $754–850 million in cash and cash equivalents in 2026, providing financial runway estimated to extend to early 2028. SP025
CP020 Recursion's partnerships with Roche/Genentech, Sanofi, Bayer, and NVIDIA collectively carry potential milestone payments exceeding $20 billion, with approximately $200 million expected in the next two years. SP016, SP025
CP021 Isomorphic Labs raised $2.1 billion in a Series B round in May 2026 led by Thrive Capital, with participation from Alphabet, GV, MGX, Temasek, CapitalG, and the UK Sovereign AI Fund, bringing total cumulative funding above $2.6 billion. SP007, SP013
CP022 Isomorphic Labs' IsoDDE platform is derived from AlphaFold-based AI and is applied to drug design programs across oncology, immunology, and cardiovascular disease. SP007, SP013
CP023 Isomorphic Labs has active drug discovery partnerships with Novartis, Eli Lilly, and Johnson & Johnson, and expects its first AI-designed drug candidates to enter human clinical trials by the end of 2026. SP007, SP013
CP024 BenevolentAI underwent multiple rounds of workforce reductions including an initial cut of approximately 180 staff and a subsequent 30% further reduction, closed its US office, and shelved its SaaS knowledge exploration tools to conserve cash. SP015, SP012
CP025 BenevolentAI's share price fell over 90% from its 2022 Amsterdam listing, reducing market capitalization to approximately £85 million, despite the company having once been valued at over £1 billion. SP015, SP012
CP026 BenevolentAI's lead clinical asset BEN-8744, a PDE10 inhibitor for ulcerative colitis, was identified through the company's AI knowledge graph and is in Phase II clinical trials, backed by a $594 million Merck KGaA partnership. SP012, SP015
CP027 Schrödinger's physics-based computational platform contributed to the discovery of zasocitinib, a TYK2 inhibitor originally discovered via Nimbus Therapeutics using the Schrödinger platform, now in Phase III clinical trials through a Takeda partnership for inflammatory and autoimmune diseases. SP012, SP013
CP028 Proscia launched the fifth generation of its Concentriq digital pathology platform with an AI-native architecture targeting drug discovery through clinical case review workflows. SP019, SP012
CP029 Foundation Medicine, a Roche subsidiary, offers FoundationOne CDx—an FDA-approved in vitro companion diagnostic for advanced solid tumors—analyzing 324 genes and genomic signatures, representing the incumbent high-throughput genomic diagnostic standard. SP022, SP006
CP030 The global AI drug discovery market was valued at approximately $1.94 billion in 2025 and is projected to reach $2.6–2.8 billion in 2026, growing at a 27% CAGR toward an estimated $16.49 billion by 2034. SP013, SP014
CP031 Over 173 AI-designed drug programs are currently in clinical trials globally as of 2026, compared to approximately 67 in 2023 and just three in 2016, reflecting rapid pipeline expansion across the sector. SP013, SP014
CP032 Phase I clinical success rates for AI-discovered drugs are reported at 80–90%, roughly double the historical average of 40–65% for traditionally designed drugs, based on analyst reporting on 2025-2026 pipeline data. SP013, SP014
CP033 Phase II success rates for AI-backed drug candidates average approximately 40%, comparable to the traditional rate of 29–40%, indicating that early-stage Phase I advantages in target selection do not yet compound into material Phase II improvements. SP014
CP034 Owkin's federated learning architecture trains AI models directly within hospital data environments so that raw patient data never leaves institutional control, enabling participation from privacy-sensitive institutions including European GDPR-constrained academic medical centers. SP001, SP004
CP035 Owkin's federated learning platform is deployed across a multi-institutional network of hospital and research institution hubs spanning Europe and North America, providing access to real-world patient data in jurisdictions where centralized data transfer is restricted. SP001, SP004
CP036 Owkin spun out its diagnostics division as Waiv in 2026, with the BRCAura breast cancer prediction solution continuing development within Waiv, while Owkin itself concentrates on K Pro and its Biological Artificial Superintelligence mission. SP002, SP003
CP037 Owkin's BRCAura histology model was found to rule out approximately 40% of patients unlikely to carry gBRCA1/2 mutations with a sensitivity of 93%, as published in a poster at ESMO 2025. SP002, SP001
CP038 Owkin's original Sanofi partnership, established in 2021 through a €90 million strategic deal, was focused on oncology target identification and patient subgrouping, and was subsequently expanded to include drug positioning for Sanofi's immunology pipeline. SP003, SP002
CP039 The combined Recursion-Exscientia platform aggregates over 60 petabytes of biological and chemical data spanning phenomics, transcriptomics, proteomics, ADME, and de-identified patient data, trained on BioHive-2, a pharma-industry supercomputer built with NVIDIA. SP008, SP009
CP040 Switching away from Owkin's federated clinical AI requires data re-integration, regulatory revalidation of AI-assisted decisions in clinical trial contexts, and retraining of scientific teams across multi-center workflows, creating material switching costs for established clients estimated at 6–18 months of migration friction. SP013, SP014
CP041 No AI-designed drug has received regulatory approval from the FDA or EMA as of June 2026, despite 173 programs in clinical development, representing a structural risk that undermines revenue-model confidence for the entire AI drug discovery sector. SP014
CP042 K Pro built an IND-ready asset positioning strategy in hours instead of months of traditional analysis, and produced a specialist scientific review in one day that was subsequently published in a peer-reviewed journal, per Owkin's official product launch. SP001, SP002
CP043 Multiple AI drug discovery companies have restructured post-2023, with Recursion shelving three pipeline programs following the Exscientia merger and several competitors cutting 20–30% of their workforces, indicating systematic sector-wide attrition despite elevated funding levels. SP014, SP015
CP044 AI drug discovery partnership announcements in 2025 exceeded $15 billion in aggregate deal value, but actual upfront payments averaged approximately 2% of headline totals, revealing a significant gap between announced deal sizes and committed capital actually paid to AI platform vendors. SP013, SP014
CP045 Owkin's competitive set as of mid-2026 spans AI drug discovery platforms (Insilico Medicine, Recursion-Exscientia, Isomorphic Labs, BenevolentAI, Schrödinger), pathology AI vendors being absorbed by diagnostics incumbents (PathAI into Roche, Paige into Tempus), and internal pharma AI programs as the dominant status-quo substitute. SP012, SP013
CP046 Pharma companies including Pfizer, Roche, Novartis, AstraZeneca, and Sanofi each invest hundreds of millions of dollars annually in internal AI R&D infrastructure, making internal build the most commonly available alternative to contracting a third-party AI drug discovery platform. SP013
CP047 Key buying criteria for AI drug discovery platforms among pharmaceutical companies include clinical validation track record, proprietary data scale and modality depth, privacy architecture, regulatory pathway clarity, and enterprise integration capability. SP013
CI001 Owkin's revenue is generated through three primary streams: pharma collaboration licensing (combining upfront equity investments with collaboration fees), milestone-triggered R&D payments, and multi-year recurring SaaS licensing of the K Pro agentic AI platform. SI001, SI012, SI015
CI002 Sanofi invested $180 million in equity into Owkin's Series B round in November 2021, valuing Owkin at over $1 billion and making it a unicorn. SI001, SI002
CI003 The November 2021 Sanofi deal also included approximately $90 million in oncology collaboration payments scheduled over three years, targeting four cancer types including non-small cell lung cancer, triple-negative breast cancer, mesothelioma, and multiple myeloma. SI001, SI022
CI004 Bristol-Myers Squibb invested $80 million upfront in Owkin's Series B extension in June 2022, with the deal including up to $100 million in additional potential milestone payments for cardiovascular-trial AI applications. SI004, SI002
CI005 In 2026, Sanofi expanded its partnership with Owkin through a five-year enterprise license for the K Pro agentic AI platform, transitioning the relationship from project-based oncology collaboration to embedded AI-agent licensing across Sanofi's R&D workflows. SI014, SI006
CI006 AstraZeneca signed a three-year licensing agreement for Owkin's K Pro agentic AI platform in May 2026, adding a third major pharma partner to Owkin's K Pro client base alongside Sanofi and BMS. SI007, SI013
CI007 Owkin does not publish a price list for K Pro or any other product; all enterprise licensing agreements are custom-negotiated with individual pharma clients at confidential terms. SI012, SI006, SI007
CI008 Owkin's pharma collaboration contracts typically combine three components: an upfront equity investment or collaboration fee at deal signing, a recurring multi-year license payment for platform access, and milestone-triggered payments tied to verifiable research achievements. SI004, SI001, SI006
CI009 Revenue recognition for Owkin's milestone-based pharma partnerships follows IFRS 15 / ASC 606 principles: upfront fees are recognised when distinct performance obligations are delivered, recurring license revenue is recognised ratably over the contract term, and milestone payments are recognised when the underlying research achievement is verified and the right to payment is established. SI008, SI009
CI010 Owkin's estimated annual revenue is approximately $125 million based on analyst workforce- intelligence platforms (Growjo, Compworth); this is an analyst estimate and Owkin has not publicly confirmed any financial metric. SI009, SI010
CI011 Owkin's hospital network of 180+ institutions serves as a federated data-acquisition asset for AI model training; it does not generate direct revenue for Owkin and requires ongoing partnership maintenance and integration costs. SI021, SI015, SI025
CI012 The Sanofi partnership expanded over five years from oncology collaboration (2021), to immunology AI drug positioning (2024), to five-year K Pro platform licensing (2026), demonstrating significant account-level expansion within a single anchor client. SI014, SI001, SI006
CI013 The cumulative value of the Sanofi relationship across equity ($180M), oncology collaboration ($90M), and the 2026 K Pro licence extension (undisclosed) is estimated at over $270 million; the financial terms of the 2026 K Pro extension have not been disclosed. SI001, SI022, SI006
CI014 Owkin's gross margin is entirely undisclosed; the company has not published financial statements and no gross-margin data appears in any reviewed public source. SI009, SI015, SI025
CI015 Owkin's cost structure encompasses AI research and model development, distributed cloud computing for federated learning nodes, hospital-network integration and maintenance for 180+ sites, regulatory and legal compliance for multi-jurisdiction patient-data frameworks, and enterprise sales and business development. SI021, SI015, SI008
CI016 Owkin's federated learning architecture distributes model training compute across hospital nodes rather than centralising data, which may structurally elevate infrastructure and coordination costs relative to centralised AI competitors that can operate on single-cloud data lakes. SI021, SI009
CI017 At approximately 470 employees (analyst workforce estimate, December 2025), Owkin's implied revenue-per-employee ratio is roughly $217K per year, derived from the $125M revenue estimate; this is typical for a pre-profit AI services company at the validation-to-scale inflection point. SI009, SI011
CI018 Owkin has not publicly disclosed its customer acquisition cost (CAC), average enterprise sales cycle length, or payback period for any pharma client segment. SI015, SI009
CI019 Owkin has not publicly disclosed its net revenue retention (NRR) for any product or client segment; no customer cohort data or renewal metrics have been reported. SI015, SI025
CI020 Owkin has not publicly disclosed its monthly burn rate or current cash runway, and no analyst or investor source has published a verified burn figure. SI009, SI015
CI021 Owkin has raised approximately $321 million in total financing from inception through early 2026, across a seed round, multiple Series A tranches, a $180M Series B (Sanofi, 2021), and an $80M Series B extension (BMS, 2022), plus small EU grants. SI008, SI019, SI020
CI022 Owkin achieved unicorn status (valuation exceeding $1 billion) upon closing the $180M Sanofi Series B in November 2021; this $1 billion post-money valuation is the last formally established equity-round valuation. SI001, SI002
CI023 Premier Alts estimated an implied secondary-market valuation of approximately $1.6 billion for Owkin in 2026, above the $1 billion post-money from the 2021 Series B, reflecting secondary- market optimism but no confirmed new primary-equity pricing event. SI010, SI008
CI024 Key Owkin investors include Sanofi, Bristol-Myers Squibb, GV (Google Ventures), Bpifrance, MACSF, Mubadala, Cathay Innovation, Eight Roads Ventures, F-Prime Capital, NJF Capital, Frst Capital, and Otium Capital. SI008, SI020, SI023
CI025 In March 2026, Owkin spun out its diagnostics division — formerly branded Owkin Dx — as a standalone company called Waiv, which simultaneously raised $33 million from OTB Ventures and Alpha Intelligence Capital. SI005, SI018
CI026 Owkin retains a minority equity stake in Waiv following the March 2026 spinout, maintaining an ongoing financial interest in the diagnostics business without consolidating its operations into Owkin's P&L. SI005, SI018
CI027 The Waiv $33 million Series A financing is on Waiv's own capital table and does not augment Owkin's cash position; it is a spinout financing event, not a primary Owkin funding event. SI005, SI018
CI028 Owkin's last primary equity capital raise at the parent-company level was the $80 million BMS Series B extension in June 2022, creating an approximately four-year gap to the runDate of June 2026 without a disclosed primary equity round for Owkin itself. SI004, SI008, SI020
CI029 Owkin's Series A rounds, conducted between 2018 and 2020, raised approximately $59 million in aggregate from investors including GV, Bpifrance, MACSF, Mubadala, Cathay Innovation, Eight Roads Ventures, F-Prime Capital, NJF Capital, and Frst Capital. SI008, SI020, SI023
CI030 Owkin has not publicly disclosed gross margin, operating margin, EBITDA, burn rate, cash position, ARR breakdown, NRR, CAC, or payback period for any period or product segment. SI015, SI009, SI025
CI031 Owkin's customer-revenue concentration is materially high: Sanofi and BMS are the two largest disclosed partners and together account for the overwhelming majority of cumulative disclosed partnership value; a strategy change or in-house AI build by either partner would be financially material. SI003, SI004, SI001
CI032 Sifted reported in 2025 that Owkin staff publicly criticised "huge managerial issues" at the company during its pivot to agentic AI, citing poor organisational transparency, inadequately communicated job cuts, and unclear strategy communication as persistent problems. SI003
CI033 Pharma drug-discovery R&D cycles typically span five to ten years from target identification to regulatory approval, creating inherent revenue-timing risk and milestone-payment lumpiness for Owkin's collaboration revenue. SI008, SI009
CI034 Owkin's federated learning architecture is not uniquely protected by an IP moat; Big Tech incumbents (e.g., Google DeepMind, Microsoft) and emerging AI biotech firms are actively developing privacy-preserving AI capabilities, creating competitive pressure on Owkin's core technological differentiator. SI003, SI021, SI009
CI035 Owkin has not disclosed any IPO timeline, registered any securities with a public regulator, announced a secondary liquidity programme, or indicated any near-term path to public markets. SI015, SI025
CI036 The five-year Sanofi K Pro enterprise license signed in 2026 provides Owkin with multi-year recurring revenue visibility from its largest partner, which is an unusual and positive durability signal for a pre-profit AI biotech company. SI006, SI014, SI016
CI037 The AstraZeneca K Pro licensing deal (May 2026) expands Owkin's active enterprise platform clients to at least three top-ten global pharma companies, modestly reducing the two-partner customer-concentration risk. SI007, SI013
CI038 Without disclosed gross margin, burn rate, cash position, or ARR breakdown, it is not possible for external investors to model Owkin's return-to-capital timeline, profitability path, or next equity-round trigger from public data alone. SI009, SI010, SI015
CI039 The approximately four-year interval between Owkin's last primary equity raise (June 2022) and the runDate (June 2026) suggests either sufficient partnership-revenue cash generation to fund operations, a managed drawdown of the 2022 raise, or proximity to a capital event — none of which can be confirmed from public data. SI004, SI008
CI040 The Waiv diagnostics spinout reduces near-term revenue diversification for Owkin's consolidated entity but allows the diagnostics business to raise capital independently and pursue focused growth, potentially creating future equity value for Owkin as a minority shareholder. SI005, SI018
CE001 Owkin launched K Pro in October 2025 as an agentic AI co-pilot for biopharma, queried in natural language to support decisions across the drug-development pipeline. SE004, SE020
CE002 K Pro uses a single intelligent orchestrator that dynamically selects and combines specialized AI skills, grounded in multimodal patient data, to return publication-ready answers. SE001, SE004
CE003 K Pro integrates OwkinZero, a fine-tuned biological LLM, as the reasoning engine at the core of the platform. SE004, SE003
CE004 Owkin reports K Pro accelerated internal drug-target identification by 70%, from more than 12 months to about 3 months. SE004
CE005 K Pro FREE, formerly branded K Navigator, is a free agentic AI tool for biologists that assists with scientific writing, literature review, hypothesis testing and data visualization. SE002
CE006 K Pro FREE ensures results are credible and traceable back to trusted sources such as PubMed and biomedical databases. SE002
CE007 OwkinZero is a domain-specialized biological reasoning model trained on more than 300,000 verified question-and-answer pairs across eight biomedical tasks. SE003, SE006
CE008 OwkinZero models were produced by post-training open-source 8B–32B LLMs with Reinforcement Learning from Verifiable Rewards. SE006
CE009 Owkin's arXiv paper reports that specialized 8B–32B OwkinZero models outperform larger state-of-the-art commercial LLMs on its biological benchmarks and show cross-task generalization. SE006
CE010 In January 2026 Owkin announced a collaboration with NVIDIA using the Nemotron open-model family and the NVIDIA NeMo RL framework to scale the next generation of OwkinZero. SE005
CE011 Owkin's federated-learning approach keeps patient data locked on hospital servers while only the predictive models travel between servers. SE017, SE016
CE012 Owkin published a peer-reviewed Nature Medicine study applying federated learning to predict histological response to neoadjuvant chemotherapy in triple-negative breast cancer. SE008
CE013 Substra, the open-source federated-learning framework created by Owkin, is now hosted by the Linux Foundation for AI & Data and logs all ML operations in an auditable read-only database. SE016, SE013
CE014 Owkin maintains FLamby, an open-source cross-silo federated-learning benchmark spanning seven real-world healthcare datasets. SE014, SE013
CE015 Phikon is an 85.8M-parameter Vision Transformer Base trained with iBOT on 40 million pan-cancer tiles from TCGA and released under an Owkin non-commercial license. SE011, SE009
CE016 Phikon-v2 is a Vision Transformer Large pre-trained with DINOv2 on PANCAN-XL, a 450M-image histology corpus sampled from 60,000 whole-slide images from public collections (TCGA, CPTAC, GTEx). SE012
CE017 The ENLIGHT–DeepPT framework imputes genome-wide tumor expression from H&E slides and predicted responders across five independent cohorts with an odds ratio of 2.28. SE007
CE018 ENLIGHT–DeepPT reported a 39.5% increased response rate among predicted responders across four treatments and six cancer types in the Nature Cancer study. SE007
CE019 Owkin markets K Pro as secure, interpretable and interoperable, and lets clients securely upload proprietary datasets to be reasoned over alongside Owkin's data. SE001
CE020 The K Pro page describes Owkin's patient-data network as the world-largest at 104 hospitals, feeding the MOSAIC spatial and multiomics dataset. SE001
CE021 The MOSAIC dataset comprises 2,700 cancer samples across 11 tumor types as stated on the K Pro product page. SE001
CE022 Owkin reports K Pro built an IND-ready asset-positioning strategy in hours and produced a specialist review in a day that was later published in a peer-reviewed journal. SE004
CE023 K Pro's adoption is validated through enterprise collaborations with AstraZeneca, Bristol Myers Squibb and Sanofi. SE004, SE020
CE024 AstraZeneca signed a three-year licensing agreement for Owkin's K Pro agentic AI platform in May 2026. SE020, SE021
CE025 Sanofi entered a multi-year (reported five-year) agreement with Owkin to build bespoke drug-development AI agents, extending into immunology drug positioning. SE022, SE023, SE027
CE026 Owkin's federated training over MOSAIC and the hospital network gives clients access to a larger data pool for repositioning and biomarker discovery without moving patient records. SE017, SE001
CE027 Owkin's K platform depends on external suppliers including NVIDIA compute and open-source base LLMs, plus renewable hospital data-sharing agreements. SE005, SE006, SE017
CE028 Owkin integrates a patient validation hub so that K Pro outputs can be reality-checked against real-world patient data. SE001
CE029 Because raw patient data stays on hospital servers under Owkin's federated design, the architecture is intended to respect data ownership and patient privacy, supporting GDPR alignment. SE017
CE030 Owkin's CE-marked diagnostics MSIntuit CRC and RlapsRisk BC were moved to the Waiv spin-out in March 2026, removing the CE-marked IVDs from Owkin's own portfolio. SE018, SE026
CE031 K Pro carries no disclosed FDA clearance or CE mark as a medical device and is positioned as a research and decision-support tool. SE001, SE019
CE032 Owkin's public surface does not disclose a SOC 2 or ISO 27001 certification or a published model-risk-management framework for K Pro. SE001
CE033 Owkin frames K Pro as research-to-care intelligence whose AI insights are validated by real patients, connecting discovery to clinical reality. SE001
CE034 Following the March 2026 Waiv spin-out, Owkin's portfolio is concentrated on the K platform, the OwkinZero model, and the federated data network rather than on shipped diagnostic devices. SE018, SE001
CE035 Owkin's stated deployment outcomes for K Pro are company-reported single instances and are not independently audited or benchmarked across customers. SE004
CE036 No public independent audit, SOC 2 / ISO 27001 attestation, or model-governance report backs K Pro's enterprise data-handling and accuracy claims as of June 2026.
CE037 Owkin states an ambition to build Biological Artificial Super Intelligence, including an internal AI Scientist to automate discovery of new cures and diagnostics. SE005
CE038 Owkin's published science, open-source FL infrastructure and pathology foundation models are mature and externally validated, whereas the agentic K Pro product is early-stage and company-validated. SE006, SE008, SE016
CE039 Owkin's public developer footprint is real but modest, anchored by repositories such as FLamby (about 238 GitHub stars) and three Hugging Face models and spaces. SE013, SE010
CE040 Owkin staff publicly criticized internal management during the agentic-AI pivot accompanied by 2023 (17% of staff) and 2025 (~15% of workforce) reductions, adding execution risk to the roadmap. SE025
CU001 Owkin's primary commercial customer segment is large-cap biopharma companies that pay through multi-year enterprise SaaS licensing (K Pro) and collaboration agreements combining upfront fees and milestone payments. SU020, SU025
CU002 As of June 2026, Owkin's publicly named enterprise pharma partners include Sanofi, Bristol-Myers Squibb, AstraZeneca, Servier, MSD/Merck, Amgen, and Johnson & Johnson. SU021, SU023, SU006
CU003 Owkin's hospital and academic research center network participants are non-revenue-generating strategic data-generation partners, not paying commercial customers, providing multimodal patient data in exchange for federated learning infrastructure access. SU024, SU005
CU004 Owkin's hospital data access network spans 800+ institutions, with over 18 academic research centers equipped with active federated learning infrastructure nodes. SU016, SU024
CU005 K Pro is positioned for pharmaceutical researchers, executives, and biotech companies as an agentic AI scientist co-pilot for drug discovery, clinical trial optimization, and competitive intelligence. SU011, SU012
CU006 Hospital network participants contribute patient data to the MOSAIC consortium and Owkin's broader multimodal dataset under managed research and federated learning agreements, providing Owkin with a privacy-preserving biomedical data moat. SU024, SU005
CU007 Since the March 2026 Waiv spinout, AstraZeneca and MSD are customers of the independent Waiv entity for diagnostic products and their revenues are no longer consolidated in Owkin's accounts. SU023
CU008 Micregen, a regenerative medicine biotech, is a confirmed K Pro user representing an emerging biotech customer segment, with the company's MD providing a named testimonial in the October 2025 K Pro launch press release. SU011
CU009 Sanofi made a $180 million equity investment and a $90 million three-year oncology collaboration payment with Owkin in November 2021, covering drug target discovery in non-small cell lung cancer, triple-negative breast cancer, mesothelioma, and multiple myeloma. SU020, SU022, SU006
CU010 In October 2022, Owkin and Sanofi expanded their collaboration to cover drug positioning for Sanofi's immunology therapeutic pipeline, adding a new disease area scope to the original oncology-focused 2021 agreement. SU019, SU007
CU011 In June 2026, Sanofi signed a new multi-year K Pro collaboration with Owkin including a five-year enterprise license, under which Owkin will build purpose-built biopharma agents deployed inside Sanofi's own IT infrastructure and decision workflows. SU012, SU018
CU012 Sanofi Chief Digital Officer Emmanuel Frenehard publicly endorsed the June 2026 K Pro collaboration, stating it would enable Sanofi teams "to operate with greater speed, depth, and confidence as we continue to work to deliver transformative outcomes for patients." SU012
CU013 Bristol-Myers Squibb signed a multi-year strategic collaboration with Owkin in June 2022, with $80 million in upfront equity and collaboration fees plus up to $100 million in success-based milestone payments tied to cardiovascular clinical trial design. SU001, SU021, SU006
CU014 The BMS–Owkin collaboration focuses on AI-powered optimization of cardiovascular clinical trial design, including covariate adjustment, external control arms, endpoint definition, and patient subgroup identification using real-world and historical trial data. SU001, SU021
CU015 BMS Senior Vice President of Global Biometrics and Data Sciences Venkat Sethuraman publicly endorsed the Owkin collaboration, stating it would "extend our applications of AI and machine learning to enhance our discovery and development programs with high-quality, diverse data." SU001, SU021
CU016 AstraZeneca signed a three-year K Pro enterprise licensing agreement with Owkin in May 2026, to build AI agents for competitive intelligence and drug discovery decision-making integrated into AstraZeneca's IT systems; financial terms were not disclosed. SU015, SU016, SU017
CU017 The AstraZeneca K Pro license involves Owkin building competitive intelligence agents capable of analyzing clinical trial activity, recruitment trends, compound patent filings, and outcomes, integrated into AstraZeneca's IT infrastructure. SU016, SU017
CU018 The AstraZeneca K Pro license builds on a prior 2024 diagnostic AI collaboration (BRCAura) for gBRCA mutation prescreening, which achieved 93% sensitivity and ruled out approximately 40% of non-carrier patients in results presented at ESMO. SU017, SU023
CU019 In December 2023, Owkin entered a collaboration agreement with MSD to develop and commercialize AI-powered MSI-H diagnostic tools for endometrial, gastric, small intestinal, and biliary cancers in the EU market. SU010, SU023
CU020 The MSD diagnostic collaboration focused on prescreening for MSI-H biomarker status in four cancer types where MSI-H prevalence is low, to improve patient matching for immune checkpoint inhibitor therapy. SU010, SU023
CU021 Amgen and Johnson & Johnson are mentioned as early-stage research partners for Owkin AI cardiovascular prediction and broader disease applications; no contract terms or deal sizes have been publicly disclosed for either relationship. SU021, SU023
CU022 Institut Gustave Roussy's Head of Research Division publicly stated that K Pro's ability to "rapidly analyze and visualize complex patient data, including spatial omics, delivers insights in moments — it's a game-changer," providing the first named hospital-sector endorsement of K Pro. SU013
CU023 K Pro launched commercially in October 2025 targeting pharmaceutical companies, biotechs, and investors with a natural language interface for biomedical research, drug discovery, and clinical development decision-support. SU011, SU025
CU024 Within seven months of K Pro's October 2025 commercial launch, Owkin secured multi-year enterprise licenses from Sanofi (5-year, June 2026) and AstraZeneca (3-year, May 2026), representing the first two publicly confirmed K Pro enterprise customers. SU012, SU015, SU016
CU025 Owkin claims K Pro accelerated drug target identification by 70%, compressing the timeline from 12+ months to approximately 3 months, based on internal validation; no independent benchmarking of this performance claim has been published. SU011
CU026 Owkin claims that at least one unnamed pharma collaboration used K Pro-derived insights to reduce clinical trial duration by three years (35%) through improved patient segment identification; this claim is company-asserted and not independently verified. SU011
CU027 A survey of over 200 pharma executives commissioned by Owkin with STAT Brand Studio ahead of the October 2025 K Pro launch found nearly half identified overcoming data challenges as the top area where agentic AI could deliver greatest value. SU011
CU028 Sanofi's partnership with Owkin has deepened across three consecutive phases from 2021 to 2026 — equity investment and oncology collaboration (2021), immunology expansion (2022), and K Pro agentic AI (2026) — representing the strongest observable retention and expansion signal in Owkin's customer base. SU020, SU019, SU012
CU029 Owkin's 2024 revenues reached approximately $86 million, representing 83% year-over-year growth from 2023 fiscal year revenues, based on French regulatory disclosures cited by Maddyness; this is the most current public revenue figure available. SU014
CU030 No public NRR, GRR, customer churn rate, CSAT, or NPS data has been disclosed for Owkin's enterprise pharma partnerships or the K Pro platform; all formal retention metrics are absent from the public record. SU025, SU023
CU031 AstraZeneca's progression from a 2024 gBRCA diagnostic collaboration to signing a K Pro enterprise license in 2026 demonstrates a confirmed cross-sell and platform expansion motion, the only publicly documented land-and-expand example across Owkin's named accounts. SU017, SU015, SU016
CU032 No pharma customer or academic partner has publicly announced exiting or declining to renew an Owkin collaboration as of June 2026; the MSD diagnostics relationship transferred to Waiv rather than churning. SU023, SU025
CU033 BMS has not announced a K Pro license or an upgrade of its 2022 collaboration to the K Pro platform as of June 2026; BMS separately signed an AI deal with Anthropic in May 2026, a potentially competing platform investment. SU021, SU018
CU034 The MSD diagnostic assets (MSI-H tools) transferred to Waiv on the March 2026 spinout; MSD is now a customer of the independent Waiv entity, not directly of Owkin, and this relationship no longer contributes to Owkin's consolidated revenue. SU023
CU035 Sanofi represents the dominant share of Owkin's cumulative disclosed deal value: $180 million in equity, $90 million in oncology collaboration payments, an undisclosed immunology expansion, and an undisclosed K Pro five-year license fee — together exceeding any other named customer by an order of magnitude. SU020, SU019, SU012, SU022
CU036 Individual customer ARR contributions are not publicly disclosed by Owkin, preventing independent assessment of revenue concentration and making it impossible to quantify Sanofi's share of annual recurring revenues without management account access. SU025, SU023
CU037 Sifted reported in September 2025 that employees leaving Owkin cited "huge managerial issues," suggesting internal execution gaps that could affect delivery quality for multi-year enterprise AI deployments. SU014
CU038 Owkin's September 2025 restructuring eliminated approximately 50 positions (~15% of its French workforce), described as part of the strategic pivot to pure agentic AI and the planned divestiture of the diagnostics division. SU014
CU039 BMS signed a separate AI partnership with Anthropic in May 2026, and Sanofi extended a concurrent collaboration with CytoReason for up to $16 million, confirming that major pharma customers are managing diversified AI vendor portfolios and are not exclusively committed to Owkin. SU018, SU021
CU040 The March 2026 Waiv spinout removed AstraZeneca and MSD as consolidated Owkin customers, reducing visible revenue diversification while the remaining named commercial accounts are more concentrated on fewer, larger pharma relationships. SU023
CU041 Owkin's 800+ institution hospital network does not generate direct revenue; it provides the proprietary multimodal patient data foundation that differentiates K Pro from generic AI platforms and creates a structural moat against competitors without equivalent data network access. SU024, SU016
CU042 Owkin's K Pro deployment model — agents running inside customer IT stacks — creates institutional switching costs for pharma customers who integrate K Pro into their research workflows, but does not prevent parallel use of competing AI platforms. SU018, SU012
CR001 OwkinZero is a general-purpose AI (GPAI) model subject to EU AI Act GPAI obligations, which became active in August 2025, including training-data transparency and systemic-risk assessment requirements. SR003
CR002 The EU AI Act becomes fully applicable on 2 August 2026, with high-risk AI rules for general-area systems (biometrics, employment, etc.) delayed to December 2027 under the May 2026 AI Omnibus political agreement. SR003
CR003 GPAI providers under the EU AI Act must disclose training data content summaries, implement copyright policies, and assess whether their model poses systemic risk above high-capability thresholds. SR003
CR004 K Pro deployed as clinical decision-support — rather than pure pharma R&D — could trigger EU AI Act high-risk AI classification, requiring conformity assessment and CE marking. SR003, SR011
CR005 GDPR Article 9 prohibits processing of health data as special-category data unless an exception applies, such as scientific research (Art. 9(2)(j)) or explicit consent (Art. 9(2)(a)); violations carry penalties up to €20M or 4% of global annual turnover. SR005
CR006 Owkin processes pseudonymized patient health data from 180+ academic hospitals across multiple jurisdictions under GDPR special-category health data rules. SR001, SR002
CR007 Owkin's privacy policy was last updated on June 8, 2026, confirming active GDPR compliance maintenance as of the research date. SR001
CR008 Owkin uses HDS (Hébergeur de Données de Santé)-certified cloud providers for European patient data storage, as required under French health data hosting law. SR002
CR009 Owkin's cross-border data transfers to US affiliates rely on Standard Contractual Clauses approved by the European Commission, plus specific clauses for UK and Swiss data protection law. SR002
CR010 A GDPR enforcement action against Owkin for health data misuse could carry penalties up to €20M or 4% of global annual turnover, whichever is higher — on estimated 2024 revenue of $86M, this represents a potential penalty of up to ~$3.4M. SR005, SR009
CR011 No public GDPR enforcement action, data breach notification, or regulatory investigation against Owkin was identified in this research as of June 2026. SR001, SR003
CR012 Owkin has raised $304.1M in total equity as of its own FAQ disclosure, comprising a $2.1M seed, $72M Series A, and $230M Series B (Sanofi $180M + BMS $80M extension). SR013
CR013 Maddyness reported in September 2025 that Owkin held approximately $200 million in cash while planning the collective rupture procedure affecting ~50 employees. SR009
CR014 Owkin's 2024 revenue was $86 million (up 83% year-over-year from approximately $47M in 2023), with a net loss of $5.6 million, improved from a $31.3 million net loss in 2023. SR009
CR015 Maddyness reported that Owkin was seeking a new primary equity funding round of between $200 million and $300 million as of September 2025. SR009
CR016 The last primary equity raise at the Owkin parent entity closed in June 2022 (BMS $80M Series B extension), creating a funding gap of approximately four years as of June 2026. SR012, SR013
CR017 Owkin executed a collective rupture procedure (rupture conventionnelle collective) in September 2025, separating approximately 50 employees — roughly 15% of its ~325-person French workforce. SR009, SR010
CR018 Sources cited by Sifted described 'huge managerial issues' and employee uncertainty at Owkin during and following the September 2025 workforce restructuring. SR010
CR019 Thomas Clozel has served as CEO since Owkin's founding in 2016 and no succession plan or designated successor has been publicly disclosed as of June 2026. SR021, SR024
CR020 Owkin's disclosed pharma partnerships — Sanofi (anchor equity + 5-year K Pro license), BMS ($80M equity + up to $100M milestones), and AstraZeneca (3-year K Pro license) — collectively constitute nearly all named revenue relationships. SR014, SR012, SR016
CR021 Sanofi holds simultaneous roles as Owkin's largest equity investor (2021 $180M Series B anchor), primary revenue partner (€90M oncology collaboration 2021–2024; 5-year K Pro license from 2026), and most visible external validator of Owkin's technology. SR014, SR022
CR022 Bristol-Myers Squibb invested $80M in Owkin's Series B extension in June 2022 and is party to a collaboration worth up to $180M total (equity plus up to $100M in milestone payments) for cardiovascular and clinical AI applications. SR012
CR023 AstraZeneca signed a three-year K Pro licensing agreement in May 2026 for AI agents focused on competitive intelligence, expanding Owkin's named pharma customer base to three major accounts. SR016
CR024 Sanofi expanded a separate collaboration with CytoReason (up to $16M) in January 2026, demonstrating that Sanofi diversifies its agentic AI spend across multiple vendors and does not treat Owkin as its exclusive AI partner. SR022
CR025 The March 2026 Waiv diagnostics spinout removed MSIntuit CRC and RlapsRisk BC (CE-marked diagnostics) from Owkin's regulatory perimeter; Owkin retains a minority equity stake in Waiv. SR006, SR007
CR026 Waiv competes in AI diagnostics against PathAI, Tempus (which acquired Paige in 2024), Foundation Medicine, Ibex Medical Analytics, and Caris Life Sciences — a highly competitive field. SR007
CR027 Federated learning faces a fundamental data-heterogeneity challenge when training on non-IID data across geographically distributed hospital sites, requiring specialized aggregation algorithms that can introduce model drift. SR017
CR028 Federated learning architectures, despite not sharing raw data, are theoretically vulnerable to model inversion and membership inference attacks through gradient-sharing, which can partially reconstruct training data attributes. SR017, SR002
CR029 OwkinZero's claimed superior performance on biological reasoning tasks versus other leading LLMs is based on company-published benchmarks; no independently peer-reviewed external replication has been confirmed as of June 2026. SR015, SR021
CR030 K Pro's reported 70% acceleration in internal drug target identification (from 12+ months to 3 months) is a company self-reported claim published in the K Pro launch press release; no independent audit or external replication has been cited. SR015
CR031 EU AI Act high-risk rules for general-area AI systems (biometrics, employment, etc.) now apply from December 2027 under the AI Omnibus agreement; systems embedded in regulated products face an extended deadline of August 2028. SR003
CR032 EFPIA's position paper states that AI tools used solely for medicines R&D are likely exempt from EU AI Act high-risk classification under Articles 2.6 and 2.8 (scientific research exemption). SR011
CR033 Owkin's patient data network is listed in the French Health Data Hub's public registry of research projects, confirming French regulatory oversight of Owkin's use of French hospital patient data. SR030
CR034 French HDS certification mandates that health data be hosted by cloud providers with physical servers in France or Europe; Owkin's compliance requires ongoing certification maintenance by its cloud infrastructure partners. SR002
CR035 Owkin states it adheres to 'Privacy by Design' principles from earliest development stages through deployment, including pseudonymization, encryption, and access controls, but has not published an independent privacy or security audit. SR002
CR036 BMS milestone payments to Owkin are tied to verifiable drug-discovery research achievements; pharma R&D cycles span 5–10 years from target identification to regulatory approval, making milestone timing inherently lumpy. SR012
CR037 Maddyness reported that Owkin expected to reach profitability within 5–6 years from late 2025, implying continued cash consumption through approximately 2030–2031. SR009
CR038 At approximately 470 employees (post-restructuring estimate) and typical AI-biotech cost structures, Owkin's estimated monthly cash consumption ranges from $3 million to $9 million, implying $200M cash covers roughly 22–67 months of operations. SR009, SR018
CR039 The Waiv spinout transferred CE-marked diagnostics (MSIntuit CRC, RlapsRisk BC) to an independent entity; Owkin retains a minority equity stake but the diagnostics assets, revenue, and associated regulatory obligations are no longer on Owkin's books. SR006, SR008
CR040 Owkin's hospital partner network of 180+ institutions is the foundational data moat for AI training; loss of access at major hub hospitals would reduce training data diversity and degrade model performance. SR021
CR041 K Pro is built on MCP-server architecture that embeds AI agents inside pharma customers' own IT stacks; while this creates deeper workflow integration, it also means Sanofi holds negotiating leverage in future renewals. SR014, SR022
CR042 No known litigation, IP dispute, or regulatory enforcement action against Owkin is on public record as of June 2026; the absence of disclosure does not confirm absence of unresolved disputes. SR021
CR043 Owkin's K Pro relies on open-weight base models including Qwen, which are fine-tuned for biological reasoning; changes in Qwen's license terms or deprecation would require retraining on an alternative base, creating operational risk. SR022
CR044 GPAI providers must submit a public summary of training content under the EU AI Act GPAI Code of Practice template; no such public disclosure for OwkinZero has been located as of June 2026. SR003
CR045 A Sanofi exit from its K Pro partnership would simultaneously remove what is likely Owkin's largest single revenue stream, its most prominent equity investor, and its primary external validation signal to future investors. SR014, SR021
CR046 The European AI Office has the power to conduct evaluations of GPAI models, request information and corrective measures from providers, and apply sanctions under the EU AI Act — making it the direct enforcement authority for any OwkinZero GPAI compliance failure. SR034
CR047 The UK ICO's AI and Data Protection guidance imposes accountability, transparency, and fairness obligations on AI systems processing personal data under UK GDPR — applicable to Owkin's UK hospital data partnerships and any UK-based clinical research data flows. SR035
CV002 Owkin extended its Series B in June 2022 with an $80 million strategic investment from Bristol-Myers Squibb, accompanied by up to $100 million in potential milestone payments. SV013, SV023
CV003 Owkin's own FAQ confirms total capital raised of $304.1 million from investors including Sanofi, BMS, GV, Bpifrance, and Mubadala. SV014, SV021
CV004 As of June 2026, Owkin's last publicly disclosed post-money valuation remains $1 billion from the November 2021 Series B; no new primary round valuation has been publicly confirmed. SV012, SV020
CV005 PremierAlts, a secondary market aggregator and broker-dealer, cited an estimated Owkin valuation of approximately $1.6 billion in 2026, which is not based on a disclosed transaction and should be treated as indicative only. SV018, SV031
CV006 Owkin reported revenue of approximately $86 million for fiscal year 2024, representing year-over-year growth of 83% from FY2023, based on French regulatory financial data cited by Maddyness. SV017
CV007 Owkin posted a net financial loss of $5.6 million in FY2024, a significant improvement from a $31.3 million loss in FY2023, according to Maddyness reporting of French regulatory data. SV017
CV008 Owkin held approximately $200 million in cash at the end of FY2024, providing an estimated multi-year runway, per Maddyness reporting. SV017
CV009 As of September 2025, Owkin was seeking a new primary equity round of $200-300 million; no closing had been announced as of the June 2026 run date, per Maddyness reporting. SV017
CV010 Owkin completed the spinout of its diagnostics unit as Waiv (formerly Owkin Dx) in March 2026, with Waiv raising $33 million independently; Owkin retains a minority equity stake but no longer consolidates Waiv revenue in its P&L. SV015, SV022
CV011 Owkin announced a multi-year K Pro enterprise license with Sanofi in June 2026 for a five-year term; no financial terms were publicly disclosed. SV011, SV026
CV012 Owkin has disclosed no gross margin, net revenue retention rate, customer acquisition cost, or payback period in any public communication as of June 2026. SV014, SV017
CV013 Third-party analysts Growjo and Compworth estimate Owkin's annual revenue at approximately $125 million; this estimate has not been confirmed or denied by Owkin. SV019, SV025
CV014 Owkin has not raised a primary equity round in approximately four years as of June 2026, making the $1 billion mark structurally stale and creating elevated risk of re-pricing at the next fundraise. SV012, SV017
CV015 Recursion Pharmaceuticals (RXRX), listed on Nasdaq, had a market capitalization of approximately $1.71 billion as of June 2026, making it one of the most directly comparable publicly traded AI drug discovery companies to Owkin. SV002, SV030
CV016 Recursion reported total revenue of $74.7 million for FY2025, consisting primarily of revenue from collaboration agreements with Sanofi and Roche/Genentech, implying a trailing price-to-sales ratio of approximately 22.9x. SV001, SV002
CV017 Recursion held $753.9 million in cash and equivalents as of December 31, 2025, with a runway estimated to extend into early 2028 based on current operating plans. SV001
CV018 Recursion reported cash operating expenses of approximately $400 million in 2025 and guided for less than $390 million in 2026, reflecting a burn rate more than five times its revenue. SV001
CV019 Insilico Medicine completed its December 2025 IPO on the Hong Kong Stock Exchange, raising approximately $293 million and debuting with a market capitalization of approximately $2.1 billion. SV006
CV020 Insilico Medicine's market capitalization grew to approximately $2.52 billion by June 2026, implying a price-to-sales ratio of approximately 29x on FY2024 revenue of $85.83 million. SV006, SV004
CV021 Insilico Medicine's revenue fell 54% year-over-year in H1 2025 to $27.5 million, primarily due to the lumpy nature of licensing fees, illustrating the milestone revenue volatility that affects the entire AI drug discovery sector. SV006
CV022 Roche announced the acquisition of PathAI for $750 million upfront plus up to $300 million in milestone payments (total potential $1.05 billion) in May 2026, building on a multi-year 2021 partnership. SV007, SV008
CV023 PathAI had raised approximately $355-660 million in total funding (sources vary) and generated estimated revenue of $100-250 million prior to the Roche acquisition, implying an M&A multiple of approximately 3-7x trailing revenue. SV007, SV008
CV024 Tempus AI acquired Paige AI for $81.25 million (paid predominantly in Tempus stock) in August 2025; Paige had developed the first FDA-cleared AI tool in pathology and raised over $220 million in prior funding. SV009
CV025 Paige AI's $81.25 million acquisition price represented a significant discount to its prior total funding of over $220 million, constituting a down-exit that illustrates the risk of valuation compression in digital pathology AI. SV009, SV010
CV026 The median EV/Revenue multiple for biotech and genomics companies in Q1 2026 stood at 4.4x, down from a peak of 15.19x in Q4 2020, a 71% compression that reflects post-pandemic risk normalization and sector rotation. SV003
CV027 AI drug discovery platforms with confirmed clinical proof-of-concept command premium EV/Revenue multiples estimated at 6-10x as of 2026, according to Axis Intelligence and deal tracker analysis. SV004, SV003
CV028 Preclinical and platform-only AI biotechs without human clinical validation trade at below 4x EV/Revenue, a significant discount to the sector median, as market participants penalize proof-of-concept risk. SV003, SV004
CV029 At Owkin's last primary valuation of $1 billion and FY2024 revenue of approximately $86 million, the implied EV/Revenue multiple is approximately 11.6x — a 2.6x premium to the 4.4x sector median but substantially below Recursion (22.9x) and Insilico Medicine (29x). SV003, SV017
CV030 At the PremierAlts secondary estimate of $1.6 billion, the implied EV/Revenue multiple on FY2024 revenue of $86 million rises to approximately 18.6x, above the sector median but still below public clinical-stage peers. SV018, SV017
CV031 Sifted reported in September 2025 that Owkin staff publicly criticized the company's management for 'huge managerial issues', poor organisational transparency, and poorly communicated strategy shifts during the pivot to agentic AI. SV016
CV032 Owkin laid off approximately 50 staff (roughly 15% of its then-workforce of 325 French-entity employees) in late 2025 as part of a strategic restructuring toward agentic AI and away from diagnostics. SV016, SV017
CV033 No disclosed preference stack, anti-dilution provisions, or cap-table waterfall is available in the public record for Owkin, making the dilution impact of future rounds on existing investors unquantifiable. SV014, SV021
CV034 The biotech IPO market recovered strongly in 2025-2026, with Eikon Therapeutics raising $381 million in February 2026 and multiple companies pricing at or above range, signalling restored institutional appetite for clinical-stage biotech equities. SV005
CV035 Venture capital funding for AI drug discovery surged 78% to $5.7 billion in 2025 from $3.2 billion in 2024, creating a favorable financing environment for validated AI drug discovery platforms in 2026. SV004
CV036 Recursion's market capitalization declined 22.6% year-over-year in 2026 despite strong cash reserves, reflecting ongoing market skepticism about profitability timelines in AI drug discovery. SV002
CV037 Owkin's federated learning architecture deployed across 180+ research hospitals represents a privacy-preserving data access advantage that is structurally difficult for centralized competitors to replicate quickly in GDPR-constrained markets. SV011, SV024
CV038 Owkin retains a minority equity stake in Waiv following the March 2026 spinout, which provides some optionality on diagnostics exit value but is immaterial to the parent-company valuation thesis at this stage. SV015, SV022
CV039 K Pro's multi-year enterprise license structure (5 years for Sanofi, 3 years for AstraZeneca) introduces recurring SaaS-like revenue components, but contract values, ACV, and milestone schedules are entirely confidential. SV011, SV034
CV040 PathAI's total funding of $355-660 million was acquired by Roche at $750 million upfront, implying strategic acquirers are willing to pay approximately 1.1-2.1x total invested capital for clinical-stage AI diagnostics companies with regulatory-grade algorithms. SV007, SV008
CV041 Owkin CEO Thomas Clozel publicly targeted profitability within five to six years from 2025, implying continued cash consumption through 2030-2031 and dependence on K Pro partnership revenues and future capital raises. SV017, SV044
CV042 Insilico Medicine's $2.52 billion June 2026 market cap on $85.83 million FY2024 revenue (29x P/S) trades at a roughly 2.5x premium to Owkin's implied multiple at $1 billion — suggesting Owkin appears relatively undervalued on public market comps but the premium reflects Insilico's active drug pipeline, not platform software alone. SV006, SV017
CV043 PathAI's $750 million acquisition by Roche demonstrates that platform-plus-regulatory-approval AI diagnostics companies can achieve exits exceeding total capital raised even with sub-$660M total funding, providing a valuation floor reference for Owkin's residual diagnostics optionality via Waiv. SV007, SV010
CV044 The only verifiable financial filing available publicly for Owkin as a private company is French regulatory data reported by Maddyness; no consolidated IFRS or US GAAP audited accounts have been voluntarily disclosed. SV017, SV014
CV045 Owkin's total disclosed capital of $304.1 million against $200 million remaining cash (end-2024) implies cumulative cash consumption of approximately $104 million or more over its history, suggesting relatively capital-efficient deployment for a nine-year-old AI biotech at this scale. SV014, SV017
来源
编号出版方标题引文
SO001 Owkin Owkin Homepage — AI for Biology K Pro today is an AI scientist that informs decision making and generates insights across biopharmaceutical R&D.
SO002 Owkin About Us | Owkin Get answers in minutes, not hours, with K Pro — the AI agent for biology, powered by multimodal patient data.
SO003 Owkin Owkin Press Kit FAQs — How much investment has Owkin received? Owkin has raised $304.1m USD from investors such as Bpifrance Large Ventures, GV, NJF Capital, Cathay Innovation, F-Prime, Otium Capital, Eight Roads Ventures, Plug and Play, Mubadala and MACSF (French Pension Fund for Clinicians), as well as corporate institutional partners such as Sanofi and Bristol-Myers Squibb.
SO004 Owkin Owkin to build AI agents as part of a multi-year K Pro collaboration with Sanofi K Pro's development is part of Owkin's mission to achieve Biological Artificial Superintelligence to unlock new therapeutic discoveries beyond the limits of human cognition alone.
SO005 Owkin Owkin Launches K Pro — The First Agentic AI Co-Pilot for Biopharma Powered by Biological Reasoning Models K Pro has already accelerated internal drug target identification by 70% (from 12+ months to 3 months); built an IND-ready asset positioning strategy in hours instead of months of traditional analysis.
SO006 Owkin Owkin creates new spin out Waiv, formerly Owkin Dx, with $33m financing The spin out of Waiv represents the natural evolution of our strategy to validate AI in the real world.
SO007 Owkin Owkin announces multi-year clinical data science strategic collaboration with Bristol-Myers Squibb We are excited to start this collaboration with Bristol Myers Squibb, a global leader in cardiovascular with a strong development pipeline which has the potential to change the lives of millions of patients.
SO008 Owkin Owkin expands reach of AI diagnostic for colorectal cancer through integration with Sectra Owkin expands reach of AI diagnostic for colorectal cancer through integration with Sectra.
SO009 Owkin How AI can make clinical trials faster, more efficient and more personalised Both our strategic drug development partnership with BMS and our strategic drug discovery partnership with Sanofi use the same Owkin data platform, the first time one platform has been used to both discover and develop new treatments.
SO010 Sifted Owkin staff criticise 'huge managerial issues' as French unicorn moves to AI agents Staff criticise 'huge managerial issues' as Owkin pivots from diagnostics to AI agents and announces approximately 50 redundancies.
SO011 Maddyness Owkin prévoit de se séparer d'une cinquantaine de salariés Owkin a réalisé un chiffre d'affaires de 86 millions de dollars l'année dernière, en hausse de 83% rapport à 2023. Elle a enregistré l'année dernière une perte financière de 5,6 millions de dollars.
SO012 Pharmaphorum Owkin bags $180m BMS alliance for cardiovascular clinical trials BMS' senior vice president of global biometrics and data sciences said the partnership will extend applications of AI and machine learning to enhance discovery and development programmes.
SO013 Yahoo Finance Owkin Launches K Pro — The First Agentic AI Co-Pilot for Biopharma K Pro integrates Owkin Zero, a fine-tuned biological LLM that performs biological reasoning with higher performance than other leading LLMs.
SO014 European Biotechnology Owkin spins out AI diagnostics startup Waiv with $33m financing By separating units such as Waiv into independent companies while maintaining strategic ties, Owkin appears to be building a network of businesses that connect data generation, AI development, and clinical applications.
SO015 Actuia AI and cancer — OWKIN presents two AI-based diagnostic solutions with CE-IVD certification OWKIN's new AI diagnostic solution, MSIntuit CRC, will account for about half of the number of cases requiring screening.
SO016 ACN Newswire Owkin Launches K Pro — The First Agentic AI Co-Pilot for Biopharma (Press Release) K Pro is described as the first agentic AI co-pilot for biopharma, built on Owkin Zero biological LLM.
SO017 Benzinga Sanofi-Owkin Expand AI Collaboration With New Agentic AI Initiative Sanofi and Owkin expand AI collaboration with new agentic AI initiative.
SO018 Waiv Waiv Achieves Dual CE Marking Under IVDR for AI Precision Testing Waiv achieves dual CE marking under IVDR, propelling AI precision testing for breast and colorectal cancer for clinical routine.
SO019 Tracxn Owkin Company Profile — Tracxn
SO020 Latka OWKIN Revenue 2023 — $101.4M ARR, $1B Valuation
SO021 PremierAlts Owkin Private Stock Price & Valuation ($1.6B) — 2026 Data
SO022 Craft.co Owkin CEO and Key Executive Team
SO023 MedCity News AI Startup Waiv Spins Out of Owkin With $33M to Bring AI Precision to Oncology
SO024 Tracxn Owkin Funding Rounds and List of Investors
SO025 AllSci Sanofi expands Owkin partnership for purpose-built AI agents Sanofi expands Owkin partnership for purpose-built AI agents.
SO026 Pharmaphorum How Owkin uses AI for clinical trial optimization with BMS The initial deal included $80M upfront for Owkin and up to $100M more in milestone payments, centered on using Owkin's AI to design more precise and efficient clinical trials for BMS.
SM001 Owkin Owkin | Building Biological Artificial Superintelligence We are building an autonomous AI scientist to automate biopharma R&D and support clinical research.
SM002 Owkin What is federated learning? | Owkin Instead of gathering data on a single server, the data remains locked on servers as the algorithms and only the predictive models travel between the servers.
SM003 Owkin Owkin becomes 'unicorn' with $180m investment from Sanofi Owkin is now a 'unicorn' - a startup valued at more than $1 billion - through a new $180 million investment from Sanofi. The total payment for three years is $90 million, plus additional research milestone-based payments.
SM004 World Pharma News Sanofi invests $180 million equity in Owkin's artificial intelligence and federated learning to advance oncology pipeline Sanofi announced an equity investment of $180 million and a new strategic collaboration with Owkin comprised of discovery and development programmes in four exclusive types of cancer.
SM005 Sifted Owkin staff criticise 'huge managerial issues' as French unicorn moves to AI agents French biotech unicorn Owkin is planning to cut dozens of staff, Sifted has learned, as the company shifts to a business model focused on AI agents.
SM006 Wikipedia Owkin Owkin has raised over $255 million and became a 'unicorn' in November 2021 through a $180 million investment from Sanofi.
SM007 Global Market Insights Artificial Intelligence in Drug Discovery Market Size, Share - 2035 Market Size By Component (Software, Services), By Technology (Machine Learning, Other Technologies), By Application Type (Molecular Library Screening, Target Identification, Drug Optimization and Repurposing, De Novo Drug Designing, Preclinical Testing).
SM008 Mordor Intelligence Artificial Intelligence In Drug Discovery Market Size and Share The Artificial Intelligence In Drug Discovery Market size was valued at USD 2.58 billion in 2025 and is estimated to grow from USD 3.25 billion in 2026 to reach USD 10.29 billion by 2031, at a CAGR of 25.94%.
SM009 Fortune Business Insights Artificial Intelligence (AI) In Drug Discovery Market Report, 2034 The global artificial intelligence in drug discovery market size was estimated at USD 4.46 billion in 2025... expected to rise from USD 5.00 billion in 2026 to USD 12.56 billion by 2034, expanding at a CAGR of 12.20%.
SM010 Research and Markets AI in Drug Discovery Market Report 2026 Artificial Intelligence (AI) in Drug Discovery Market Characteristics: Market Definition & Scope, Market Segmentations.
SM011 Mordor Intelligence Precision Oncology Market Size & Share Outlook to 2031 The Precision Oncology Market size was valued at USD 115.51 billion in 2025 and is estimated to grow from USD 127.68 billion in 2026 to reach USD 201.27 billion by 2031, at a CAGR of 9.53%.
SM012 Precedence Research Precision Oncology Market Size To Hit USD 338.89 Bn By 2035 The global precision oncology market size accounted for USD 133 billion in 2025 and is predicted to increase from USD 146.24 billion in 2026 to approximately USD 338.89 billion by 2035, expanding at a CAGR of 11.33%.
SM013 Fortune Business Insights Precision Oncology Market Size, Share | Industry Report 2034 The global precision oncology market size was valued at USD 122.60 billion in 2025... projected to grow from USD 133.06 billion in 2026 to USD 317.16 billion by 2034, exhibiting a CAGR of 11.47%.
SM014 EurekAlert! (Owkin) Major study validates Owkin's best in class AI diagnostic for colorectal cancer MSIntuit CRC, the only CE-marked AI diagnostic for MSI testing for colorectal cancer from digital H&E slides... with 96% sensitivity, can rule out almost half of the MSS population of colorectal cancer patients.
SM015 EFPIA The Pharmaceutical Industry in Figures - Key Data 2024 In 2023 it invested an estimated EUR 50,000 million in R&D in Europe. It directly employs some 900,000 people.
SM016 BioSpace Undeterred by Political, Economic Headwinds, Pharma Ups R&D Investment in 2024 and Beyond Drugmakers the world over pumped nearly $288 billion into research and development... could reach just under $340 billion by 2030.
SM017 R&D World Global pharma R&D hits $276B, triples marketing spend The study found that global biopharma R&D investment hit $276 billion in 2021 - more than double the commonly cited estimates.
SM018 EFPIA EFPIA Statement on the use of AI in the medicinal product lifecycle in the context of the AI Act When used solely for the purpose of medicines R&D, such AI enabled tools are exempt from the requirements of the EU AI Act... the majority of these tools would in any case not be considered high risk.
SM019 Bird & Bird (Bioslice Blog) EMA adopts reflection paper on the use of Artificial Intelligence (AI) The Paper reflects the views of the EMA on the use of AI from drug discovery and non-clinical development, to clinical trials, manufacturing and post-authorisation... encourages a risk-based approach.
SM020 MarketsandMarkets Drug Discovery Technologies Market - Global Forecast to 2030 The Drug Discovery Technologies market... stood at US$30.58 billion in 2025 and is projected to advance at a resilient CAGR of 11.0% from 2025 to 2030, culminating in a forecasted valuation of US$51.51 billion.
SM021 U.S. Food and Drug Administration Artificial Intelligence and Machine Learning (AI/ML)-Enabled Medical Devices The agency lists numerous AI/ML-enabled medical devices cleared through early 2026, including imaging and diagnostic algorithms from multiple manufacturers.
SM022 Nature Biotechnology Pooled perturbation screens with high-content readouts Pooled perturbation screens with high-content readouts ranging from single-cell RNA sequencing to imaging-based spatial profiling.
SM023 World Health Organization Cancer - Fact sheet Cancer is a leading cause of death worldwide, accounting for nearly 10 million deaths in 2022.
SM024 Evaluate World Preview 2026 Breakthrough categories such as GLP-1s, accelerating patent expirations, pricing pressure, changing R&D economics, and rising competition from China are redefining how value is created.
SM025 Precedence Research Federated Learning Market Size to Hit USD 17,462.60 Million by 2035 The global federated learning market size accounted for USD 1,219.00 million in 2025 and is predicted to increase from USD 1,590.80 million in 2026 to approximately USD 17,462.60 million by 2035, expanding at a CAGR of 30.50%.
SP001 Owkin Owkin Launches K Pro: the First Agentic AI Co-Pilot for Biopharma Powered by Biological Reasoning Models K Pro has already accelerated internal drug target identification by 70% (from 12+ months to 3 months); built an IND-ready asset positioning strategy in hours instead of months of traditional analysis; and written a specialist review in one day, which was later published in a peer-reviewed scientific journal.
SP002 Owkin Owkin to Build AI Agents as Part of a Multi-Year K Pro License Agreement with AstraZeneca Owkin, the agentic AI company pioneering Biological Artificial Superintelligence to revolutionize drug discovery and development, today announced an agreement with AstraZeneca to build biopharma agents as part of their three-year licensing of K Pro.
SP003 Owkin Owkin to Build AI Agents as Part of a Multi-Year K Pro Collaboration with Sanofi Owkin and Sanofi have collaborated since 2021 through a €90 million strategic partnership focused on target identification in oncology and patient subgrouping. This new collaboration represents the next evolution in the partnership, backed by a five-year license for K Pro.
SP004 MOSAIC Research / Owkin MOSAIC: AI-Ready Spatial Multiomics Dataset in Oncology — About Owkin is an AI biotechnology company startup that uses AI to find the right treatment for every patient. Using federated learning, a pioneering collaborative AI framework, Owkin enables partners to unlock valuable insights from siloed datasets while protecting patient privacy and securing proprietary data.
SP005 Owkin MOSAIC: Spatial Omics and Multimodal Dataset in Oncology
SP006 Roche Roche Enters into a Definitive Merger Agreement to Acquire PathAI Under the terms of the agreement, Roche will pay a purchase price of USD 750 million upfront and additional milestone payments of up to USD 300 million. PathAI's AISight IMS software interface is efficient and user-friendly, seamlessly integrating advanced analysis and workflow capabilities within the digital pathology laboratory.
SP007 Isomorphic Labs Isomorphic Labs Announces Series B Investment Round Isomorphic Labs announces it has raised $2.1 Billion in Series B funding. The financing round is led by Thrive Capital, and includes participation from existing backers Alphabet and GV alongside new investors MGX, Temasek, CapitalG, and the UK Sovereign AI Fund.
SP008 Recursion Pharmaceuticals Recursion Pharmaceuticals — Company Homepage We have generated and aggregated one of the largest fit-for-purpose proprietary biological and chemical datasets in the world — >50 petabytes spanning phenomics, transcriptomics, proteomics, ADME, and de-identified patient data.
SP009 Recursion Pharmaceuticals The Recursion Operating System — Platform
SP010 Insilico Medicine Insilico Medicine — Company Homepage
SP011 Insilico Medicine Insilico Medicine — Pipeline Overview The rapid progress of internal pipeline demonstrates the generative-AI driven drug discovery capabilities of our Pharma.AI platform consisting of Biology42, Chemistry42, and Medicine42.
SP012 BiomedNexus 25 AI Drug Discovery Companies Actually Delivering Clinical Candidates (2026) Eighteen months after launch, Chai Discovery closed a $130 million Series B at a $1.3 billion valuation. Integration of Recursion and Exscientia has not been without friction — several legacy candidates were deprioritized post-merger.
SP013 Axis Intelligence AI Drug Discovery 2026: 173 Programs, FDA Framework & Market 2025: $1.94B (28% YoY growth). 2026: $2.6-2.8B projected. Venture capital $8B+ annually. Clinical validation premium (+$400-500M) — rentosertib success eliminates proof of concept discount. 81% pharma deploy AI, 30% new 2026 drugs use AI.
SP014 Science Reader AI Drug Discovery in 2026: 5 Signs That It's Still Experimental No AI-designed drug has received regulatory approval yet. Phase II success rates match traditional drugs despite early-stage gains. In 2025, AI drug discovery partnerships exceeded $15 billion in so-called "biobucks" — the actual upfront payments averaged around 2% of those headline figures.
SP015 Pharmaphorum Shakeup at BenevolentAI Continues with Job, Programme Cuts BenevolentAI started life as a public company with a valuation of more than £1 billion but has seen its share price fall by more than 90% since then and is now valued at around £85 million, despite signing a major $594 million alliance last year with German group Merck KGaA.
SP016 Drug Discovery Trends Recursion-Exscientia Merger Consolidates AI in Drug Discovery Field Recursion Pharmaceuticals and Exscientia, two of the most prominent AI-driven drug discovery companies, are merging. The deal would create a portfolio of clinical and near-clinical programs with an expected 10 clinical readouts over the next 18 months with peak sales opportunities in excess of $1 billion.
SP017 Paige Paige — Redefining Precision Oncology Our AI-powered solutions provide a deep understanding of cancer's molecular and cellular complexities, helping to deliver reliable detection and characterization of disease. From tissue analysis to molecular profiling, we're redefining what's possible in precision diagnostics.
SP018 PathAI PathAI — AISight Enterprise Digital Pathology Platform AISight is a cloud-native, open platform enterprise workflow solution used by the world's leading laboratories and research centers to power their digital pathology workflows and AI applications.
SP019 Proscia Proscia Launches Fifth Generation of Concentriq Built on a new AI-native architecture, Concentriq delivers clinically and scientifically grounded intelligence to decisions throughout drug discovery, development, and case review.
SP020 BenevolentAI BenevolentAI — Company Homepage BenevolentAI brings together a unique cross-section of science and technology teams who push AI and machine learning to new limits. Building on a decade of investment in our knowledge graph and proprietary ontologies.
SP021 Schrödinger Schrödinger — Opening New Worlds for Molecular Discovery Schrödinger's computational platform, powered by physics, is transforming the way therapeutics and materials are discovered to make innovations of the future achievable.
SP022 Foundation Medicine FoundationOne CDx — Foundation Medicine FoundationOne CDx and FoundationOne Liquid CDx are for prescription use only and are FDA-approved qualitative next-generation sequencing based in vitro diagnostic tests for advanced cancer patients with solid tumors. FoundationOne CDx analyzes 324 genes.
SP023 mHealthSpot Owkin Partners with Sanofi to Build Custom AI Agents for Drug Discovery
SP024 Pharmaphorum AI Biotechs Exscientia and Recursion Agree $688M Merger
SP025 MarketBeat Recursion Pharmaceuticals Details AI-Driven Drug Pipeline, Sanofi/Roche Milestones, Runway to 2028 (2026-03-22)
SI001 Owkin Owkin becomes 'unicorn' with $180m investment from Sanofi and four new collaborative projects Owkin has raised $180 million in a Series B financing led by Sanofi, achieving a valuation above $1 billion.
SI002 Owkin / BusinessWire Owkin Raises $180 Million Series B Financing Owkin Raises $180 Million Series B Financing
SI003 Sifted Owkin staff criticise 'huge managerial issues' as French unicorn moves to AI agents huge managerial issues
SI004 Pharmaphorum Owkin bags $180m BMS alliance for cardiovascular clinical trials Bristol-Myers Squibb, which is paying $80 million upfront to tap into its expertise. The deal is split between a Series B financing led by BMS and upfront fees, taking the total raised by the Paris, France-based start-up to $300 million, and also includes another $100 million in potential milestone payments.
SI005 Owkin / BusinessWire Owkin Creates New Spin-out Waiv, Formerly Owkin Dx, With $33M Financing Owkin Creates New Spin-out Waiv, Formerly Owkin Dx, With $33M Financing
SI006 RD World Online Sanofi deepens its Owkin bet with a five-year deal to build bespoke drug development agents
SI007 FierceBiotech AstraZeneca pens agentic AI licensing deal with Owkin
SI008 Tracxn Owkin — 2026 Funding Rounds and List of Investors
SI009 Growjo OWKIN: Revenue, Competitors, Alternatives
SI010 Premier Alts Owkin Private Stock Price and Valuation ($1.6B) — 2026 Data
SI011 Revelio Labs Owkin — Employee Count and Workforce Data
SI012 Owkin AI Agent for Biopharma Decision Making — Owkin K Pro
SI013 mHealthSpot AstraZeneca licenses Owkin's K Pro AI platform in three-year deal
SI014 Owkin Owkin expands collaboration with Sanofi leveraging AI for drug positioning in immunology
SI015 Owkin Press Kit — Owkin
SI016 AllSci Sanofi Expands Owkin Partnership for Purpose-Built AI Agents
SI017 FirstWord Pharma Sanofi, Owkin partnership moves from cancer discovery work to AI agents
SI018 HLTH Owkin Spins Out Waiv with $33M Financing from OTB Ventures and Alpha Intelligence Capital
SI019 Signalbase Owkin Secures $320 Million in Funding to Revolutionize AI-driven Biotechnology Solutions
SI020 Clay How Much Did Owkin Raise? Funding and Key Investors
SI021 Owkin What is Federated Learning? — Owkin
SI022 World Pharma Today Sanofi invests $180 million equity in Owkin's artificial intelligence and federated learning to advance oncology pipeline
SI023 Wikipedia Owkin — Wikipedia
SI024 World Pharma News Sanofi invests $180 million equity in Owkin's artificial intelligence and federated learning
SI025 Owkin Owkin — Official Website
SE001 Owkin AI Agent for Biopharma Decision Making | Owkin K Pro K Pro, the AI agent for biology, powered by multimodal patient data, for smarter biopharma decision making. MOSAIC spatial & multiomics dataset (2700 cancer samples, 11 tumor types); World-largest patient data network (104 hospitals).
SE002 Owkin Agentic AI built for scientists | K Pro FREE K Navigator is now called K Pro FREE ... it ensures every result is credible and easily traceable back to trusted sources (PubMed and biomedical databases).
SE003 Owkin Biological Reasoning Model | Owkin Zero Trained on 300,000+ verified Q&A pairs across eight key biomedical tasks, it draws on proprietary patient data, spatial and bulk transcriptomics, and Owkin's domain expertise.
SE004 Owkin Owkin Launches K Pro: the First Agentic AI Co-Pilot for Biopharma Powered by Biological Reasoning Models K Pro has already accelerated internal drug target identification by 70% (from 12+ months to 3 months); built an IND-ready asset positioning strategy in hours ... At its core, K Pro integrates Owkin Zero, a fine-tuned biological LLM.
SE005 Owkin / BusinessWire Owkin Advances Frontier Model Development for Biological Artificial Super Intelligence, Accelerated by NVIDIA The new OwkinZero model leverages the NVIDIA NeMo RL ... It will power the next generation of Owkin's agentic AI capabilities, available through Owkin's interoperable agentic infrastructure, Owkin K, for biopharma and Owkin's internal AI Scientist.
SE006 arXiv (Owkin authors) OwkinZero: Accelerating Biological Discovery with AI Specialized 8-32B OwkinZero models substantially outperform larger, state-of-the-art commercial LLMs on these biological benchmarks ... over 300,000 verifiable question-and-answer pairs.
SE007 Nature Cancer A deep-learning framework to predict cancer treatment response from histopathology images through imputed transcriptomics ENLIGHT-DeepPT predicted responders across five independent cohorts with an overall odds ratio of 2.28 and a 39.5% increased response rate among predicted responders.
SE008 Nature Medicine Federated learning for predicting histological response to neoadjuvant chemotherapy in triple-negative breast cancer Federated learning for predicting histological response to neoadjuvant chemotherapy in triple-negative breast cancer.
SE009 medRxiv (Owkin authors) Scaling Self-Supervised Learning for Histopathology with Masked Image Modeling Scaling Self-Supervised Learning for Histopathology with Masked Image Modeling (Filiot et al., medRxiv 2023), releasing the Phikon model.
SE010 Hugging Face owkin (Owkin) — organization hub Understand complex biology through agentic AI ... spaces 3 ... models 3 (Federated Learning with Substra; Substra Trainer; Agents).
SE011 Hugging Face owkin/phikon · Hugging Face model card Phikon is a self-supervised learning model for histopathology trained with iBOT ... Params (M): 85.8 ... Pretrain Dataset: 40 million pan-cancer tiles extracted from TGCA ... License: Owkin non-commercial license.
SE012 Hugging Face owkin/phikon-v2 · Hugging Face model card Phikon-v2 is a Vision Transformer Large pre-trained with Dinov2 ... on PANCAN-XL, a dataset of 450M 20x magnification histology images sampled from 60K whole slide images.
SE013 GitHub Owkin · GitHub organization Popular repositories ... Cross-silo Federated Learning playground in Python ... Python 238 ... We publicly release Phikon.
SE014 GitHub owkin/FLamby — Cross-silo Federated Learning benchmark FLamby is a benchmark for cross-silo Federated Learning with natural partitioning, currently focused in healthcare applications ... 7 real-world federated datasets.
SE015 GitHub owkin/HistoSSLscaling — Phikon code release Code associated to the publication: Scaling self-supervised learning for histopathology with masked image modeling ... We publicly release Phikon.
SE016 Substra (Linux Foundation) Substra documentation — open-source federated learning Substra is an open source federated learning (FL) software ... Traceable — All machine learning operations are logged in an auditable read-only database ... Substra was created by Owkin and is now hosted by the Linux Foundation for AI and Data.
SE017 Owkin What is federated learning? | Owkin Instead of gathering data on a single server, the data remains locked on servers ... and only the predictive models travel between the servers ... while respecting data ownership and privacy.
SE018 Owkin / BusinessWire Owkin Creates New Spin-out Waiv, Formerly Owkin Dx, With $33M Financing Owkin Creates New Spin-out Waiv, Formerly Owkin Dx, With $33M Financing.
SE019 U.S. Food & Drug Administration Artificial Intelligence and Machine Learning (AI/ML)-Enabled Medical Devices FDA guidance on Software as a Medical Device (SaMD) and AI/ML-enabled medical devices.
SE020 FierceBiotech AstraZeneca pens agentic AI licensing deal with Owkin AstraZeneca pens agentic AI licensing deal with Owkin.
SE021 mHealthSpot AstraZeneca licenses Owkin's K Pro AI platform in three-year deal AstraZeneca licenses Owkin's K Pro AI platform in a three-year deal.
SE022 R&D World Online Sanofi deepens its Owkin bet with a five-year deal to build bespoke drug development agents Sanofi deepens its Owkin bet with a five-year deal to build bespoke drug development agents.
SE023 Owkin Owkin expands collaboration with Sanofi, leveraging AI for drug positioning in immunology Owkin expands collaboration with Sanofi, leveraging AI for drug positioning in immunology.
SE024 Wikipedia Owkin Owkin is an AI biotechnology company co-founded by Thomas Clozel and Gilles Wainrib using federated learning and multimodal data.
SE025 Sifted Owkin staff criticise 'huge managerial issues' as French unicorn moves to AI agents huge managerial issues.
SE026 HLTH Owkin spins out Waiv with $33M financing Owkin spins out Waiv with $33M financing from OTB Ventures and Alpha Intelligence Capital.
SE027 MobiHealthNews Owkin, Sanofi expand AI collaboration to build drug development AI agents Owkin and Sanofi expand their AI collaboration to build drug-development AI agents.
SU001 Owkin Owkin Raises $80 Million from Bristol-Myers Squibb to Advance AI for Oncology "We look forward to collaborating with Owkin to extend our applications of AI and machine learning to enhance our discovery and development programs with high-quality, diverse data from clinical and real world sources." — Venkat Sethuraman, SVP Global Biometrics and Data Sciences, BMS
SU002 Owkin Sanofi and Owkin Announce a Major Collaboration to Advance Cancer Research
SU003 Owkin Owkin and Sanofi Sign Expanded Collaboration
SU004 Owkin Owkin — Sanofi Announced Collaboration
SU005 Owkin Owkin Raises $180 Million to Power Federated Learning for Medical Research
SU006 TechCrunch Owkin Raises $180M in Series B Led by Sanofi
SU007 Pharma Journalist Owkin and Sanofi Expand AI Drug Development Partnership
SU008 Awaira Owkin Company Profile
SU009 Crunchbase Owkin Organization Profile
SU010 Owkin Owkin Enters Collaboration Agreement with MSD to Develop AI-Powered Diagnostics for Cancer
SU011 Owkin Owkin Launches K Pro — The First Agentic AI Co-Pilot for Biopharma Powered by Biological Reasoning Models "The task would have been nearly impossible to achieve at the scale that Owkin K handled in just a few work sessions." — Ben Mellows, Managing Director, Micregen
SU012 Owkin Owkin to Build AI Agents as Part of a Multi-Year K Pro Collaboration with Sanofi "By implementing purpose-built agentic systems into our workflows, we aim to empower our teams to operate with greater speed, depth, and confidence." — Emmanuel Frenehard, Chief Digital Officer, Sanofi
SU013 Owkin K Pro — AI Scientist for Biopharma "Its ability to rapidly analyze and visualize complex patient data, including spatial omics, delivers insights in moments. It's a game-changer." — Head of Research Division, Institut Gustave Roussy
SU014 Sifted French AI Biotech Owkin Plans to Cut Dozens of Staff in Pivot to AI Agents Departing and current staff cited "huge managerial issues" following Owkin's September 2025 workforce reduction.
SU015 Owkin Owkin to Build AI Agents as Part of a Multi-Year K Pro License Agreement with AstraZeneca
SU016 Fierce Biotech AstraZeneca Pens Agentic AI Licensing Deal with Owkin
SU017 mHealthSpot AstraZeneca Licenses Owkin's K Pro AI Platform in Three-Year Deal
SU018 R&D World Sanofi Deepens Its Owkin Bet with a Five-Year Deal to Build Bespoke Drug Development Agents Sanofi is not placing its bet on Owkin alone: in January, the company expanded a parallel collaboration with CytoReason, another agentic-AI vendor, in a deal valued at up to $16 million.
SU019 Owkin Owkin Expands Collaboration with Sanofi Leveraging AI for Drug Positioning in Immunology
SU020 Owkin Owkin Becomes a Unicorn with $180M Investment from Sanofi and Four New Collaborative Projects
SU021 Pharmaphorum Owkin Bags $180M BMS Alliance for Cardiovascular Clinical Trials
SU022 World Pharma News Sanofi Invests $180 Million Equity in Owkin's AI and Federated Learning to Advance Oncology Pipeline
SU023 Wikipedia Owkin — Wikipedia
SU024 Owkin Owkin — Federated Learning
SU025 Owkin Owkin Homepage
SR001 Owkin Owkin Privacy Policy — Last updated June 8, 2026 Owkin is firmly committed to conducting its business in accordance with the applicable data protection regulations and, in particular, the General Data Protection Regulation (EU) 2016/679.
SR002 Owkin Owkin Patient Information — Data Privacy, HDS, and International Transfers Owkin uses cloud service providers with servers located in the same region as the region as its research partners providing access to the data is established.
SR003 European Commission EU AI Act — A Risk-Based Approach: Regulatory Framework and Application Timeline The AI Act entered into force on 1 August 2024, and will be fully applicable 2 years later on 2 August 2026.
SR004 U.S. Food and Drug Administration Artificial Intelligence-Enabled Medical Devices — FDA AI Device List (2026) The FDA encourages the development of innovative, safe, and effective medical devices, including devices that incorporate artificial intelligence (AI).
SR005 GDPR-Info.eu Art. 9 GDPR — Processing of Special Categories of Personal Data Processing of personal data revealing racial or ethnic origin... and the processing of genetic data, biometric data for the purpose of uniquely identifying a natural person, data concerning health... shall be prohibited.
SR006 Owkin Owkin Creates New Spin-Out Waiv, Formerly Owkin Dx, With $33M Financing The spin out of Waiv represents the natural evolution of our strategy to validate AI in the real world.
SR007 MedCity News AI startup Waiv spins out of Owkin for precision medicine in oncology and digital pathology A growing number of companies offer AI technologies that inform drug selection in clinical care and clinical trials, many of them focused on cancer.
SR008 European Biotechnology Owkin spins out AI diagnostics startup Waiv with $33M financing By separating units such as Waiv into independent companies while maintaining strategic ties, Owkin appears to be building a network of businesses that connect data generation, AI development, and clinical applications.
SR009 Maddyness Owkin prévoit de se séparer d'une cinquantaine de salariés (Owkin plans to part with about fifty employees) Elle a enregistré l'année dernière une perte financière de 5,6 millions de dollars, soit tout de même une somme inférieure aux 31,3 millions de dollars engloutis en 2023.
SR010 Sifted Owkin job cuts: AI agents pivot forces restructuring at French biotech unicorn Employees and former employees... described 'huge managerial issues' and 'not knowing what to expect'.
SR011 EFPIA EFPIA Statement on the Use of AI in the Medicinal Product Lifecycle in the Context of the AI Act AI systems and models specifically developed and put into service for the sole purpose of scientific research and development are excluded from its scope.
SR012 Pharmaphorum Owkin bags $180M BMS alliance for cardiovascular clinical trials The deal is split between a Series B financing led by BMS and upfront fees, taking the total raised by the Paris, France-based start-up to $300 million, and also includes another $100 million in potential milestone payments.
SR013 Owkin Owkin FAQ — How much investment has Owkin received? Owkin has raised $304.1m USD from investors such as Bpifrance Large Ventures, GV, NJF Capital...
SR014 Owkin Owkin to Build AI Agents as Part of a Multi-Year K Pro Collaboration with Sanofi Owkin and Sanofi have collaborated since 2021 through a €90 million strategic partnership focused on target identification in oncology and patient subgrouping.
SR015 Yahoo Finance / Access Newswire Owkin Launches K Pro, the First Agentic AI Co-Pilot for Biopharma K Pro has already accelerated internal drug target identification by 70% (from 12+ months to 3 months).
SR016 Owkin Owkin to Build AI Agents as Part of a Multi-Year K Pro License Agreement with AstraZeneca Under the three-year licensing agreement, Owkin will lead the end-to-end development of AI agents to run on K Pro, integrated within AstraZeneca's IT infrastructure.
SR017 Precedence Research Federated Learning Market Size, Share, and Growth to 2035 The market faces challenges due to the lack of standardization and the technical complexity of federated learning systems. Data across various clients is often in a non-uniform state, called Non-IID data, which creates a challenge in model conversion and precision.
SR018 Revelio Labs Owkin Employee Headcount and Workforce Trends
SR019 Tracxn Owkin — Funding and Investors
SR020 Premier Alternatives Owkin — Secondary Market Intelligence
SR021 Wikipedia Owkin — Wikipedia
SR022 R&D World Sanofi Deepens Its Owkin Bet with a Five-Year Deal to Build Bespoke Drug Development Agents Sanofi is not placing on Owkin alone: in January, the company expanded a parallel collaboration with CytoReason, another agentic-AI vendor, in a deal valued at up to $16 million.
SR023 BusinessWire Owkin Creates New Spin-Out Waiv, Formerly Owkin Dx, With $33M Financing
SR024 Craft.co Owkin Executive Team
SR025 Clay Owkin Funding History
SR026 Signal Base Owkin Secures $320 Million in Funding
SR027 BioSpace Undeterred by Political, Economic Headwinds, Pharma Ups R&D Investment in 2024 and Beyond Drugmakers the world over pumped nearly $288 billion into research and development in 2024.
SR028 Owkin Owkin Expands Collaboration with Sanofi Leveraging AI for Drug Positioning in Immunology
SR029 R&D World How Much Does the Pharma Industry Spend on R&D? Probably More Than You Thought
SR030 Health Data Hub (France) Health Data Hub — Owkin Research Projects Registry
SR033 Owkin Owkin Newsroom — K Pro Platform
SR034 European Commission — AI Office European AI Office — Structure, GPAI Enforcement, and Mandate The AI Office also enforces the rules for GPAI models and supports the governance bodies in Member States in their tasks. This is underpinned by the powers given to the Commission by the AI Act, including the ability to conduct evaluations of GPAI models, request information and measures from model providers, and apply sanctions.
SR035 UK Information Commissioner's Office (ICO) Guidance on AI and Data Protection — Accountability, Transparency, and Fairness The Guidance on AI and Data Protection has been updated after requests from UK industry to clarify requirements for fairness in AI.
SV001 Recursion Pharmaceuticals Recursion Reports Fourth Quarter and Full Year 2025 Financial Results "Cash, cash equivalents and restricted cash were $753.9 million as of December 31, 2025... Total revenue was $74.7 million for the year ended December 31, 2025... cash operating expense of approximately $400 million."
SV002 CompaniesMarketCap Market capitalization of Recursion Pharmaceuticals (RXRX) As of June 2026 Recursion Pharmaceuticals has a market cap of $1.71 Billion USD.
SV003 Finerva BioTech & Genomics: 2026 Valuation Multiples "In Q1 2026, the median EV/Revenue Multiple for BioTech & Genomics companies sat at 4.4x."
SV004 Axis Intelligence AI Drug Discovery 2026: 173 Programs, FDA Framework and Market "AI biotech multiples: 8-12x revenue clinical-stage. 2025 VC: $5.7B (78% surge)."
SV005 Vision Life Sciences Biotech Funding and IPO Landscape 2026 "In the first week of February 2026, Eikon ($381M), Veradermics ($256M), and two additional biotechs collectively raised over $1.2 billion in just three days."
SV006 Forbes AI Drug Discovery Unicorn Insilico Medicine Jumps In Hong Kong Trading Debut "Shares of Insilico Medicine rose 24.7% in their Hong Kong stock market debut Tuesday, boosting its market cap to HK$16.7 billion ($2.1 billion)."
SV007 MedDeviceGuide Roche Acquires PathAI for $750M to Bolster AI-Driven Diagnostic Capabilities "Roche will pay a purchase price of US$750 million upfront and additional milestone payments of up to US$300 million, valuing the deal at US$1.05 billion. Target Total Funding: ~$355–660M. Target Revenue (est.): $100–250M."
SV008 Latham & Watkins LLP Latham & Watkins Advises PathAI on Acquisition by Roche "Under the terms of the agreement, Roche will pay a purchase price of US$750 million upfront and additional milestone payments of up to US$300 million, valuing the deal at US$1.05 billion."
SV009 HLTH Tempus Announces the Acquisition of Paige "Tempus AI has acquired Paige, a pioneer in AI-powered digital pathology, in a deal valued at $81.25 million."
SV010 StartupFortune Roche's $750 million PathAI deal shows diagnostics is the most credible near-term exit for vertical AI
SV011 Owkin Owkin to Build AI Agents as Part of a Multi-Year K Pro Collaboration with Sanofi "Owkin and Sanofi have collaborated since 2021 through a €90 million strategic partnership... This new collaboration represents the next evolution in the partnership."
SV012 Owkin Owkin becomes 'unicorn' with $180m investment from Sanofi and four new collaborative projects "Owkin is now a 'unicorn' — a startup valued at more than $1 billion — through a new $180 million investment from Sanofi."
SV013 Owkin Owkin announces multi-year clinical data science strategic collaboration with Bristol-Myers Squibb
SV014 Owkin How much investment has Owkin received? Owkin has raised $304.1m USD from investors.
SV015 Owkin Owkin Creates New Spin-out Waiv (formerly Owkin Dx) with $33M Financing
SV016 Sifted Owkin staff criticise 'huge managerial issues' as French unicorn moves to AI agents "Owkin staff publicly criticized the company's internal management during its pivot to agentic AI, flagging 'huge managerial issues', poor organisational transparency, and strategy shifts as poorly communicated."
SV017 Maddyness Owkin prévoit de se séparer d'une cinquantaine de salariés "Owkin a réalisé un chiffre d'affaires de 86 millions de dollars l'année dernière, en hausse de 83% rapport à 2023... elle a enregistré l'année dernière une perte financière de 5,6 millions de dollars... Owkin dispose encore en banque de 200 millions de dollars... cherche à boucler un nouveau tour de table d'un montant compris entre 200 et 300 millions de dollars."
SV018 PremierAlts Owkin — Private Market Data
SV019 Latka Owkin.com — Revenue, Customers, Growth
SV020 Tracxn Owkin — 2026 Company Profile and Funding Rounds
SV021 Crunchbase Owkin — Funding & Investors
SV022 MedCity News AI startup Waiv, spun out of Owkin, is bringing precision medicine testing to oncology
SV023 Pharmaphorum Owkin bags $180M+ BMS alliance for cardiovascular clinical trials
SV024 Wikipedia Owkin — Wikipedia
SV025 Clay How Much Did Owkin Raise? Funding & Key Investors
SV026 Benzinga Sanofi Owkin Expand AI Collaboration with New Agentic AI Initiative
SV027 European Biotechnology Owkin spins out AI diagnostics startup Waiv with $33M financing
SV028 World Pharma News Sanofi invests $180 million equity in Owkin's artificial intelligence
SV029 TechCrunch Owkin raises $180M in Series B led by Sanofi
SV030 Yahoo Finance Recursion Pharmaceuticals Overview
SV031 Awaira Owkin: $1B Valuation — Funding & Investors 2026
SV032 Craft.co Owkin Executives
SV033 Allsci Sanofi Expands Owkin Partnership for Purpose-Built AI Agents
SV034 Owkin Owkin Launches K Pro, the First Agentic AI Co-Pilot for Biopharma
SV044 Maddyness Thomas Clozel (Owkin): 'Dans 10 ans, nous aurons remplacé les Big Pharma'