Chai Discovery
蓝筹资本背书的 AI 抗体平台,估值 $1.3B;声称实现突破性基准表现,但没有同行评审验证,公开收入伙伴仅有一家
Chai Discovery 是技术上可信的 AI 抗体平台,背后有蓝筹投资人和标志性 Lilly 合作;但 $1.3B 估值完全押在未经验证的预印本基准、单一披露收入伙伴和零公开财务数据上,任何投入前都需要更深尽调。
封面要素
公司概况
Chai Discovery 是一家 AI 药物发现平台公司,2024 年初由四位来自 OpenAI、Meta FAIR、Stripe 和 Absci 的研究人员在 San Francisco 创立。不到两年,公司以 $1.3 billion 估值融资约 $230 million,并在 2026 年 1 月宣布与 Eli Lilly 达成标志性合作。公司采用双层产品:开源的 Chai-1 用于分子结构预测,自研的 Chai-2 用于从头设计抗体,目标是生物制药 R&D 工作流。基于公司撰写、仍待同行评审的预印本,Chai-2 声称在 52 个多样化靶点上实现约 16–20% 的从头设计抗体命中率(是此前计算方法的 100×)。公司以轻资产平台运作,没有内部药物管线、没有湿实验基础设施,也没有公开披露收入指标。
- 成立时间
- 2024-01-01
- 创始人
- Joshua Meier, Jack Dent, Matthew McPartlon, Jacques Boitreaud
- 创立地点
- San Francisco, California
- 总部
- San Francisco, California
- 产品
- Chai 采用双层产品架构。Chai-1 是开源多模态基础模型,用于生物分子结构预测(蛋白质、小分子、DNA、RNA、糖基化),2024 年 10 月以 Apache 2.0 发布,可通过网页和 PyPI 免费商用。Chai-2 是用于完全从头设计抗体的自研生成模型;输入只需要目标抗原和表位,就能从零生成所有 CDR,在 52 个多样化靶点上实现 16% 命中率(每个靶点 ≤20 个设计),超过 86% 的全长 mAb 满足与已获批疗法相当的可开发性标准。Chai-2 按 Responsible Deployment 政策控制访问,需要合作伙伴申请并获批。
- 客户
- 客户主要是希望加速早期生物药疗法发现的大型生物制药公司和生物技术公司,尤其是抗体方向。截至 2026 年 5 月,Eli Lilly 是唯一公开具名的商业伙伴,另有数量未披露的生物技术公司获得 Chai-2 早期访问。Menlo Ventures 称,「相当一部分生物技术行业」已经申请访问;尚未确认更多具名伙伴。
- 商业模式
- 公司采用平台即服务模式,并带有 freemium 结构:Chai-1 免费开源,用来积累开发者心智;Chai-2 则按 Responsible Deployment 政策,通过定制合作开放访问。收入预计来自前期平台授权费、定制模型开发费(例如 Lilly 交易中基于 Lilly 专有数据训练的模型),也可能来自临床候选药物推进的里程碑付款或 royalties。所有交易经济性均未公开。
- 阶段
- early-stage private (Series B)
- 融资情况
- 公司三轮累计融资约 $230 million:$30M 种子轮(2024 年 9 月,由 Thrive Capital、OpenAI、Dimension Capital 领投,估值约 $150M);$70M Series A(2025 年 8 月,由 Menlo Ventures/Anthology Fund 领投,估值约 $550M);$130M Series B(2025 年 12 月,由 Oak HC/FT 和 General Catalyst 共同领投,投后估值 $1.3B)。其他投资方包括 Lachy Groom、Yosemite、Neo、SV Angel、Emerson Collective 和 Glade Brook Capital。董事会成员包括 Mikael Dolsten(Pfizer 前 CSO)、Annie Lamont(Oak HC/FT)和 Hemant Taneja(General Catalyst)。
执行摘要
主要优势
- 突破性基准表现:Chai-2 声称在 52 个多样化靶点上 de novo 抗体命中率约 16–20%,比此前低于 0.1% 的计算领域最优水平高出 100× 以上;设计出的全长 mAbs 中,86%+ 达到已获批疗法的可开发性基准,并在 GPCR、肿瘤新抗原等历史难题靶点上实现功能成功。
- 一线创始团队,罕见跨学科深度:Meier(Meta FAIR ESM1 共同负责人、Absci AI 负责人)、McPartlon(Absci de novo 抗体设计)、Dent(Stripe 基础设施)和 Boitreaud(Aqemia ML)把前沿蛋白 AI 研究与大规模工程履历结合在一起。
- 顶级投资人和董事会构成:Oak HC/FT、General Catalyst、Thrive Capital、OpenAI 和 Menlo Ventures 参与三轮融资;董事 Mikael Dolsten(前 Pfizer CSO,150+ 个临床项目、36 个获批项目)以及投资人 Annie Lamont、Hemant Taneja,显示机构信心高,也能打开深度生物制药网络。
- Eli Lilly 合作验证平台质量:与全球前五药企开展定制模型合作,覆盖多个 biologic 靶点,并用 Lilly 专有数据训练;这是现阶段最严格商业验证,也削弱了“没有真实药企客户”的叙事风险。
- 轻资产、高潜在毛利模式:公司不自建湿实验室基础设施(相比 Generate:Biomedicines 的 140,000 sq ft),约 29 名员工即可运营;若平台靠合作交易规模化,软件经济性有望带来结构性更高毛利。
主要风险
- 基准证据完全由公司撰写且未经评审:所有 Chai-2 性能主张都来自 biorxiv 预印本,尚未经历同行评审;截至 2026-05-22,独立湿实验室复现实验包仍未发表。若后续可开发性筛查(聚集、稳定性、免疫原性、表达不佳)暴露公司实验未捕捉的缺口,复现风险会很大。
- 客户极端集中且没有财务披露:Eli Lilly 是唯一公开点名的商业伙伴;交易条款、ARR、合同数量和收入全部未披露,仅凭公开来源无法验证商业牵引力和财务健康度。
- 估值溢价要求多个正面结果同时兑现:在无披露收入情况下拿到 $1.3B 估值,Series C 投资人需要湿实验室验证、Lilly ARR 高于约 $15M、以及至少新增一家具名药企伙伴同时成立,才能支撑该估值;若三项都缺席,基于公开可比 DCF 分析,估值结构上应下调 30–50% 至 $700M–$900M。
- 没有临床概念验证,也缺监管路径清晰度:截至 May 2026,没有 Chai 设计分子进入 IND 或临床试验;FDA 没有为 AI 设计 biologics 提供专门审批路径;Chai 也未披露任何资产的制造伙伴、工艺规模或稳定性包。
- 关键人物集中,团队可见度有限:Joshua Meier 支撑技术身份、投资人关系和合作战略;约 29 人团队公开治理面有限;若 Meier 或其他联合创始人离开,技术路线图和合作伙伴 pipeline 都会明显受损。
- 竞争和生物安全逆风:Isomorphic Labs(融资 $600M、Alphabet 背书)、Generate:Biomedicines(估值 $1.9B、湿实验室一体化)、AbSci(上市公司)和开源替代方案(Boltz-2、ESMFold)都在争夺药企注意力和人才;此外,前沿蛋白设计 AI 的 dual-use 生物安全担忧,可能带来监管压力。
未决问题
- Eli Lilly 合作的收入、ARR、烧钱速度和交易经济性完全未披露;没有 data room access,就无法独立验证财务风险画像和隐含估值。
- Chai-2 de novo 抗体命中率和可开发性主张仍没有独立、同行评审的湿实验室复现;复现风险是科学尽调中最高优先级。
- Lilly 之外尚无其他具名 pharma 或 biotech 伙伴获确认;early-access partner pipeline 的规模和状态未知,客户集中度风险无法量化。
- 股权结构、优先股堆叠和稀释画像不公开;缺少这些信息,下行情景收益建模无法完成。
- 没有披露 IND 阶段项目、制造伙伴、CMC 包或监管沟通,无法确认任何 Chai 设计分子是否走在可信临床入组路径上。
- 估计员工数(约 29 人)和计算支出未经确认;烧钱速度估计($20–35M/year)置信度低,没有官方披露就不能用于评估现金跑道。
目录
01公司概况
1.1 身份、总部与使命
Chai Discovery 是一家人工智能公司,总部位于 California 州 San Francisco,2024 年初注册成立。公司自称使命是「把生物学从科学变成工程」,用前沿 AI 预测并重写生化分子之间的相互作用——蛋白质、抗体、核酸和小分子几乎支撑着所有生物过程。Chai 不建设自己的药物管线,而是以平台公司运作:开发 AI 基础模型,并部署给生物制药伙伴,让对方加速早期疗法发现。公司将愿景描述为打造分子的「计算机辅助设计套件」,类似 CAD 软件在机械和土木工程中的角色——把上游工作从蛮力实验筛选,转向有目的、由计算引导的设计。Chai 创立时,团队在 OpenAI 的 San Francisco 办公空间工作,反映出公司与更广泛 AI 研究社区的深度联系。四位联合创始人 Joshua Meier、Jack Dent、Matthew McPartlon 和 Jacques Boitreaud,在 Harvard 研究、OpenAI、Meta FAIR、Absci、Stripe 和 Aqemia 之间有长达十年的交集。OpenAI 是公司最早的种子轮投资方之一。Chai 按「Responsible Deployment」政策运营,选择性开放伙伴访问,而非全面商业开放。截至 2026 年初,据报道公司约有 29 名员工,但该数字未在公开披露中得到官方确认。 [CO001, CO002, CO003, CO004, CO005, CO006]
| 指标 | 数值 / 状态 | 日期 | 置信度 | 缺口 / 限制 |
|---|---|---|---|---|
| 创立时间 | Early 2024 | 2024 | 高 | 确切月份未公开确认 |
| 总部 | 美国加州旧金山 | 2026-05 | 高 | None |
| 累计融资 | >$225M(2026 年 1 月新闻稿约 $230M) | 2026-01 | 高 | 各来源之间存在轻微取整差异 |
| 最新轮次 | $130M Series B 轮 | 2025-12 | 高 | None |
| 估值(Series B) | $1.3 billion | 2025-12 | 高 | None |
| 员工数 | ~29 名员工(估计) | early 2026 | 低 | 官方未披露;由间接报道推导 |
| 收入 / ARR | 未公开披露 | 2026-05 | unknown | 非上市公司;无财务披露 |
| 进入临床阶段的分子 | 截至运行日没有 | 2026-05 | 高 | 仍处于临床前平台阶段 |
估值和融资金额来自公司新闻稿(官方),并由 TechCrunch 和 Bloomberg 佐证。员工数(~29)是基于间接报道的估计,置信度低。收入未披露。
[CO001, CO006, CO007, CO024, CO026, CO030]Chai 的创始团队、AI 平台、资本和合作伙伴如何连成商业价值。
[CO005, CO006, CO008, CO023, CO029, CO030]1.2 领导层、创始人与治理
Chai Discovery 的创始团队把深度 AI 研究能力、生物制药经验和大规模工程经验组合在一起。联合创始人兼 CEO Joshua Meier 2018 年曾在 OpenAI 研究和工程团队任职,之后加入 Meta FAIR,共同主导 ESM1 的开发——ESM1 是首个 transformer 蛋白语言模型,也是现代蛋白 AI 的基础前身。Meier 随后在 Absci 工作约三年,他与联合创始人 Matthew McPartlon 领导 AI 部门,率先开展从头设计抗体的早期研究,并贡献了如今进入临床试验的多个候选药物。联合创始人兼总裁 Jack Dent 曾任 Stripe 工程和产品负责人,具备构建高韧性大规模机器学习系统的经验;他和 Meier 最早在 Harvard 计算机科学课堂相识,后来由 OpenAI CEO Sam Altman 介绍潜在合作。Jacques Boitreaud 曾任 Aqemia AI 负责人,将面向小分子发现的 ML 工具推向生产。投资者把四位创始人的合作称为「酝酿了十年的伙伴关系」。治理层面,Mikael Dolsten 于 2025 年加入 Chai 董事会;他曾任 Pfizer 首席科学官,负责推进超过 150 个分子进入临床试验,并交付 36 款获批药物。2025 年 12 月 Series B 之后,Oak HC/FT 联合创始人兼管理合伙人 Annie Lamont,以及 General Catalyst 董事总经理 Hemant Taneja 也加入董事会。关键人物风险较高:Meier 支撑技术愿景和团队声誉,也是顶级投资者与合作伙伴信任的核心。 [CO011, CO012, CO013, CO014, CO015, CO016]
| 人物 | 职务 | 背景 | 创始人-市场匹配度 | 关键人依赖 |
|---|---|---|---|---|
| Joshua Meier | CEO 兼联合创始人 | OpenAI(2018 年研究 / 工程),Meta FAIR(共同负责 ESM1),Absci(AI 负责人,从头抗体设计) | 深厚 AI 与药物发现经验;蛋白质语言模型先行者 | 极高——技术愿景、投资人关系和 CEO 品牌 |
| Jack Dent | 总裁兼联合创始人 | Stripe(工程和产品负责人,大规模 ML 系统);Harvard CS(在此结识 Meier) | ML 系统规模化和产品商业化 | 高——业务拓展、合作伙伴和公司运营 |
| Matthew McPartlon | CTO 兼联合创始人 | Absci(与 Meier 同任 AI 负责人;从头抗体设计管线进入临床) | 抗体设计生成式 AI 实操经验;临床转化经验 | 高——模型架构和核心 R&D |
| Jacques Boitreaud | 联合创始人 | Aqemia AI 负责人(小分子发现 ML) | 生成式 ML 用于分子设计 | 中——模型开发;四位创始人之一 |
| Mikael Dolsten | 董事 | 前 Pfizer 首席科学官;负责推动 150+ 个分子进入临床试验、36 款药物获批 | 大型药企监管、临床和商业化经验 | 顾问型——董事会层面的战略和科学指导 |
角色和背景经 BusinessWire 新闻稿、TechCrunch、The Pharmaletter 等多个来源核验。根据 Series A 公告,Dolsten 于 2025 年加入董事会。关键人依赖为定性推断,依据公开的角色描述和投资人评论。
[CO011, CO012, CO013, CO014, CO015, CO016]1.3 融资历史与投资者版图
对一家成立两年的公司来说,Chai Discovery 的融资节奏异常快。2024 年 9 月,成立约六个月后,公司完成 $30 million 种子轮,由 Thrive Capital、OpenAI 和 Dimension Capital 领投,估值约 $150 million。种子轮提供了初始现金跑道,并支持 Chai-1 发布。2025 年 8 月,公司完成 $70 million Series A,由 Menlo Ventures 通过 Anthology Fund 领投;Anthology Fund 是 Menlo 与 Anthropic 的联合基金。新投资方包括 Yosemite、DST Global Partners、SV Angel、Avenir 和 DCVC,老股东 Thrive Capital、OpenAI 和 Dimension 也继续参与。Series A 将 Chai 估值推至约 $550 million,累计融资约 $100 million。仅四个月后,2025 年 12 月,Chai 完成 $130 million Series B,由 Oak HC/FT 和 General Catalyst 共同领投,估值 $1.3 billion,成为独角兽。Series B 新投资方包括 Emerson Collective(Laurene Powell Jobs)和 Glade Brook Capital。Series B 后累计融资超过 $225 million;Chai 2026 年 1 月新闻稿称公司「迄今已融资近 $230M」,与其他公开披露存在轻微四舍五入差异。投资者组合值得注意:前沿 AI 投资方(OpenAI、Thrive、Menlo/Anthology)与医疗健康成长股权投资方(Oak HC/FT)和转型导向风投(General Catalyst)同时在场,说明技术和商业化两个维度都获得了信任。 [CO020, CO021, CO022, CO023, CO024, CO025]
| 利益相关方 | 角色 | 参与轮次 | 披露金额 | 战略重要性 | 尽调问题 |
|---|---|---|---|---|---|
| Thrive Capital | 种子轮领投方;连续投资人 | 种子轮、Series A 轮、Series B 轮 | 未披露 | 先发投资人;连续参投体现信心;拥有 AI 组合的 VC | 确认各轮按比例跟投权和治理角色 |
| OpenAI | 种子轮投资人;连续投资人 | 种子轮、Series A 轮、Series B 轮 | 未披露 | 战略 AI 伙伴;人才管道;Sam Altman 与创立故事相连 | 厘清股权之外的战略协议范围;是否有数据或算力安排? |
| Dimension Capital | 种子轮投资人;连续投资人 | 种子轮、Series A 轮、Series B 轮 | 未披露 | AI 聚焦的深科技 VC;长期财团锚点 | 确认治理参与情况 |
| Menlo Ventures (Anthology Fund) | Series A 轮领投方 | Series A 轮 | 领投 $70M 轮次 | Anthology Fund 是 Menlo / Anthropic 联合设立的投资载体;押注 AI + 生物学 | 评估 Anthropic 战略重叠和潜在竞争张力 |
| Oak HC/FT | Series B 轮共同领投方;董事席位 | Series B 轮 | 共同领投 $130M 轮次 | 医疗 + 金融科技成长股权;Annie Lamont 进入董事会 | 确认董事会构成、保护性条款和按比例跟投权 |
| General Catalyst | Series B 轮共同领投方;董事席位 | Series B 轮 | 共同领投 $130M 轮次 | 聚焦转型的 VC;Hemant Taneja 进入董事会;2027 年临床试验论点 | 确认董事会构成、否决权和商业承诺 |
| Yosemite (Reed Jobs) | Series A 和 B 轮投资人 | Series A 轮、Series B 轮 | 未披露 | 聚焦肿瘤学;Reed Jobs 基金;连续支持者 | 理解肿瘤学战略覆盖如何匹配 Chai 平台逻辑 |
| Emerson Collective(Laurene Powell Jobs 关联) | Series B 新投资人 | Series B 轮 | 未披露 | 使命导向的影响力投资;Series B 新进 | 评估治理诉求和使命一致性条件 |
投资人名称和参与轮次来自公司新闻稿(BusinessWire),并由 TechCrunch、Bloomberg 和 Observer 报道确认。各轮出资规模未公开披露。DST Global Partners、SV Angel、Avenir、DCVC、Neo、Lachy Groom、Fred Ehrsam 和 Glade Brook 也是已披露投资人, 但因覆盖不完整未纳入本表;见证据缺口 EG-investor-coverage。
[CO020, CO022, CO024, CO025, CO027, CO028]截至 2026 年 5 月,Chai Discovery 的关键性能和技术指标。
命中率数字来自公司撰写、尚未同行评审的预印本。估值和融资来自官方新闻稿。
[CO030, CO033, CO035, CO045, CO026, CO024]1.4 技术平台与产品主张
Chai 的平台围绕两代基础模型构建。Chai-1 于 2024 年末作为开源模型发布,通过在生物分子结构预测中实现最先进表现,迅速在研究社区建立声誉——某些类别的表现大致持平或超过 Google DeepMind 的 AlphaFold 基准。Chai-2 于 2025 年 6 月 30 日发布,是公司的旗舰产品,声称在抗体设计上带来根本性跃迁。模型只需要输入目标抗原和表位,就能在 zero-shot 场景下从零生成所有互补决定区(CDR)——不需要模板、大规模筛选或既有实验样本。Chai 在 2025 年 6 月提交给 bioRxiv 的预印本中报告称,模型在 52 个多样化抗体靶点上的命中率为 16%,每个靶点提示少于 20 个设计,并在不到两周内完成湿实验验证;约一半靶点(26/52)至少产出一个已验证命中。2025 年 11 月的配套预印本把结果延伸到全长单克隆抗体,报告称超过 86% 的 Chai-2 设计呈现强可开发性特征——热稳定性、表达、纯度、人源性——可与已获批疗法相比。同一预印本还展示了功能性 GPCR 激动作用和对肿瘤特异性新表位的选择性结合,这两类靶点历来难以成药。公司还报告称,小蛋白结合物设计的湿实验成功率为 68%。Chai 声称,一个传统 R&D 需要投入超过 $5 million、耗时三年以上的药物发现挑战,被其在数小时内用计算解决,并在不到两周内完成实验室验证。上述公开结果均为公司撰写的预印本,截至运行日期尚未经过正式同行评审。 [CO029, CO030, CO031, CO032, CO033, CO034]
1.5 里程碑、合作与反向背景
Chai 在前两年的里程碑节奏异常快。公司从 2024 年 9 月拿到种子轮,到 2025 年 12 月成为独角兽,全程约 15 个月。Chai-1 和 Chai-2 的发布,让公司很快在研究社区和生物制药行业获得技术可信度。2026 年 1 月 8 日,Chai 宣布与全球最大制药公司之一 Eli Lilly 合作:Lilly 将在多个药物靶点上部署 Chai 的 AI 平台,Chai 则会开发一个专用模型,仅用 Lilly 专有的大规模数据集训练。至少一位分析师将该合作描述为生物技术领域最大的 AI 软件交易之一;财务条款未公开。General Catalyst 预计,像 Chai 这类工具的早期制药采用者,可能会在 2027 年底前看到 first-in-class 分子进入临床试验。 动能之外,投资者和分析师也指出了实质性反向背景。LoonBio 发布的一份详细行业分析显示,自 2015 年以来,更广泛的 AI 药物发现行业吸引了超过 $60 billion 风险投资,但截至 2025 年初仍未产出一款获得 FDA 批准的 AI 设计药物。Exscientia、BenevolentAI、Recursion 等高知名度第一代 AI 生物技术公司,都经历了重大临床失败、管线缩减和股价崩跌。批评者质疑,早期计算命中率高能否转化为临床成功,也质疑 Chai 在没有临床阶段资产的情况下,能否支撑 $1.3 billion 估值。Chai 的 Chai-2 性能数据建立在公司撰写、尚未同行评审的预印本之上。截至运行日期,没有任何 Chai 设计的分子进入人体试验。Chai 的直接竞争对手包括 Isomorphic Labs——由 Nobel laureate Demis Hassabis 领导的 Alphabet 子公司,2025 年 3 月融资 $600 million——以及 Formation Bio 和 Reid Hoffman 支持的 Manas。 [CO038, CO039, CO040, CO041, CO042, CO043]
| 日期 | 事件 | 类型 | 金额 / 估值 / 状态 | 参与方 | 影响 |
|---|---|---|---|---|---|
| Early 2024 | 公司在旧金山创立 | 创立 | — | 创始人:Joshua Meier、Jack Dent、Matthew McPartlon、Jacques Boitreaud | 由合作十年的协作网络组建;早期在 OpenAI 旧金山办公室办公;OpenAI 在种子轮投资 |
| Sep 2024 | $30M 种子轮完成 | 融资 | $30M 融资;估值约 $150M | Thrive Capital(领投)、OpenAI、Dimension Capital、Amplify Partners | 验证团队可信度;为 Chai-1 开发提供资金;让 Chai 早早进入顶级 AI VC 财团 |
| Late 2024 | Chai-1 开源发布 | 产品 | — | Chai Discovery | 开源分子结构预测模型;基准表现达到 SOTA;建立研究社区可信度 |
| Jun 30, 2025 | Chai-2 发布;从头抗体设计取得突破 | 产品 | 声称命中率约 16–20% | Chai Discovery | 首个实验命中率达到两位数的零样本平台;较此前计算方法提升 100×;开放早期合作伙伴访问权限 |
| Aug 2025 | $70M Series A 轮;Mikael Dolsten 加入董事会 | 融资 | $70M 融资;估值约 $550M;累计约 $100M | Menlo Ventures/Anthology Fund(领投)、Yosemite、DST Global、SV Angel、Avenir、DCVC、Thrive、OpenAI、Dimension | Mikael Dolsten(前 Pfizer CSO)加入董事会;Series A 与 Chai-2 发布公告绑定 |
| Nov 2025 | 挑战性靶点预印本发表 | 产品 | >86% 全长 mAbs 可开发性 | Chai Discovery(公司撰写的 bioRxiv 预印本) | 将 Chai-2 扩展到具备药物属性的全长单克隆抗体;展示 GPCR 激动和新抗原表位选择性;尚未经过同行评议 |
| Dec 15, 2025 | $130M Series B 轮;达到独角兽估值 | 融资 | $130M 融资;$1.3B 估值;累计 >$225M | Oak HC/FT + General Catalyst(共同领投)、Thrive、OpenAI、Dimension、Menlo、Emerson Collective、Glade Brook、 Yosemite | 进入独角兽行列;Annie Lamont(Oak HC/FT)和 Hemant Taneja(GC)加入董事会;为商业化阶段补足资金 |
| Jan 8, 2026 | 宣布与 Eli Lilly 合作 | 合作 | 财务条款未披露 | Chai Discovery + Eli Lilly | 首个具名药企伙伴;基于 Lilly 专有数据的定制模型;验证商业牵引力和平台适用性 |
| May 2026(报告运行日) | 没有 Chai 设计的分子进入人体试验 | 负面 | 临床阶段分子为零 | — | 平台仍处于临床前;计算性能与临床部署之间的转化缺口尚未解决 |
日期和事件细节主要来自 BusinessWire 新闻稿,并由 TechCrunch、Bloomberg 和 FierceBiotech 佐证。预印本由公司撰写,尚未经过同行评议。种子轮估值(约 $150M)来自二手报道,置信度中等。里程碑类型分类遵循 founding|financing|product|scale|regulatory|partnership|governance|adverse schema。
[CO001, CO020, CO022, CO024, CO029, CO030]Chai Discovery 从成立到 2026 年 5 月的里程碑,展示融资、产品和合作事件。
成立和 Chai-1 发布的时间线日期使用近似季度锚点;具体日期未公开确认。
[CO001, CO010, CO020, CO024, CO029, CO030]1.6 展示材料
02市场分析
2.1 市场边界与范围
精确定义市场很重要,因为分析师估算会随「AI 药物发现」的窄口径或宽口径相差一个数量级。最窄定义由 Axis Intelligence 和 Grand View Research 使用,只覆盖在 IND 申报前,对靶点识别、命中物和先导物生成、先导物优化、从头分子设计以及临床前候选物选择提供实质支持的 AI 软件和相关服务。这个定义排除了用于临床试验运营、监管申报、药物警戒、制造和商业分析的 AI。按这个窄口径,2026 年市场规模为 $2–5 billion。Mordor Intelligence 使用的中等定义加入 AI 赋能的制剂和药物再利用分析,2026 年规模为 $4.36 billion。Global Market Insights 和 Towards Healthcare 采用最宽定义,把触及整个制药价值链的所有 AI 软件都纳入——临床试验、制造、患者匹配——得到 2026 年 $24.51 billion。对 Chai Discovery 来说,相关边界是最窄定义:Chai-2 平台瞄准实验筛选前的早期生物药设计(抗体、纳米抗体、小蛋白),而不是临床或制造 AI。相邻但排除在外的支出类别包括:不依赖 AI-native 设计的 CRO 生物信息服务、传统湿实验抗体发现(杂交瘤、噬菌体展示)以及 AI 临床试验招募工具。主要现状替代方案是传统抗体发现方法(杂交瘤和噬菌体展示,2024 年占更广泛抗体发现市场的 38.1%)、手工计算 docking 流水线,以及制药公司完全自建的内部 AI 团队。全球制药 R&D 基数约为 2025 年 $300 billion 年支出,构成需求背景,但 AI 药物发现软件目前只占其中约 1.5%。 [CM001, CM002, CM003, CM004, CM005, CM007]
| 细分 / 类别 | 纳入支出 | 排除支出 | 买方 / 付款方 | 对 Chai 的意义 |
|---|---|---|---|---|
| AI 药物发现(狭义) | IND 前靶点识别、苗头 / 先导物生成、先导物优化、从头设计软件 | 临床 AI、制造 AI、药物警戒 | 药企 R&D、生物科技创始人 | 核心市场;最直接可比的估算 |
| AI 制药 R&D(中等范围) | 狭义范围 + 重定位分析、制剂 AI、临床前预测服务 | 临床试验运营、监管申报 | 药企 R&D、CRO | 扩大 TAM,但模糊 Chai 的竞争边界 |
| AI 制药(广义生态) | 制药价值链中的全部 AI 软件,包括临床运营、制造和商业化 | 硬件、耗材、无 AI 的湿实验室服务 | 所有制药细分 | 抬高 TAM;GMI / Towards Healthcare 估算采用该口径 |
| 抗体发现(所有方法) | 基于 AI 的设计、杂交瘤、噬菌体展示、B 细胞工程、转基因小鼠 | 小分子药物发现 | 药企生物制剂部门、生物科技公司 | Chai 的直接子细分;2026 年 $10.75B |
| AI 抗体设计(从头设计) | 无需实验种子的纯 AI 平台,用于从头生成抗体序列 | AI 辅助亲和力成熟、现有抗体优化 | 前沿药企、AI-native 生物科技 | Chai 的核心市场;最小但增长最快的子切片 |
边界定义造成 2026 年估算出现 9x 差距($1.94B–$24.51B)。Chai 竞争的是最狭义类别。相邻类别代表未来扩张路径或买方预算语境。
[CM003, CM010, CM011, CM037]三层规模测算,从全球药企 R&D 支出到 de novo 生物制品 AI 设计子细分,展示 Chai Discovery 的可触达市场。
TAM 数字来自 New Market Pitch 引用的全球药企 R&D 支出。SAM 中点来自 Mordor Intelligence AI 药企 R&D 市场($4.36B,2026)。SOM 是分析师基于 Chai 商业化前阶段、伙伴选择性和抗体设计重点估算的近期可获取市场;并非来自单一报告。
[CM001, CM002, CM005]2.2 市场规模测算:TAM、SAM、SOM 与相互矛盾的估算
Chai 的总可用市场(TAM)可以分两层来看。最宽层面,全球制药 R&D 支出约 $300 billion,代表更好的发现工具最终可能替代或增强的全部支出。更实际地看,Statista 估计 2026 年管线接近 23,000 个在研候选药物,来自超过 7,000 家公司——这些都是 AI 辅助分子设计的潜在受益者。可服务市场(SAM)是其中流向 AI 发现软件和生物药平台的支出子集。以 Mordor Intelligence 对 AI 制药 R&D 的 $4.36 billion(2026 年)作为最宽可信商业估算,再与 $10.75 billion(2026 年)的抗体发现市场交叉校验,Chai 的 SAM——聚焦生物药的 AI 设计——可估计为 2026 年约 $1–2 billion,并会随从头设计能力成熟而上升。可获取市场(SOM)要小得多:Chai 仍处商业化前,依赖选择性伙伴访问,并与既有玩家竞争;对 2026–2028 规划期而言,$100–300 million 的早期部署收入 SOM 属于保守且合适的估算。分析师对窄口径 AI 药物发现市场的估算相差 9 倍,主要由定义范围驱动。Grand View Research 将 2025 年市场定为 $2.35 billion(到 2033 年 CAGR 24.8%)。Axis Intelligence 将其收窄至 $1.94 billion(2025 年),CAGR 27%。Fortune Business Insights 报告为 $4.46 billion(2025 年),CAGR 更保守,为 12.2%。最乐观报告(Global Market Insights/Towards Healthcare)把临床 AI 纳入后,给出 2026 年 $24.51 billion;这个数字高估了纯发现平台的相关市场。相互矛盾的 CAGR 估算(11.3%–32.25%)大体跟随范围差异:窄范围 CAGR 较低,宽范围吸收了增长更快的临床分部。 [CM001, CM004, CM005, CM006, CM007, CM008]
| 发布方 | 地理范围 | 2025 年估算 | 2026 年估算 | 预测终年 | CAGR | 方法 / 范围 | 置信度 | 局限 |
|---|---|---|---|---|---|---|---|---|
| Grand View Research | 全球 | $2.35B | ~$2.91B | $13.77B (2033) | 24.8% | 狭义平台:靶点识别、优化、重定位软件 | 中高 | 不含临床 AI;CAGR 相比 Fortune 偏乐观 |
| Mordor Intelligence(AI Pharma R&D 口径) | 全球 | $3.30B | $4.36B | $17.66B (2031) | 32.25% | 中等范围:包括服务、制剂、重定位 | 中高 | 比纯发现类业务更宽;相对 GVR 抬高 TAM |
| Fortune Business Insights | 全球 | $4.46B | $5.00B | $12.56B (2034) | 12.2% | 中等范围:小分子和大分子软件,范围适中 | 中 | CAGR 最保守;可能低估 AI-native 初创公司收入 |
| Business Research Insights | 全球 | N/A | $2.68B | $8.67B (2035) | 13.95% | 狭义-中等范围:面向药企和生物科技的药物发现 AI | 低中 | 方法透明度有限 |
| Axis Intelligence | 全球 | $1.94B | $2.6–2.8B | $16.49B (2034) | ~27% | 狭义:仅纳入已验证 AI 参与的 pre-IND 项目 | 中 | 已验证样本组较小;排除自报项目 |
| Towards Healthcare / GMI | 全球 | $19.89B | $24.51B | $160.49B (2035) | 23.22% | 广义:整个 AI 赋能制药价值链 | 低 | 把药物发现 AI 与临床和制造 AI 混在一起;高估 Chai 的适用市场 |
| Mordor Intelligence (Antibody Discovery) | 全球 | $9.09B | ~$10.0B | $15.45B (2030) | 11.3% | 所有抗体发现方法(杂交瘤 38.1%、AI/ML、噬菌体展示) | 高 | 包括非 AI 方法;Chai 只竞争 AI/ML 子细分 |
| ResearchAndMarkets(AI Protein Structure Prediction 口径) | 全球 | $1.80B | $2.33B | $6.62B (2030) | ~30% | AI 蛋白质结构预测软件和服务 | 中 | 与从头抗体设计市场相邻,但并不相同 |
2026 年估算从 $1.94B(狭义)到 $24.51B(广义)不等。四个最可信的中点估算集中在狭义至中等范围的 $2.68B 至 $5.00B。Chai 的竞争版图更接近 $2–5B 区间。抗体发现($10.75B)是最贴近的相邻市场。
[CM004, CM005, CM006, CM007, CM008, CM009]五份分析师报告给出的 2026 年 AI 药物发现市场规模低—高区间,并解释每个边界的口径。
数值以 $B(2026)计。GVR 2026 年数字由 2025 年基数($2.35B)按 24.8% CAGR 外推。其他数值均来自已发布报告。$24.51B 数字不能与窄口径估计比较,应视为定义口径离群值。Chai 竞争于 $2–5B 的窄到中口径区间。
[CM004, CM008, CM009, CM005, CM007, CM010]2.3 买方、用户与付款方分层
AI 药物发现平台的主要买方是制药公司和生物技术公司,两者合计占 2025 年 AI 制药 R&D 支出的 59.45%。在这一群体中,大型制药公司(年 R&D 预算超过 $8 billion,例如 Eli Lilly、AstraZeneca、Roche 和 Novartis)支出最高,但采用外部新平台较慢。它们通常通过研究合作、股权投资或里程碑驱动的授权协议接触 AI 发现公司;Eli Lilly 仅在 2026 年 Q1 签署的 AI 药物发现交易总额就超过 $3.75 billion,正是例证。中型和专科制药公司往往每年向 AI 药物发现分配 $10–50 million,主要通过里程碑驱动合作,而不是大额前期承诺。AI-native 生物技术公司是增长最快的买方分部,AI 采用率比大型制药公司高 73%。它们在整个管线中部署 AI,也是重要的转介绍和验证渠道。传统生物技术公司更挑剔,通常只运行一两个 AI 试点。合同研究组织(CRO)是按 CAGR 计算增长最快的终端用户分部(到 2031 年为 33.15%),因为它们整合 AI 能力,为制药客户提供扩展后的发现服务。高校和研究机构是增长中的分部,但不是主要商业预算持有者。预算负责人通常是 Discovery Biology 负责人或首席科学官,采购中也越来越多涉及首席 AI 官和平台架构师。采用触发点通常是:某一具体靶点类别上,传统方法已经失败或太慢,同时专利悬崖时间表带来压力。 [CM026, CM027, CM028, CM029, CM022]
| 细分 | 买方 | 用户 | 付款方 | 典型年度预算 | 采用触发点 | 交易结构 |
|---|---|---|---|---|---|---|
| 大型药企 | R&D 副总裁 / CSO | 发现生物学家、计算化学家 | R&D 预算(总计 $8B+) | AI 项目 $100–500M+ | 专利悬崖压力;传统方法在特定靶点失败 | 多年平台合作,附带里程碑;股权投资 |
| 中型 / 专科药企 | CSO / 发现负责人 | 发现团队、外部 CRO | R&D 预算($500M–$3B) | 每年 $10–50M | 特定管线缺口;寻求更快的先导物生成 | 基于里程碑的授权;按项目收费的合作 |
| AI 原生生物技术公司 | 创始人 / CTO | 内部 AI 与生物学团队 | 风险投资支持的运营预算 | $10–100M(内部或合作伙伴平台) | 核心商业模式;采用新平台最快 | 平台授权;共同发现;数据共享合作 |
| 传统生物技术公司 | CSO / 生物学负责人 | 实验室科学家、生物信息学 | Series B/C 轮风险资本 | $1–20M(选择性试点) | 验证单个高难靶点;加快管线 | 试点项目;按设计付费模式 |
| 合同研究组织(CRO) | 商务拓展 / R&D 负责人 | 药物发现服务团队 | 药企客户合同 | $5–30M 用于 AI 基础设施 | 差异化竞争;药企客户对 AI 的需求 | 内部能力建设;从 AI 供应商获得平台授权 |
2025 年,药企和生物技术公司占 AI 药物研发支出的 59.45%。CRO 增长最快,年复合增长率(CAGR)为 33.15%。 大型药企($100-500M+ 支出)是价值最高的细分市场,但采用外部新平台最慢。 AI 原生生物技术公司采用速度比大型药企快 73%。
[CM026, CM027, CM028, CM029]五个主要 AI 药物发现市场细分中的买方—用户—付费方关系和采用画像。
预算数字来自 Business Research Insights 调研和 Mordor 分项数据。采用速度为定性判断,来自 CAS Life Sciences Summits 2025 和 AllAboutAI 采用统计。Chai 适配度为分析师推断,依据是当前 Lilly 合作和平台阶段。
[CM026, CM027, CM028, CM022]2.4 增长驱动因素与采用约束
AI 药物发现采用的首要结构性驱动,是 R&D 生产率危机。开发一个新的分子实体平均成本为 $2.8 billion(资本调整后),耗时 12–15 年,并承受约 90% 的 Phase I 失败率。2024 至 2030 年间,年美国收入超过 $180 billion 的药物将面临专利独占期丧失,董事会层面急需比传统时间表更快地补上收入。AI 能带来可衡量的效率提升:在特定工作流中,它可以把临床前开发周期从 5–6 年压缩到 12–18 个月,并将开发成本降低 25–50%。IQVIA Global R&D Trends 2026 报告给出最权威的行业信号:AI 赋能的新兴生物制药项目 Phase I 成功率达到 75%,传统项目为 40–65%;这一结果在 EBP 分部可见,即便尚未扩展到全行业。AlphaFold 映射 200M+ 蛋白结构,是基础性使能因素;FDA 在 2024 年对 12 款 AI 识别的肿瘤药物给予 fast-track designation,也释放了监管接纳信号。2026 年 1 月,FDA 和 EMA 联合发布药物开发中 AI 实践的 10 条指导原则,降低了买方的监管不确定性。风险投资也验证了行业:2025 年投入 $5.7 billion,较 2024 年增长 78%。采用约束同样关键。最主要障碍不是数据量,而是数据就绪度:组织拥有足够数据,却难以完成清洗、语境化,并对齐具体发现问题;这是 CAS Life Sciences Summits 在 2025 年的发现。尽管投入很高,只有 22% 的生命科学领导者成功规模化 AI。切换成本高:整合 AI 平台需要深入连接数据流水线、重新培训员工适应新工作流、按监管标准验证模型输出,并推动化学和生物部门的变革管理。模型可解释性仍是约束——监管方和实验室科学家需要可解释输出,黑箱深度学习模型在内部会遭遇怀疑。人才缺口(AI + 生物学复合能力)和预算限制卡住了较小生物技术公司。也许最重要的是:截至 2026 年 5 月,没有 AI 设计药物获得 FDA 批准,意味着买方购买的是尚未被证明的长期结果,这会拉长决策周期并限制前期合同规模。 [CM017, CM018, CM019, CM020, CM021, CM023]
| 因素 | 类型 | 方向 | 时间窗口 | 对 Chai 的影响 | 尽调要点 |
|---|---|---|---|---|---|
| R&D 成本危机($2.8B/NME,90% 失败率) | 驱动因素 | 正向 | 当前且持续 | 更快、更便宜药物发现的结构性需求——Chai 的核心价值主张 | 跟踪现有合作伙伴验证的平台 ROI,并与传统周期对比 |
| 专利悬崖(2030 年前 $180B+ LOE) | 驱动因素 | 正向 | 2024–2030 年紧迫窗口 | 大型药企董事会优先推动大规模管线提速 | 监测 Chai 与 2026–2028 年专利到期公司之间的交易管线 |
| IQVIA:AI EBP 的 Phase I 成功率为 75% | 驱动因素 | 正向 | 已验证的 2022–2025 队列 | 临床验证信号增强买方对 AI 平台的信心 | 评估 Chai 合作伙伴是否属于呈现该优势的 EBP 细分群体 |
| FDA/EMA 联合 AI 指南(2026 年 1 月) | 驱动因素 | 正向 | 监管清晰度立即提升 | 降低投资人和药企合作伙伴对 AI 设计资产的不确定性 | 核验 Chai 技术报告是否符合新的 FDA 可信度框架 |
| AlphaFold:已映射 200M+ 结构 | 驱动因素 | 正向 | 基础层(2020–2024) | 结构生物学底座支撑 AI 原生抗体设计管线 | 评估 Chai-2 的结构预测能力,对比 AlphaFold 3 / Isomorphic |
| VC 激增:2025 年 $5.7B(增长 78%) | 驱动因素 | 正向 | 2025–2026 年峰值 | 平台估值处于高位;Chai 的 $1.3B 估值符合市场水平 | 监测 2026 年回调是否压低交易量 |
| 数据就绪瓶颈 | 约束 | 负向 | 持续存在;需要 2–4 年修复 | 没有经过整理的自有数据集,合作伙伴无法充分使用 AI 平台 | 评估 Chai 如何将合作伙伴数据就绪度作为售前或入驻要求处理 |
| 仅 22% 生命科学领导者已规模化 AI | 约束 | 负向 | 当前(2025–2026) | 显示组织阻力普遍存在;限制部署速度 | 获取 Lilly 合作中内部部署成功的案例证据 |
| 尚无 AI 药物获得 FDA 批准(2026 年 5 月) | 约束 | 负向 | 首个批准前持续存在 | 限制合同规模,也压低把整条管线押注在 AI 平台上的意愿 | 跟踪 Insilico Medicine rentosertib Phase IIb(最接近获批的 AI 药物) |
| 切换成本与组织惯性 | 约束 | 负向 | 高;需要 2–4 年投入 | 需要多年集成;买方谨慎,避免锁定在早期平台上 | 了解 Lilly 交易期限和排他条款;评估切换条款 |
截至 2026 年 5 月,增长驱动因素与采用约束大致均衡。来自 IQVIA 的临床验证信号(Phase I 成功率 75%) 和监管清晰度,是短期最重要的正向催化。尚无 FDA 批准的 AI 药物,以及数据就绪瓶颈,仍是两个最显著逆风。
[CM017, CM018, CM019, CM024, CM025, CM031]四阶段采购与部署漏斗,展示药企从认知 AI 平台到触发主动合作里程碑的层层流失。
漏斗数量为指示性估算。69% 采用率来自 AllAboutAI 2026 报告;22% 规模化数字来自 Business Research Insights 和 Ardigen 引用的行业调研。尚无 AI 药物走到审批通过关口。
[CM021, CM022, CM032]2.5 AI 抗体设计子赛道
Chai Discovery 的核心方向——完全从头设计 AI 抗体——位于两个市场的交汇处:更广泛的抗体发现市场(2025 年 $9.78 billion,2026 年增至 $10.75 billion)以及该市场中由 AI/ML 赋能的子分部;后者 2025–2030 年 CAGR 为 22.4%,高于整体抗体市场的 10.1%。抗体发现市场涵盖识别治疗性抗体候选物的所有方法:杂交瘤技术(2024 年 38.1% 份额)、噬菌体展示、B 细胞工程、转基因小鼠平台,以及 AI/ML in-silico 设计。AI/ML 分部装机基础最小,但增长最快,驱动因素是 AI 平台可以把命中物识别周期从数月压缩到数周,并降低下游可开发性测试中的流失。制药和生物制药公司占 2024 年抗体发现支出的 48.3%;生物技术初创公司则以 14.8% CAGR 推进。North America 在 2024 年占据 41.5% 市场,Asia-Pacific 以 13.5% CAGR 增长最快。在 AI 抗体子赛道中,关键差异点是从头设计,而不是对实验发现先导物做 in-silico 优化。Chai 的 Chai-2 平台瞄准从头设计用例——只给定目标蛋白,不给实验种子分子,就从零设计抗体。这个问题明显比序列优化或亲和力成熟更难,相比 AI 辅助筛选,它也对应更高价值主张(以及更高采用门槛)。AI 蛋白结构预测市场是紧密相关的使能层,估计 2025 年为 $1.8 billion,2026 年以约 30% CAGR 增至 $2.33 billion,并在 2030 年达到 $6.62 billion。这一结构生物学基础设施支撑所有 AI-native 抗体设计平台,也是 Chai 与竞争对手竞争的计算基础。Chai 在这一子赛道中的 SAM 受限于当前伙伴访问模式,以及从头设计抗体市场作为独立商业类别仍处早期。 [CM011, CM012, CM013, CM014, CM002]
2.6 展示材料
03竞争格局
3.1 竞争格局概览
截至 2026 年 5 月,AI 药物发现竞争格局包含五类功能上不同的竞争者,各自瞄准相互重叠但有差异的买方需求。第一类是 AlphaFold 谱系的结构预测和通用 AI 平台——主要是 Isomorphic Labs(Alphabet spinout),其在 2024 年 11 月 Series A 融资约 $600 million,并把 AlphaFold 衍生模型用于小分子药物设计。Isomorphic 主要聚焦小分子,公开伙伴包括 Eli Lilly 和 Novartis,与 Chai 生物药优先的方向不同,但争夺同一制药 R&D 预算和伙伴关系。第二类是全栈 AI 药物发现编排平台——Recursion Pharmaceuticals(NASDAQ: RXRX)于 2025 年完成收购 Exscientia,形成覆盖基于 phenomics 的靶点发现到 AI 化学候选物设计的端到端平台;Insilico Medicine 则运营 Pharma.AI 平台,覆盖靶点识别(Biology42)、分子设计(Chemistry42)和临床优化(Medicine42)。Recursion 通过 BioHive-2 机器人基础设施生成超过 50 petabytes 专有实验数据,构成行业内最强的数据护城河。第三类是面向蛋白和抗体设计的生成生物学公司——Generate:Biomedicines 的 GB-0895(anti-TSLP)已进入重度哮喘 Phase 3,且已生成并测试 42,000+ 个蛋白;AbSci 的 ABS-201 是首个进入人体临床试验的 AI 设计从头抗体。两者都是 Chai 生物药设计抱负的直接类比对象,但临床验证更早。第四类是物理 + ML 混合既有玩家:Schrödinger(NASDAQ: SDGR)运营超过 35 年,与 1,750+ 制药和生物技术客户保持授权关系,并用可解释的物理方法(FEP+、Glide、BioLuminate)区别于纯深度学习路径。第五类是开源和学术替代方案,下一节会详细讨论;它们对 Chai 的结构价值主张构成最不对称的商品化威胁。 [CP001, CP002, CP003, CP004, CP005, CP006]
| 公司 | 成立时间 | 融资 / 状态 | 药物类型重点 | 临床阶段 | 核心护城河 | 对 Chai 的威胁 |
|---|---|---|---|---|---|---|
| Isomorphic Labs | 2021 | 已融资约 $600M(Series A,2024 年 11 月);Alphabet 支持 | 小分子;部分生物药(AlphaFold 系谱) | 临床前(多项药企共同开发交易) | Alphabet 资本、AlphaFold 架构、Eli Lilly + Novartis 合作 | 高——若扩展到生物药,将以 $600M+ 平台直接对上 Chai |
| Generate:Biomedicines | 2019 | 已融资约 $470M+;IPO 阶段;Flagship Pioneering 支持 | 蛋白、抗体、肽(Generative Biology™) | Phase 3(GB-0895,抗 TSLP,重度哮喘) | 已测试 42K+ 蛋白;140K sq ft 湿实验室;临床阶段推进最远的生成式蛋白公司 | 高——临床推进最远的直接生物药设计类比公司 |
| AbSci | 2011 | 已融资约 $350M+;NASDAQ:ABSI | AI 从头设计抗体和蛋白 | Phase 1(ABS-201;首个进入人体试验的 AI 从头设计抗体) | 先发临床验证;ACE Assay;77K sq ft 湿实验室;6 周周期 | 高——在直接类比的抗体设计上,临床阶段领先 Chai |
| Recursion (+ Exscientia) | 2013(Recursion);2012(Exscientia) | 已融资约 $1B+;NASDAQ:RXRX | 小分子(为主);基于表型组学的靶点发现 | Phase 1/2(合并平台的多项候选药物) | 50+ PB 自有数据;BioHive-2 / NVIDIA;Recursion OS 平台锁定 | 中——主要药物类型不同;Exscientia 曾有 3 个 Phase 1 候选药停止开发 |
| Insilico Medicine | 2014 | 已融资约 $400M+;IPO 前 | 小分子(为主) | Phase 2(ISM001-055 治疗 IPF——推进最远的 AI 设计小分子) | 13 项 IND 批准;40+ 项项目;全栈 Pharma.AI 平台 | 中——聚焦小分子,但证明了 Chai 尚缺的 AI 药物设计记录 |
| Schrödinger | 1990 | 上市公司(NASDAQ:SDGR);35+ 年运营历史 | 小分子(基于物理 + ML) | 商业化软件平台;1,750+ 家药企 / 生物技术客户 | 35+ 年客户关系;小分子领域基于物理的准确性;LiveDesign 锁定 | 低-中——买方群体不同,但争夺药企计算预算 |
| 开源 / 学术(Boltz-2、ESMFold、OpenFold) | 2020–2024 | N/A(资助金和机构经费) | 蛋白结构预测;生物药结构 | 仅临床前工具;无商业开发项目 | MIT/Apache 2.0 许可证;Boltz-2 明确以 Chai-1 为基准对比 | 高——Boltz-2 直接把 Chai-1 的核心结构预测价值商品化 |
竞品概况基于公司官方披露;融资金额为披露轮次的近似值。临床阶段反映截至 2026 年 5 月公开宣布的项目。 “对 Chai 的威胁”为分析师判断,依据是药物类型重叠和临床接近度,并非独立来源。
[CP002, CP003, CP004, CP005, CP006, CP007]将七个主要竞争主体放在两条有证据支撑的轴上:模态焦点(x 轴,0 = 纯小分子,1 = 纯生物制剂 / 抗体)和临床验证阶段(y 轴,0 = 仅临床前工具,1 = 3 期或商业化)。Chai Discovery 位于高生物制剂焦点、低临床阶段象限,生物制剂地盘主要与 AbSci 和 Generate:Biomedicines 重叠;后两者临床进展更靠前。
X 轴模态分数是基于各公司披露的产品组合和管线作出的顺序估计;并非来自单一量化来源。Y 轴临床阶段分数映射:0=临床前 / 工具,0.1–0.3=pre-IND,0.4–0.5=1 期,0.6=2 期,0.9=3 期,1.0=商业化批准。Schrödinger 主要是软件公司、无临床开发管线,因此排除。
[CP001, CP003, CP005, CP006, CP039]3.2 直接同行与相邻平台
在 Chai 最直接的竞争同行中,三家公司在 modality、能力和商业阶段对比上尤其相关:Generate:Biomedicines、AbSci 和 Isomorphic Labs;Recursion/Exscientia 与 Insilico Medicine 则是相邻但覆盖更广的平台。Generate:Biomedicines(Cambridge, MA)成立于 2019 年,由 Flagship Pioneering 和 ARCH Venture Partners 支持,已经在抗体、酶和其他功能蛋白中生成并实验验证 42,000+ 个设计蛋白。其领先候选物 GB-0895 是一款从头设计的 anti-TSLP 抗体,用于重度哮喘,已进入 Phase 3 临床试验,使 Generate 成为临床推进最靠前的生成蛋白设计公司。Generate 拥有 140,000+ 平方英尺实体湿实验空间,形成一体化设计-制造-测试优势;Chai 的轻资产、依赖伙伴模式无法复制这一点。AbSci Corporation(Vancouver, WA;NASDAQ: ABSI)成立于 2011 年,聚焦用 ACE Assay、SoluPro® 表达系统和深度学习平台推进 AI 驱动的抗体与蛋白工程。AbSci 的 ABS-201 由 AI 从头设计,截至 2025 年成为首个进入人体 Phase 1 临床试验的 AI de novo 抗体,比 Chai 更早提供临床概念验证。AbSci 声称拥有 6 周设计到表征周期,直接竞争 Chai-2 的快速生物药设计工作流。Isomorphic Labs(London)是 Alphabet spinout,也是行业内资本最充足的纯 AI 药物发现公司,2024 年 11 月 Series A 融资约 $600 million。尽管 Isomorphic 平台能力延伸到生物药,其历史聚焦和与 Eli Lilly、Novartis 的共同开发组合主要仍是小分子。Schrödinger 呈现出不同竞争动态:其 LiveDesign 协作平台和基于物理的工具(用于 docking 的 FEP+、Glide,以及用于生物药建模的 BioLuminate)经过 35+ 年已经深度嵌入制药工作流,形成行业最高的切换成本护城河。Insilico Medicine 的 ISM001-055 用于特发性肺纤维化,已进入 Phase 2,是 AI 设计小分子推进最远的案例;即使截至 2026 年 5 月没有 AI 药物最终获得 FDA 批准,它也表明 AI 药物设计可以产出可行临床候选物。这些竞争者的定价均未公开披露:合作通常是一单一议的定制安排,Schrödinger 是一个显著例外,它作为授权软件平台拥有经常性企业订阅收入。 [CP009, CP010, CP011, CP012, CP013, CP014]
| 能力 | Chai Discovery | Isomorphic Labs | Generate:Bio | AbSci | Recursion | Schrödinger |
|---|---|---|---|---|---|---|
| 从头抗体设计 | ✓(Chai-2;零样本) | 部分(主要是小分子) | ✓(已设计 42K+ 蛋白) | ✓(ABS-201 Phase 1) | 有限(聚焦小分子) | ✗(无深度学习生物药设计) |
| 小分子设计 | 有限(不是核心重点) | ✓(核心重点;AlphaFold 系谱) | 部分(部分酶 / 肽) | ✗(聚焦抗体) | ✓(核心重点 + 表型组学) | ✓(同类最佳 FEP+/Glide) |
| 蛋白复合物结构预测 | ✓(Chai-1 开放权重) | ✓(AlphaFold3 系谱) | ✓(生成式 + 结构) | ✓(ACE Assay + DL 底座) | ✓(集成到 Recursion OS) | ✓(BioLuminate 生物药) |
| 一体化湿实验室验证 | ✗(依赖合作伙伴;无内部湿实验室) | ✗(依赖合作伙伴) | ✓(140K sq ft;内部) | ✓(77K sq ft;ACE Assay 实验) | ✓(BioHive-2 机器人;2.2M 样本 / 周) | ✓(通过药企客户实验室) |
| 临床阶段管线 | ✗(截至 2026 年 5 月无 IND) | ✗(仅临床前) | ✓(Phase 3:GB-0895) | ✓(Phase 1:ABS-201) | ✓(多项 Phase 1/2) | ✗(软件平台,不是药物开发商) |
| 开放权重 / 开放 API 访问 | ✓(Chai-1 开放权重;API) | ✗(自研;仅限合作伙伴) | ✗(自研) | ✗(自研) | ✗(企业许可证) | ✗(企业许可证) |
| 基于物理的结合预测 | ✗(仅深度学习) | ✗(以 ML 为主) | ✗(以 ML 为主) | ✗(以 ML 为主) | ✗(以 ML 为主) | ✓(FEP+;小分子金标准) |
标记 ✗ 的矩阵单元表示没有公开证据证明具备该能力;标记为“有限”或“部分”的单元表示已有文档支持但不是核心重点。 鉴于公开披露有限,Isomorphic Labs 和 Recursion 的生物药能力单元属于部分未知;应视为部分证据缺口。 截至 2026 年 5 月,Chai Discovery 没有公开披露内部湿实验室能力。
[CP008, CP009, CP010, CP013, CP037]| 竞品 | 访问模式 | 定价 / 合同 | 包含的关键能力 | 切换成本 |
|---|---|---|---|---|
| Chai Discovery | API 访问 + 选择性合作 | 未披露;仅向合作伙伴开放(Eli Lilly 交易条款未公开) | Chai-2 从头抗体设计;Chai-1 结构预测 API | 低——仍处早期;尚无深度工作流集成 |
| Isomorphic Labs | 仅限合作(无公开 API 或许可证) | 未披露,逐笔定制交易(Eli Lilly、Novartis 条款未公开) | AlphaFold 系谱分子设计;共同开发项目 | 中——共同开发集成形成数据依赖 |
| Generate:Biomedicines | 研究合作 + IPO 资本(上市公司) | 未披露;合作经济条款未公开 | Generative Biology™ 平台;Phase 3 湿实验室制造 | 中——自研设计平台一旦集成,难以拆开 |
| AbSci | 合作式 R&D 交易;NASDAQ 上市 | 逐笔交易未披露;与 Merck、AbbVie 的历史交易已披露;条款未公开 | ACE Assay;SoluPro® 表达;从头抗体设计周期 | 中——ACE Assay 在抗体表征工作流中形成平台依赖 |
| Recursion | 企业合作 + NASDAQ 披露的 SaaS 授权 | NASDAQ:RXRX 披露显示平台授权收入;逐笔药企合作 | Recursion OS(表型组学 + AI 化学);BioHive-2 数据;Exscientia AI 化学 | 高——Recursion OS 在靶点发现和设计中形成数据与工作流锁定 |
| Schrödinger | 企业软件许可证 + 专业服务 | 数百万美元级年度许可证;公开 SaaS 模式(NASDAQ:SDGR 披露) | FEP+、Glide、BioLuminate、LiveDesign 协作平台;1,750+ 客户安装基础 | 高——35+ 年后,LiveDesign 已嵌入药企团队工作流;培训深 |
所有竞品定价要么未披露,要么来自 NASDAQ 季度披露。切换成本为基于集成深度的定性评估,并非独立量化。 鉴于当前访问仅向合作伙伴开放,Chai 的定价模式可能在商业化后显著演变。
[CP025, CP028, CP029]矩阵展示 Chai Discovery 与五家商业竞争对手在六项关键采购标准上的有无覆盖。不支持或不确定的单元格另行标注。Chai 的独特差异点是开放权重模型策略叠加从头抗体设计;相较直接同业,明显缺口在于缺少临床管线和内部湿实验室能力。
标注为“部分”或“有限”的单元格,表示依据公开披露判断,能力属于次要或新兴方向;“未知”表示缺少公开信息。矩阵不衡量能力深度或质量,只记录是否达到门槛。
[CP009, CP010, CP013, CP015, CP036, CP037]3.3 开源与现状替代方案
开源和免费工具对 Chai Discovery 构成一类独立且被低估的竞争威胁。AlphaFold Database(alphafold.ebi.ac.uk)由 EMBL-EBI 与 Google DeepMind、NVIDIA 和 Seoul National University 合作维护,按 Creative Commons Attribution(CC-BY-4.0)许可提供超过 200 million 个蛋白的预测结构。2026 年 3 月更新后,覆盖扩展到蛋白复合物结构。蛋白结构预测作为独立价值层由此被永久商品化——任何具备标准生物信息能力的制药团队都可以免费访问这些预测。Google DeepMind 发布的 AlphaFold3 把覆盖范围扩展到蛋白-核酸和蛋白-小分子相互作用;不过,模型权重下载需要 Google DeepMind 明确批准,商业使用条款也较严格,限制了它作为 Chai 潜在客户自托管替代方案的可部署性。最尖锐的开源商品化风险是 Boltz-2(github.com/jwohlwend/boltz),它按 MIT License 发布。Boltz-2 在结构预测准确性上明确对标 Chai-1,并额外预测结合亲和力——这是 Chai-1 不提供的能力,也把它的用途从结构预测扩展到配体排序。完全宽松的许可(MIT)意味着任何制药公司或学术团队都可以在生产环境部署 Boltz-2,无需授权费或限制。Meta AI 按 MIT license 开发的 ESMFold(github.com/facebookresearch/esm),不需要多序列比对就能从单一序列预测蛋白结构,支持早期筛选所需的快速推理。OpenFold(github.com/aqlaboratory/openfold)提供按 Apache 2.0 授权的 AlphaFold2 复现,让学术实验室可以用专有数据微调结构预测模型,进一步模糊商业能力与学术能力边界。除这些计算工具外,现状替代方案还包括传统抗体发现方法(杂交瘤融合和噬菌体展示);这些方法在 2024 年占更广泛抗体发现市场的 38.1%,并且仍被充分理解、风险更低,也有成熟 CRO 基础设施支持。Genentech、AstraZeneca、Pfizer 等大型制药公司的内部 AI 团队,则代表自建替代方案,可能压缩外部 AI 设计平台的总可用市场。战略含义很直接:Chai 必须在从头设计和候选物生成层做出差异化,而不是结构预测层;结构预测正越来越像免费的公共品。 [CP015, CP016, CP017, CP018, CP019, CP020]
3.4 护城河耐久性与竞争风险
Chai Discovery 的竞争护城河仍处早期,同时面临近期和结构性风险。截至 2026 年 5 月,公司最主要的差异化资产是:(1)Chai-2 带来的 zero-shot 从头设计抗体能力,已在已发表技术报告中的难靶点上验证;(2)Eli Lilly 合作,它既提供商业验证,也可能带来真实世界生物药设计数据,用于强化训练;(3)开放权重的 Chai-1 模型,它推动社区采用,但展示能力的同时也会加速竞争者商品化。上述优势在多个方向上面临耐久性挑战。最尖锐的风险是开源替代:Boltz-2 的 MIT license 以及直接对标 Chai-1,意味着免费使用的 Chai-1 网页服务器,要与一个可商业部署、且增加了 Chai-1 所缺结合亲和力预测的开源替代方案竞争。AbSci 凭 ABS-201 进入 Phase 1 的先发位置,意味着临床基准和制药关系数据会在 Chai 推进自己的 IND 候选物之前由竞争对手设定——可能围绕 AbSci 的设计路径和时间表塑造商业伙伴预期。Isomorphic Labs 背靠 Alphabet 的资本优势显著:其融资约 $600 million,而 Chai 累计约 $200 million,Isomorphic 可以在模型开发、BD 能力和制药共同开发安排上投入更多。如果 Isomorphic 从小分子扩展到从头设计生物药——其 AlphaFold 谱系架构在技术上具备这种能力——它将成为资本充足的直接竞争者。Recursion 的 50+ PB 数据护城河在结构上很难被较小竞争者复制:它需要与 NVIDIA 共同开发的专用机器人生物学基础设施(BioHive-2)、实体实验室设施,以及多年化合物库筛选。虽然这道数据护城河目前主要适用于小分子,但如果 Recursion 把基于 phenomics 的数据生成扩展到生物药靶点,它就会成为数据不对称如何变成耐久竞争壁垒的先例。Schrödinger 通过 35+ 年 LiveDesign 工作流整合形成的切换成本护城河,与 Chai 的目标买方画像(发现阶段生物药团队)不完全直接相关,但提供了一个警示模型:长期竞争位置最终可能由深度整合带来的锁定效应决定,而不是模型优越性。Generate:Biomedicines 的实体湿实验基础设施优势,也向 Chai 提出资本配置问题:是继续保持轻资产并依赖伙伴做实验验证,还是建设内部湿实验能力,以增强可信度并降低对 Eli Lilly 等伙伴的依赖。对 Chai 最不利的情景,是三件事叠加:(1)Boltz-2 或后继者完全商品化基于结构的设计;(2)AbSci 或 Generate 在 Chai 推进自己的 IND 之前建立临床先例;(3)Isomorphic Labs 借 Alphabet 资本扩展到生物药。这些风险单独看都可管理,但合在一起会压缩 Chai 在市场结构固化前建立差异化的窗口。 [CP021, CP022, CP023, CP024, CP025, CP026]
| 风险 / 护城河因素 | 类型 | 严重性 | Chai 暴露 | 证据 | 缓解措施 |
|---|---|---|---|---|---|
| Boltz-2 开源商品化 | 技术替代 | 高 | 高——Boltz-2 直接对标 Chai-1;MIT 许可证支持生产部署 | Boltz-2 GitHub(MIT)明确对标 Chai-1;新增结合亲和力预测 | Chai-2 从头设计(不只是结构预测)未被 Boltz-2 复现;强化合作伙伴排他性 |
| AbSci 在临床阶段 AI 从头抗体中的先发优势 | 先发风险 | 高 | 高——ABS-201 Phase 1 在 Chai IND 前建立临床基准 | AbSci 通过公司官方沟通披露 ABS-201 进入 Phase 1 | 加快 IND 申报路径;利用 Chai-2 设计优越性主张;借助 Eli Lilly 合作获取临床数据 |
| AlphaFold DB 永久商品化结构预测层 | 技术风险 | 中 | 中——Chai-1 结构预测要与免费 200M+ 结构数据库竞争 | alphafold.ebi.ac.uk CC-BY-4.0;200M+ 结构;2026 年 3 月复合物更新 | Chai 的价值从预测转向从头生成;结构预测是前提,不是产品 |
| Isomorphic Labs 资本与 Alphabet 支持 | 资本不对称 | 高 | 高——已融资 $600M,对比 Chai 约 $200M;Isomorphic 可在模型开发和 BD 上投入更多 | Isomorphic Labs 官方融资公告;Alphabet 公司所有权 | Chai Series B($130M)部分缩小差距;靠生物药专业化做差异,而 Isomorphic 尚未优先投入该方向 |
| Recursion 50+ PB 数据护城河 | 数据不对称 | 中 | 低-中——Recursion 数据护城河主要是小分子表型组学;直接威胁较小 | recursion.com/technology:50+ PB 数据;通过 BioHive-2 机器人基础设施每周 2.2M 样本 | 通过 Eli Lilly 合作积累生物药数据;专有从头设计结果可作为训练信号 |
| Generate:Biomedicines 湿实验室集成优势 | 运营集成 | 中 | 中——Chai 缺少内部湿实验室;依赖合作伙伴的模式增加验证延迟 | generatebiomedicines.com:140K sq ft 实验室;内部设计-制造-测试能力 | 轻资本模式保留可选性;Lilly 合作提供湿实验室访问;固定成本基数更低 |
| Schrödinger 35 年工作流锁定 | 市场进入 | 低-中 | 低——Schrödinger 主要面向小分子计算化学家;买方群体不同 | schrodinger.com/company:35+ 年历史;1,750+ 客户;LiveDesign 作为企业平台 | 生物药设计团队(Chai 主要目标)不是 Schrödinger 的核心存量客户群;工作流切入点不同 |
严重性评级是基于药物类型重叠、资本强度和时间线接近度的定性评估。暴露反映 Chai 在没有缓解措施时的近期脆弱性。 证据引用来自主要公司披露和官方来源;鉴于多数竞品定价和合作经济条款不公开,独立竞争情报有限。
[CP023, CP024, CP025, CP026, CP027, CP029]对 Chai Discovery 五项关键竞争护城河做紧凑评估,评级口径为截至 2026 年 5 月的持久性。模型性能和开放权重采用是当前评分最高的护城河;相较同业,临床验证和专有数据是最大缺口。
KPI 评级是基于截至 2026 年 5 月公开信息的定性评估。“高 / 中 / 低”反映相对最接近同业组的持久性,并非绝对尺度。临床管线评级下调 Chai,是因为截至报告日期,同业组合(AbSci、Generate)领先 1–3 个临床阶段。
[CP021, CP022, CP023, CP024, CP026]3.5 展示材料
04财务情况
4.1 收入模式与变现架构
Chai Discovery 的收入架构分两层,核心是对平台模型开放不同级别的访问。第一代分子结构预测模型 Chai-1 于 2024 年末发布,所有用户都可通过网页界面免费使用,商业用途也包括在内。Chai-1 模型权重按 Apache 2.0 许可证开源,但仅限非商业用途;商业网页访问则由 Chai 平台直接免费提供。Chai Discovery 的 Crunchbase 资料特别写明,“Chai-1 可免费用于商业应用”,Chai-1 在 bioRxiv 上的预印本也确认了这种商业网页访问模式。这个免费增值底座意在拉动平台采用,沉淀专有使用数据,并让 Chai 成为生物制药公司计算药物发现团队默认使用的分子建模工具。 Chai-2 是公司的旗舰级从头抗体设计系统,目前不公开开放。访问由一套“负责任部署”政策约束,Chai 只向经过选择的合作伙伴开放;目前仅限具名药企合作方。2026 年 1 月与 Eli Lilly 的合作,是该模式下首个公开确认的商业交易:Chai 将开发一个专用 AI 模型,只用 Lilly 的大规模专有数据训练,并部署到 Lilly 的 TuneLab 前沿 AI 单元,用于多个药物靶点的生物制剂发现。除定制训练系统外,Lilly 还可访问 Chai 的核心平台模型。Lilly 交易的财务条款没有公开披露,包括任何预付许可费、与发现成果挂钩的里程碑付款或特许权使用费安排。 这套架构暗示的中期收入模型有四块:(1)面向 Chai-2 合作伙伴访问的平台许可费,可能按年度软件协议设计;(2)面向希望基于专有数据微调版本的生物制药公司的定制模型开发费,Lilly 即属此类;(3)如果 Chai 生成的候选物推进到临床前或临床开发,可能获得发现里程碑付款;(4)更长期的共同开发收益层,即药企伙伴选择共同开发结构时,Chai 参与其生成项目的经济收益。共同领投 Series B 的 General Catalyst 公开预测,Chai 等 AI 药物设计工具的早期采用者,可能在 2027 年底前看到同类首创生物制剂进入临床试验;这意味着 GC 预计收入模型将在两年内从软件费用升级为里程碑和特许权使用费收入。 公司公开愿景是打造一套分子“计算机辅助设计套件”,这表明其志向是做更广泛可访问的平台。但当前商业落地仍处早期,依赖合作伙伴,财务条款也不透明。截至 2026 年 5 月,Chai 的收入尚未公开披露;公司或任何第三方都没有确认 ARR、合同数量或付费客户数量。 [CI001, CI002, CI003, CI004, CI005, CI006]
| 收入来源 | 产品 / 层 | 可用性 | 交易结构 | 财务证据 | 置信度 |
|---|---|---|---|---|---|
| 免费增值 Web 平台 | Chai-1 结构预测 | 面向所有用户免费,包括商业用户 | 不收费;基于使用收集数据 | bioRxiv Chai-1 预印本确认商业 Web 访问;Crunchbase 资料称商业使用免费 | 高 |
| 开源模型权重 | Chai-1 Apache 2.0 权重 | 非商业用途免费;商业用户也可免费使用 Web 访问 | Apache 2.0 许可;非商业用途不按席位收费 | Chai-1 预印本(biorxiv.org)和 Crunchbase 资料页已证实 | 高 |
| 定制模型开发 | 基于合作伙伴专有数据微调的定制 Chai-2 | 受限——Responsible Deployment 政策下选择性向合作伙伴开放 | 假设有预付款;可能包含里程碑付款;条款未披露(Lilly 交易) | businesswire.com 上的 Lilly 新闻稿确认定制模型训练;财务条款未披露 | 中 |
| 核心平台授权 | 在定制模型之外获得 Chai 核心模型访问权(Lilly 交易) | 受限——仅限合作伙伴 | 假设按年授权或 SaaS 收费;条款未披露 | hitconsultant.net 和 businesswire.com 的 Lilly 新闻稿确认 Lilly 获得核心模型访问权 | 中 |
| 发现里程碑付款 / 特许权使用费 | Chai 设计的候选物推进到临床前或临床阶段时触发付款 | 暂不适用——尚无 Chai 设计的分子进入临床试验 | 平台交易通常采用里程碑付款 + 特许权使用费结构;Chai 条款尚未得到确认 | loonbio.com、pda.org、drugdiscoverynews.com 讨论 AI 平台交易经济模型;Chai 尚未披露特许权使用费条款 | 低 |
| 共同开发参与 | 在与药企伙伴共同开发的项目中持有股权或共同出资权益 | 尚无任何 Chai 项目获得公开确认 | 潜在未来模式;目前尚非现状 | 没有公开证据显示 Chai 拥有共同开发股权;这一点来自 TechCrunch 所述平台野心的推断 | 很低 |
所有收入流描述均根据公司架构、公开新闻稿和 AI 药物发现行业标准交易结构推断。只有 Lilly 合作(第 3、4 类收入流)已公开确认;所有收入流的财务条款均未披露。截至 2026 年 5 月,各收入流的收入金额未知。
[CI001, CI002, CI003, CI004, CI005, CI007]| 收入机制 | 价格 / 单位 / 合同类型 | 标价 vs. 实际成交 | 折扣 / 未知项 | 来源 / 置信度 |
|---|---|---|---|---|
| Chai-1 Web 平台 | 面向所有用户免费,包括商业用户 | 标价 = 实际成交 = $0;不收费 | 定价无不确定性;已确认免费访问 | biorxiv.org Chai-1 预印本;Crunchbase 资料页——高置信度 |
| Chai-1 开源权重(非商业) | Apache 2.0 许可下免费 | 标价 = 实际成交 = $0;Apache 2.0 开源 | 未经协议,商业用户可能不得以商业形式再分发权重 | biorxiv.org Chai-1 预印本——高置信度 |
| Chai-2 平台访问(合作伙伴门控) | 未公开列价;假设为年度授权费 | 标价未知;实际成交价完全未披露 | 无公开定价;可能按合作伙伴定制谈判;未确认标价 | businesswire.com Lilly 新闻稿;techcrunch.com 资料——低置信度(推断) |
| 定制模型开发费(Lilly 类型) | 未披露;假设为预付款 + 里程碑结构 | 标价未知;Lilly 经济条款未披露 | 根据行业基准,预付款可能在 $1M 到 $20M+;没有公开确认 | mavenbio.com 药企 R&D 配置;pda.org 交易结构分析——很低置信度 |
| 发现里程碑付款 | 目前不适用;没有临床阶段分子 | 尚未触发任何里程碑;无可用基准 | 生物技术交易中,每次阶段推进的典型里程碑付款为 $1M–$50M+;Chai 未披露任何里程碑 | drugdiscoverynews.com AI 经济性;pda.org——很低置信度(递延) |
| 共同开发特许权使用费 | 未确认;长期潜力 | 未披露特许权使用费条款;目前不适用 | 若设计为带特许权使用费的共同开发,典型费率为净销售额的 1–5%;Chai 未确认 | SI004 drugdiscoverynews.com;SI014 pda.org——很低置信度(推测) |
Chai-2 及之后的所有定价条目均由行业基准估计或推断。Chai 未披露 Chai-2 或 Lilly 合作的任何价目表、交易规模或变现条款。只有 Chai-1 定价(免费)得到官方来源确认。
[CI001, CI002, CI005, CI008, CI032]Chai Discovery 的收入模式从开放免费访问基础层(Chai-1 网页版)流向有门槛的商业合作(Chai-2 定制模型和平台授权),并可能随着合作方项目推进到临床开发,形成里程碑和特许权使用费收入。
收入流基于 Chai 披露的架构(businesswire.com、techcrunch.com、biorxiv.org)和标准 AI 药物发现平台交易结构(pda.org、drugdiscoverynews.com)推断。本图未纳入任何已确认收入金额或交易经济条款。
[CI001, CI002, CI004, CI005, CI007, CI008]4.2 成本结构、单位经济与资本强度
Chai Discovery 的成本结构主要由两项构成:人力资本(旧金山的 AI 研究人才)和算力基础设施(AI 模型训练与推理)。根据 BuiltInSF 的间接报道,截至 2026 年初,公司约有 29 名员工;该数字尚未得到官方确认。旧金山 AI 公司若处在这一人才层级,含税含福利的单人年成本通常在 $250,000 至 $400,000 之间,意味着估算年薪资消耗约为 $7–12 million。管理层称代码库完全自研——“我们代码库里的每一行代码都是自研的”——不使用现成大语言模型,架构也高度定制;这说明技术团队偏向资深 AI 研究员,薪酬大概率落在区间上沿。 算力成本是关键且未披露的第二条成本线。Chai 训练其规模下的前沿蛋白质结构预测和从头设计模型,需要大量 GPU 算力。行业基准显示,训练 Chai-2 这一级别的前沿 AI 生物模型,每次主要训练运行成本约在 $1–15 million,合作伙伴部署还会叠加持续推理成本。Lilly 合作明确涉及在 Lilly 专有数据上训练定制模型,这会产生额外算力支出,但公开披露中看不到。Chai 没有披露 AWS、Azure、GCP 或其他算力开支,公开报道中也没有出现对其算力预算的独立估算。 Chai 成本画像中的一个关键结构性优势,是没有湿实验室基础设施。不同于 Generate:Biomedicines(140,000 sq ft 湿实验室空间)或 AbSci(77,000 sq ft),Chai 完全依赖合作伙伴做湿实验室验证,而不是自建内部实验设施。这压低了资本开支,也让软件模式具备较高毛利潜力;但它也意味着 Chai 声称的命中率需要合作伙伴实验室确认,由此形成验证依赖,既有财务含义,也有科学含义。 大药企研发支出背景凸显机会规模,也凸显竞争成本:头部制药公司合计每年研发支出超过 $200 billion,Roche、J&J、Merck 等单家公司每年都超过 $15 billion。在这个背景下,与单一合作伙伴签下的 AI 药物发现交易,收入很可能只占成功平台最终必须捕获金额的一小部分,才足以支撑 $1.3 billion 估值。S&P Global Market Intelligence 的行业数据确认,生物制药风投活动仍集中在临床前阶段平台,Series B+ 交易需要可信的商业化证据;Chai 通过 Lilly 交易部分跨过了这道门槛,但尚未用已披露收入指标完全证明。 [CI010, CI011, CI012, CI013, CI014, CI015]
| 成本类别 | 估计规模 | 关键驱动因素 | 结构特征 | 置信度 |
|---|---|---|---|---|
| AI 研究人才(薪酬) | 估计 $7–12M/year | ~29 名员工 × $250–400K 全成本;旧金山 AI 研究员薪酬有溢价 | 最大成本项;技术团队技能密集,成员履历来自 OpenAI/Meta/Absci;员工数未获官方确认 | 低(仅估计) |
| 计算基础设施(训练) | 估计 $2–15M/year | 前沿蛋白 AI 模型的 GPU 集群成本;Lilly 合作的定制模型训练会在基准上增加成本 | 单次训练运行非经常性但会持续发生;云端或自有集群;Chai 完全未披露 | 很低(仅行业代理) |
| 计算基础设施(推理) | 估计 $1–5M/year | 面向 Chai-1 用户的 Web 平台推理;面向 Chai-2 部署的合作伙伴 API 推理 | 随用户规模扩大而增长;可能由 API 收入抵消;不透明 | 很低(仅行业代理) |
| 湿实验室 / 实验验证 | 内部 $0(依赖合作伙伴) | 不自建湿实验室;验证完全外包给药企伙伴 | 相比 Generate:Bio($140K sq ft)和 AbSci($77K sq ft),资产轻的结构优势明显;大幅降低资本开支 | 高 |
| 业务拓展 / 商业化 | 早期阶段;未披露 | Chai 在 Lilly 之外扩张时,BD 和销售员工数会增加 | 从 1 个药企合作扩展到多个合作必须补足;2026–2027 年会明显增长 | 很低(根据 Series B 资金用途推断) |
| 一般及行政 | 估计 $1–3M/year | ~29 人公司的法务、财务、人力;Series B 后的治理成本 | 对旧金山 30 人左右、独角兽阶段的初创公司属常见成本;相较技术成本不算重大 | 很低(行业代理) |
所有成本数字均由行业代理、员工数报道和前沿 AI 生物模型计算成本区间的公开披露测算。Chai 未披露任何财务报表、烧钱速度或成本拆分。通过分析 Chai 依赖合作伙伴的模式,可确认湿实验室基础设施成本为零。
[CI010, CI011, CI012, CI013, CI014, CI015]Chai 的成本结构由相对克制的人力成本(精简的 29 人团队)和前沿 AI 模型训练的大额算力开支组成。没有湿实验室基础设施是一项结构性优势;只要算力成本可控,Chai 可以跑出高毛利的软件模式。
所有数值均为代理估算,依据是旧金山 AI 劳动力市场数据、前沿 AI 算力成本基准,以及 Chai 公开描述的团队规模和架构。Chai 未披露任何财务数字。毛利率估算为方向性判断,基于可比 AI 软件平台模式。
[CI010, CI011, CI012, CI013, CI014]4.3 资本充足性、现金跑道与融资风险
Chai Discovery 三轮累计融资约 $230 million:2024 年 9 月完成 $30 million 种子轮(Thrive Capital 与 OpenAI 领投,估值约 $150 million),2025 年 8 月完成 $70 million Series A(Menlo Ventures/Anthology Fund 领投,估值约 $550 million),2025 年 12 月完成 $130 million Series B(Oak HC/FT 与 General Catalyst 共同领投,估值 $1.3 billion)。Series B 新闻稿称资金将用于“加速研究和产品开发,并扩大商业化投入”;这套表述符合公司从研发高投入阶段转向搭建商业基础设施的状态。 按每年 $20–35 million 的估算消耗率计算——这与 29 人 AI 研究团队、显著算力开支以及早期销售和商业开发成本相符——仅 $130 million Series B 就可从 2025 年 12 月交割起提供约三到六年现金跑道,且尚未计入任何合作收入。如果 Lilly 合作或其他药企交易带来的收入显著降低净消耗,现金跑道还可进一步延长。反过来,团队快速扩张、算力投资加速,或战略性湿实验室合作资本支出,都可能缩短消耗周期。在缺少已披露财务报表的情况下,所有现金跑道估算都带有推测性。 投资人联盟质量很高,构成安全垫,显著降低了稀释性过桥融资风险。General Catalyst、Oak HC/FT、Thrive Capital、OpenAI、Menlo Ventures 和 Emerson Collective 都是一线机构资本,储备雄厚。多家投资人(Thrive、OpenAI、Dimension、Menlo)参与了多轮融资,释放出对未来融资能力的信心。这套名单降低了当前现金跑道内被迫降估值融资或困境过桥融资的风险。 行业融资背景同时带来顺风和风险。S&P Global Market Intelligence 数据显示,2025–2026 年生物制药 VC 资金不成比例地流向 AI 驱动的临床前平台,这一趋势有利于 Chai 的定位。但同一时期,多家知名 AI 药物发现公司在临床阶段遭遇挫折;loonbio.com 的行业分析认为,自 2015 年以来 AI 药物发现风投资金超过 $60 billion,但截至 2025 年初产生的 FDA 批准为零——这提出了一个问题:投资人对临床前到临床转化的耐心是否正在逼近上限。Chai 必须在当前融资周期内,证明其能够走向临床里程碑或有意义的平台收入,否则下一轮融资可能面临估值重置。drugdiscoverynews.com 关于 AI 从命中到临床候选物缺口的分析,特别指出从 in silico 设计转向经验证的临床候选物,是成本最高、失败率最高的阶段;Chai 尚未用任何公开披露项目走过这道关。 [CI018, CI019, CI020, CI021, CI022, CI023]
| 情景 | 假设年度烧钱 | 收入抵消假设 | 净现金消耗 | 基于 2025 年 12 月 Series B($130M)的现金跑道 | 主要风险 |
|---|---|---|---|---|---|
| 保守 / 精简运营 | $15–20M/year | 无显著收入 | $15–20M/year | 6.5–8.7 年 | 商业牵引不足;可能意味着增长缓慢 |
| 基准情景 / 扩张 | $25–35M/year | Lilly 收入有限 + 1–2 笔新增交易($3–8M/year) | 净额 $20–30M/year | 4.3–6.5 年 | 商业化招聘推高烧钱;必须签下更多合作 |
| 激进 / 扩规模 | $40–60M/year | 合作收入增长($8–20M/year) | 净额 $30–45M/year | 2.9–4.3 年 | 需要快速签约;同时验证商业化说法;临床里程碑前可能面临 Series C 融资压力 |
| 困境情景 | >$60M/year | <$5M 收入 | >$55M/year 净额 | <2.4 年 | 临床验证前需要紧急过桥或下轮降价融资;考虑到投资人质量,这不太可能,但如果 AI 药物发现情绪转向,也并非不可能 |
所有情景值都基于公开数据估计(员工数、Series B 融资额、行业成本基准)。Chai 未披露烧钱速度、ARR 或财务报表。$130M Series B 于 2025 年 12 月 15 日完成。此前轮次($30M 种子轮、$70M Series A)假设在 Series B 完成时已大部分或全部投入使用。收入抵消高度不确定,完全取决于 Lilly 交易结构和新增交易签约。
[CI018, CI019, CI020, CI021, CI022]Chai 的 $130M B 轮(2025 年 12 月)按现实烧钱速度估算,可提供 3–7 年现金跑道,且未计入收入抵消。该区间反映实际烧钱速度高度不确定;公司尚未披露。
现金跑道估算由代理烧钱率假设(见 TI003)和已确认 B 轮融资额($130M)构建。假设此前轮次在 B 轮交割前已大体投入。收入抵消具有推测性,并取决于不同情景。由于 Chai 未披露财务数据,所有估计置信度较低。
[CI018, CI019, CI020, CI022, CI023]4.4 财务透明度缺口与尽调问题
Chai Discovery 的财务图景有一个异常特征:已融资金额很高,已披露财务信息却很少。公司通过官方新闻稿确认了融资金额、估值和投资人联盟;其他所有内容——收入、ARR、现金消耗、交易经济性、算力成本、利润率结构和资产负债表——要么未披露,要么只能推断。对一家成立两年的私营公司而言,这种不透明并不反常;但 $1.3 billion 估值让尽调所需的举证责任明显提高。 Eli Lilly 合作是最重大的未知数。作为公司唯一公开确认的商业交易,Lilly 合同的财务结构——包括任何预付款、年度许可费、里程碑触发条件和特许权使用费安排——定义了 Chai 当前的收入轨迹。Eli Lilly 是上市公司,若合作达到重大性门槛,可能会在财务申报中提及;但目前没有 Lilly 的 SEC 披露确认具体交易经济性。pda.org 对 AI 药物发现监管和商业挑战的分析指出,AI 平台交易通常采用与发现成果挂钩的里程碑付款结构,因此早期合作的经济价值高度取决于候选物推进情况。 Lilly 交易之外,还有几类财务信息在结构上缺失。Chai 没有正式披露员工数;BuiltInSF 给出的约 29 人只是间接估计。尽管算力开支可能是一条每年数百万美元的成本线,具体数额完全不透明。合同管线、销售周期数据,以及 Lilly 之外的其他合作伙伴名称,都没有得到确认。mavenbio.com 对大药企研发资本配置的分析显示,发现软件工具在每个大型药企合作中的年度平台费用通常在 $1–20 million 区间,但没有交易数量和合同结构,无法估算 Chai 的实际交易规模。 药物发现行业经济学文献一贯显示,从 AI 命中生成走向临床候选物所需资本,显著高于发现阶段本身;纳入临床成本后,通常要高出两到三个数量级。Chai 当前模式把这部分成本递延给合作伙伴,近期保住资本效率,但也可能限制上行空间,因为合作伙伴捕获临床价值。drugdiscoverynews.com 对 AI 改造药物发现经济性的分析强调,能产生发现命中的 AI 软件平台,必须及早谈下有利的里程碑和特许权使用费条款,才能捕获长期经济价值;随着时间推移,药企伙伴对这些条款会更熟练。 [CI027, CI028, CI029, CI030, CI031, CI032]
| 财务指标 | 是否披露? | 最佳可用代理指标 | 尽调路径 | 重要性 |
|---|---|---|---|---|
| 收入 / ARR | 未披露 | 零或极少(商业化前阶段;Lilly 交易是唯一确认交易) | 向管理层索取 P&L 或收入桥;查阅 Lilly 的 SEC 文件,判断交易重要性 | 关键 |
| 烧钱速度 / 现金消耗 | 未披露 | 根据员工数 + 计算成本代理估计 $20–35M/year | 索取月度烧钱报表;查看数据室里的银行余额趋势 | 关键 |
| Lilly 交易经济性 | 未披露(财务条款) | 结构上:预付款 + 里程碑付款 + 特许权使用费;规模未知 | 索取交易条款摘要,或让 Lilly(上市公司)提供 SEC 8-K 对照 | 关键 |
| 计算 / 云支出 | 未披露 | 基于前沿 AI 生物模型基准估计 $3–20M/year | 索取 AWS/GCP/Azure 发票,或 CTO 的计算成本估算 | 高 |
| 员工数(官方) | 未获官方确认 | BuiltInSF 间接报道为 ~29 名员工(低置信度) | 索取按职能拆分的 HR 员工数;交叉核对 LinkedIn 员工数 | 高 |
| 新增合作管线 | 未披露 | 公开确认的只有 Lilly;公司可能有未披露的早期访问合作伙伴 | 向管理层索取合作伙伴名单、交易阶段和财务条款 | 高 |
| 资产负债表 / 现金状况 | 未披露 | 根据累计融资减估计支出,估计约 $100–200M | 索取经审计资产负债表或管理层财务摘要 | 中 |
| 毛利率 | 未披露 | 如果计算成本适中,软件模式意味着毛利率潜力可达 60–80%+ | 向管理层索取按收入流拆分的毛利率 | 中 |
所有披露 / 未披露判断均基于截至 2026 年 5 月的公开信息。所有代理估计均由行业基准和间接报道构建。本次尽调未审阅 Chai 经审计财务、管理账或投资人材料。
[CI027, CI028, CI029, CI030, CI031, CI032]评估截至 2026 年 5 月的五项关键财务风险及其发生可能性和影响。收入不透明与交易集中度(仅 Lilly 确认)是短期评分最高的财务风险。
风险评级是基于截至 2026 年 5 月公开证据的定性评估。低、中、高评级反映对未来 3 年内 Chai 商业模式中发生可能性和潜在财务影响组合的判断。
[CI027, CI028, CI030, CI031, CI035]4.5 重点观察
05产品与技术
5.1 产品组合与模型架构
Chai Discovery 的产品组合分为两个明确层级。Chai-1 于 2024 年 10 月以 biorxiv 预印本形式发布,并在 PyPI(chai_lab,目前为 v0.6.1)开放,是一个多模态基础模型,在单一架构中统一预测蛋白质、小分子、DNA、RNA、糖基化以及混合模态复合物。它接收 FASTA 输入,可选连接 MSA 服务器(MMseqs2/ColabFold 集成),也可用交联质谱数据等实验约束提示模型,把预测准确率提升两位数百分点。Chai-1 按 Apache 2.0 发布,允许学术和商业使用。推荐推理硬件是 NVIDIA A100 80 GB 或 H100 80 GB;A10 和 A30 支持更小的复合物,用户也报告过用消费级 RTX 4090 GPU 成功运行。 Chai-2 于 2025 年 6 月发布,是一个专有多模态生成模型,专为完全从头抗体和迷你蛋白设计而建。不同于 Chai-1 预测现有或假设序列的结构,Chai-2 只用目标抗原和表位规格,就能从零生成新型抗体序列,在没有任何种子序列或既有结合体的情况下设计全部六个互补决定区(CDR)。联合创始人 Jack Dent 表示:“我们代码库里的每一行代码都是自研的。我们不是从开源 [生态] 中拿现成 LLM 再微调。这些都是高度定制的架构。”Chai-2 的模型权重和训练数据没有公开披露;访问由早期访问合作伙伴计划把关。lab.chaidiscovery.com 的浏览器界面在认证后提供免费的 Chai-1 预测,包括商业药物发现用途。 [CE001, CE002, CE003, CE004, CE005, CE006]
| 模块 / 资产 | 用户 / 客户 | 状态 / 成熟度 | 差异化 | 尽调缺口 |
|---|---|---|---|---|
| Chai-1(结构预测) | 学术研究人员、生物技术 / 药企开发者、药物发现团队 | 生产可用(Apache 2.0 开源,PyPI v0.6.1) | 多模态覆盖(蛋白、小分子、DNA、RNA、聚糖);MSA + 约束条件化 | 没有第三方运行的独立 CASP/CAMEO 基准比较 |
| Chai-2(抗体生成式设计) | 通过早期访问提供给药企伙伴;Eli Lilly 是锚定商业合作伙伴 | 早期访问 / 专有(未开源) | 16% 从头命中率(较既有方法高 100×+);仅凭表位生成完整 CDR | 所有基准均为自报;截至 2026 年 5 月,尚无同行评议的独立验证 |
| lab.chaidiscovery.com(Web 界面) | 无 GPU 基础设施的科学家;商业药物发现用户 | 生产可用(包括商业用途在内免费访问;需登录) | 降低 Chai-1 使用门槛;无需 GPU;零成本 | 需登录;容量 / 吞吐上限未披露;未记录 SLA |
| 合作伙伴 API / Chai-2 早期访问 | 选定药企和学术机构(Eli Lilly 已确认;其他未披露) | 有限早期访问(邀请制) | Chai-2 直接接入药企发现管线 | 商业条款、容量、正常运行时间和支持层级未公开披露 |
状态来自官方公告和 GitHub README;Chai-2 访问覆盖范围为公司自称;用户数或吞吐量没有第三方审计。
[CE001, CE004, CE007, CE021, CE026]从数据输入到模型核心再到输出交付的五层架构,展示开放(Chai-1)与专有(Chai-2)边界。
Chai-2 层细节来自 biorxiv 预印本和 GitHub 引用推断;确切架构边界属专有信息。
[CE002, CE012, CE021, CE022, CE026]5.2 技术表现——抗体设计基准与验证
Chai-2 的核心主张是:面对 52 个多样化抗原靶点,在完全从头抗体设计中达到 16% 命中率;这些靶点在 RCSB Protein Data Bank(PDB)中都没有既有抗体或纳米抗体结合体。每个靶点测试不超过 20 个设计,模型在单轮湿实验室检测中,让 50% 的靶点至少产生一个经实验验证的结合体。经验证抗体展现出纳摩尔级结合亲和力、对目标靶点的特异性,以及可比已获批疗法的开发属性。从 AI 设计到实验确认的端到端流程少于两周。 传统计算抗体发现方法——包括免疫、定向进化,以及基于酵母表面展示的计算筛选——报告的命中率持续低于 0.1%,这让 Chai-2 声称的表现高出 100 倍以上。在迷你蛋白设计上,Chai-2 达到 68% 的湿实验室成功率,并经常产出皮摩尔级结合体。2025 年 11 月的困难靶点预印本进一步显示,超过 86% 的全长单克隆抗体(mAb)设计,其可开发性画像与已获批治疗性抗体相当;Chai-2 设计的实验解析结构也与 in silico 预测高度吻合(原子级准确度)。Chai-2 还成功为 GPCR 激动作用设计出功能性抗体,并为肿瘤特异性新表位设计出高度特异性抗体。 关键基准注意事项:AlphaFold 3(DeepMind/Google)于 2024 年 5 月发表在 Nature,是生物分子结构预测领域主导性的同行评议基准;Chai-1 预印本明确说明,AlphaFold 3 基准值取自公开发布的预测结果,而不是独立运行。ESMFold(Meta/FAIR)于 2023 年 1 月发表在 Science,使用蛋白质语言模型,仅凭序列预测原子级蛋白质结构。AlphaFold 3 和 ESMFold 都主要是结构预测工具;二者都不是为从头抗体生成而设计,因此直接比较命中率在方法上并不等价。CASP(Critical Assessment of Protein Structure Prediction)仍是结构预测的国际金标准盲测竞赛。截至 2026 年 5 月,所有 Chai-2 抗体设计数据都只发布在公司作者撰写的 biorxiv 预印本中,尚无独立同行评议。 [CE007, CE008, CE009, CE010, CE011, CE013]
| 用户任务 / 目标 | 当前 / 既有工作流 | Chai-2 方案 | 可衡量收益(据报) | 关键限制 |
|---|---|---|---|---|
| 针对新抗原从头发现抗体命中物 | 抗体库高通量筛选(>10⁶ 个设计);周期数周到数月 | 基于靶点 + 表位生成 ≤20 个计算模拟候选物;2 周内完成湿实验验证 | 16% 命中率,对比传统方法 <0.1%;52 个靶点中目标成功率 50% | 性能为自报;没有与同类最佳文库筛选进行前瞻性头对头比较 |
| 面向紧凑型生物制剂的纳米抗体(VHH)设计 | 骆驼科动物免疫 + 噬菌体展示;周期 3–6 个月;大量使用动物 | Chai-2 零样本生成 VHH;在小蛋白基准中获得皮摩尔级结合物 | 小蛋白设计成功率 68%;报告案例达到皮摩尔级亲和力 | VHH 专用验证集小于完整 IgG 数据集;可开发性数据有限 |
| 具备类药性质的全长 mAb 设计 | 杂交瘤、噬菌体展示,或动物源抗体人源化 | 生成全长 VH-VL 格式,并集成可开发性评分 | >86% 的设计具备治疗级可开发性画像(challenging-targets 论文) | 晶体结构验证仅覆盖子集;GPCR 靶点仍具挑战 |
| 面向此前难以攻克靶点的抗体设计 | 失败模式:靶点无法通过常规手段生成有功能的抗体 | Chai-2 据称在数小时内解决了此前耗资 >$5M、耗时 3+ 年的难题 | 一个记录案例在 2 周内完成实验室验证;已设计 GPCR 激动剂抗体 | 仅为单个案例研究层级;关于「难以攻克」靶点类别的系统数据尚未发表 |
| 零成本学术蛋白结构预测 | AlphaFold2/3 Server、ESMFold API 或本地 ColabFold 安装 | lab.chaidiscovery.com 免费提供 Chai-1 预测 Web 访问,包括商业用途 | 根据公司报告基准,在 CAMEO/CASP 任务上准确度可与 AF3 相当 | Web 界面容量未记录;MSA 集成需要调用外部 MSA 服务器 |
收益数据来自公司预印本;传统方法命中率(<0.1%)来自 businesswire 和 techcrunch 对行业标准的引用;尚无独立头对头研究发表。
[CE007, CE008, CE009, CE011, CE013, CE015]| 模型 / 方法 | 任务类型 | 关键指标 | 报告值 | 来源 / 验证状态 |
|---|---|---|---|---|
| Chai-2(Chai Discovery) | 从头抗体设计命中率 | 每个设计对应的湿实验验证结合物 | ~16%(52 个靶点,≤20 个设计 / 靶点) | 公司 biorxiv 预印本(2025 年 7 月);未经同行评议 |
| 传统计算方法(此前最先进水平) | 计算抗体筛选与优化 | 湿实验命中率(结合物频率) | <0.1%(预印本引用的行业标准) | 公司引用的基线;来自 biorxiv Chai-2 论文引用 |
| AlphaFold 3(Google DeepMind) | 生物分子结构预测(蛋白、配体、核酸) | PoseBusters 基准准确度 | 2024 年同行评议的同类最佳方法;Chai-1 比较所用 AF3 数值取自 AF3 公开发布 | Nature 同行评议(2024 年 5 月);Chai 团队未重新运行 AF3 基准 |
| ESMFold(Meta FAIR) | 单序列蛋白结构预测 | CAMEO 基准上的 TM-score | 接近 AF2 的准确度;推理快得多;仅单序列 | Science 同行评议(2023 年 1 月);没有直接抗体设计能力 |
所有 Chai-2 抗体设计指标均为自报。AlphaFold3 和 ESMFold 是结构预测工具,不是抗体设计系统;直接比较命中率在方法上并不等价。Chai-1 的 AF3 比较基于 AF3 公开预测结果,并非独立运行的评估。
[CE007, CE016, CE017, CE018, CE019, CE034]从靶点规格到实验室验证抗体命中物的端到端工作流,使用 Chai-2 可在两周内完成。
[CE007, CE008, CE009, CE010, CE011, CE031]按模态展示 Chai-1 和 Chai-2 的能力覆盖,包含截至 2026 年 5 月的成熟阶段和验证依据。
成熟度标签由公开披露推断;Chai-2 访问广度未经独立审计;Chai-2 的“生产可用”指商业早期访问,并非全面开放。
[CE002, CE011, CE013, CE014, CE015, CE033]5.3 开源策略与开发者生态
Chai-1 是 Chai Discovery 技术策略中的开源锚点。Chai Discovery 按 Apache 2.0 发布 Chai-1,包括模型权重和推理代码,由此打造了一个社区飞轮,推动采用、外部验证和人才信号。Python 包(chai_lab)通过 PyPI 分发并定期更新;最新发布版本为 0.6.1,开发分支每天更新。GitHub 仓库(chaidiscovery/chai-lab)提供完整源码、可复现环境的 dev-container 设置,以及 Chai-1 和 Chai-2 预印本的引用元数据。HuggingFace 托管 Chai-1 模型卡,包含安装说明和模型文档。Chai-2 的引用块已经出现在 GitHub README 中;即便 Chai-2 模型本身仍为专有,Chai 也在把该库定位为社区接口。 这种开源核心策略对应着 MLOps(如 Hugging Face)和生物信息学(如 ColabFold)中的成功先例:免费提供强大的基线模型,建立生态依赖,再通过下一代能力的商业访问变现。Chai-2 保持专有,因为公司认为抗体设计能力是核心商业差异点。PLOS Computational Biology 的 LAP(Liability Antibody Profiler)工具包可以说明补充 Chai 平台的开源工具类型:它会对抗体负债进行测序和结构映射,并与天然和治疗性抗体库比对,用于可开发性评估;未来也可能与 Chai 平台集成。 [CE003, CE004, CE021, CE022, CE036]
| 层 / 组件 | 角色 | 关键依赖 | 风险 / 约束 |
|---|---|---|---|
| 多模态 Transformer(Chai-1 核心) | 同时编码蛋白、小分子、DNA、RNA、聚糖等模态 | 自研定制架构;CUDA GPU(优先 A100/H100) | 架构细节在 biorxiv 预印本中部分披露;未经同行评议 |
| MSA 生成模块 | 提供多序列比对上下文,以提升预测准确度 | 外部 MMseqs2/ColabFold MSA 服务器;可选但可提升性能 | 依赖第三方服务器可用性;单序列模式准确度下降 |
| 实验约束条件化 | 允许湿实验数据(如交联 MS)引导结构预测 | 来自实验室合作伙伴的上游实验数据 | 需要实验数据作为输入;早期发现阶段未必可得 |
| CDR 生成引擎(Chai-2 专有) | 根据靶点 + 表位规格从头生成全部六个抗体 CDR 环 | 专有训练数据,包括 PDB 抗体结构和表位–抗体配对 | 专有系统,无法独立复现;训练数据构成未披露 |
| 可开发性评估集成 | 评估生成抗体的类药画像(多反应性、PSR、稳定性) | 内部评分模型 + LAP 式风险画像 | 可开发性指标阈值和评分方法未公开记录 |
Chai-2 架构为专有;信息来自 biorxiv 预印本和 GitHub README。具体模型超参数、训练数据规模和微调协议未公开披露。
[CE002, CE005, CE006, CE012, CE030, CE036]支撑开源 Chai-1 和专有 Chai-2 平台的关键基础设施、数据和模型依赖。
[CE003, CE004, CE021, CE024, CE026]5.4 部署、路线图与信任控制
Chai-2 只通过早期访问合作部署给经过选择的学术机构和生物制药组织。公司运行一套自称“负责任部署框架”的机制,重点放在有利健康、低风险的应用,生物安全,以及与社会目标一致。2026 年 1 月 Eli Lilly 合作——Lilly 的 TuneLab 项目将 Chai-2 集成到生物制剂发现中——是迄今最受关注的合作伙伴部署。Menlo Ventures 合伙人 Greg Yap 在 Series A 公告中公开表示,“生物技术行业中相当一部分已经申请 Chai-2 访问权限”,但具体申请数量或批准率没有披露。 从监管合规角度看,FDA 于 2025 年 1 月发布题为《使用人工智能支持药物和生物制品监管决策的考量》的指导草案,并称自 2016 年以来已审查超过 500 份纳入 AI 的药物申报。Chai-2 设计的抗体候选物仍需完成完整的 IND 支持性研究、临床试验(I–III 期)和监管审评,才能获批;Chai 平台加速的是发现阶段设计,不是临床或监管批准。截至 2026 年 5 月,没有专门面向 AI 设计生物制剂这一独立类别的 FDA 路径。从头抗体设计带来的生物安全担忧已经被承认,包括生成针对危险靶点的新型结合体的双重用途风险,但公开文件尚未充分缓解。公司路线图包括把 Chai-2 扩展到更多模态(肽、酶、小分子),也可能扩展到更广格式(双特异性抗体、ADC),但具体时间表没有公开披露。 [CE028, CE029, CE031, CE032, CE033, CE037]
| 控制 / 框架 / 风险 | 状态 | 范围 / 覆盖 | 差距 / 尽调问题 |
|---|---|---|---|
| 负责任部署框架(RDF) | 已宣布;早期访问阶段已投入运行 | Chai-2 合作伙伴访问;按生物安全和健康正向应用做过滤 | 具体标准、审查流程和执行机制尚未公开记录 |
| FDA AI 指南合规(草案 Jan 2025) | 相关但不具约束力;仍处草案指南阶段 | 适用于包含 AI 组件的监管申报;Chai 处于临床前阶段 | 未披露 Chai 与 FDA 的专门互动;工具仍在发现阶段,审批路径不清晰 |
| 同行评议 / 科学验证 | Chai-2 相关主张尚未完成验证 | Chai-1 和 Chai-2 仅以 biorxiv 预印本发布(截至 May 2026 尚未同行评议) | 复现性缺口重大;独立实验验证尚未发表 |
| 生物安全和双重用途风险控制 | 已承认;控制措施没有完整公开说明 | 适用于针对危险病原体靶点的从头抗体生成 | 未发布双重用途风险评估;负责任部署框架细节稀疏 |
所有合规状态均基于公开披露;尚未披露任何与 Chai 的 AI 平台相关的第三方审计、SOC 2 认证或监管文件。
[CE016, CE028, CE029]| 日期 / 阶段 | 功能 / 里程碑 | 状态 | 影响 | 来源 |
|---|---|---|---|---|
| Oct 2024 | Chai-1 开源发布(模型权重 + 代码,Apache 2.0,PyPI 包) | 已完成 | 打下开发者社区和开放核心生态基础 | biorxiv 预印本 + GitHub 发布 |
| Jun 2025 | Chai-2 亮相;向特定合作伙伴开放早期访问;businesswire 公告 | 已完成(早期访问仍在进行) | 锚定商业化模式;首次大规模展示从头抗体设计 | businesswire 新闻稿;Chai-2 biorxiv 预印本(Jul 2025) |
| Nov 2025 | Chai-2 困难靶点预印本;展示全长 mAb 设计 | 已完成(预印本;尚未同行评议) | 范围扩至 GPCR 激动和新表位抗体;提高治疗门槛 | biorxiv 预印本 Nov 2025 |
| Jan 2026 | 宣布与 Eli Lilly 合作;为 Lilly 生物制剂数据定制模型 | 活跃 / 进行中 | 首个具名药企客户验证商业兴趣;释放定制模型路线图信号 | businesswire Jan 2026;techcrunch Jan 2026 报道 |
January 2026 之后的路线图项目没有公开具体披露;模态扩展(多肽、小分子、双特异性抗体)被提作未来方向,但未给出时间表。
[CE001, CE007, CE013, CE031, CE038]5.5 重点观察
06客户
6.1 客户基础、买方与用户分层
Chai Discovery 的客户结构横跨三个运营上不同的分层。第一类是大型制药公司,具体就是 Eli Lilly;它同时扮演买方、用户和付款方。Lilly 合作于 2026 年 1 月 8 日宣布,在多个生物制剂靶点上部署 Chai 的前沿 AI,并包括用 Lilly 专有数据训练一个定制 Chai 模型。这代表了典型的 Chai 商业关系:多项目合作,药企伙伴把 Chai 平台接入自身内部发现工作流。该合作之前有一段已披露的“评估期”,确认大药企客户在承诺前会先对 Chai 做尽调。财务条款没有披露;Chai 采用“负责任部署”政策,选择性把关访问,而不是开放式商业自助服务。 第二类是早期访问生物技术合作伙伴。Chai-2 在 2025 年 7 月发布时明确向“精选合作伙伴”开放合作申请;Chai Series A 领投方 Menlo Ventures 公开表示,“生物技术行业中相当一部分”已经申请 Chai-2 访问。Chai 没有公开点名这些申请方或已确认早期访问合作伙伴。它们代表管线,即潜在未来付费客户;但在正式协议宣布前,不应计入已确认收入。 第三类是 Chai-1 的学术、研究和非商业用户;Chai-1 是 Chai 的开源结构预测模型。Chai-1 可在 GitHub 和 PyPI 免费下载,也可在 lab.chaidiscovery.com 免费网页推理。这些用户增强开发者生态和品牌可信度,但除非转化为商业协议,否则不是付款方。CB Insights 将 Chai 列入 AI 100 2026 的医疗健康与生命科学类别,并指出公司估值在约十五个月内从种子轮 $150 million 增至 Series B $1.3 billion;这进一步验证了 Chai 在药企 AI 市场中的定位,即便其付费客户数量仍有限。 [CU001, CU002, CU003, CU004, CU005, CU006]
| 细分客群 | 代表例子 | 访问渠道 | 付费状态 | 证据基础 |
|---|---|---|---|---|
| 大型药企 | Eli Lilly | 谈判型合作;负责任部署门槛 | 已确认付费合作伙伴 | BusinessWire 新闻稿 Jan 2026;TechCrunch 专访 |
| AI 原生 / 早期访问生物科技公司 | 未具名(Chai-2 早期访问申请方) | 基于申请的 Chai-2 早期访问计划 | 试点 / 付费未确认 | Menlo Ventures 投资方评论;Chai-2 发布新闻稿 |
| 学术 / 研究 | 大学实验室;独立研究者 | GitHub、PyPI、HuggingFace、lab.chaidiscovery.com(免费) | 非付费(免费非商业许可) | GitHub API:1,938 stars;PyPI 包页面;HuggingFace 组织 |
细分定义来自公开公告和开源仓库数据。早期访问生物科技公司的付费状态为推断;除 Eli Lilly 外,尚未公开确认正式付费合作。
[CU001, CU007, CU015, CU016, CU017]| 客户 / 合作伙伴 | 公告日期 | 合作范围 | 财务条款 | 证据质量 |
|---|---|---|---|---|
| Eli Lilly and Company(客户) | January 8, 2026 | 部署 Chai 前沿 AI,覆盖多个生物制剂靶点;基于 Lilly 专有数据训练定制 Chai 模型 | 未披露 | 高——由官方 BusinessWire 新闻稿和 Lilly 声明确认 |
| Chai-2 早期访问生物科技合作伙伴(未具名) | July 2025(开放访问) | 为抗体设计项目提供 Chai-2 平台访问;完整条款未披露 | 未披露 | 低——仅由投资方提及确认其存在;未披露合作伙伴名称 |
| Menlo Ventures(A 轮领投方;Anthology Fund) | August 2025 | 投资方,不是付费客户;称“相当一部分生物科技行业”申请了 Chai-2 访问 | N/A(投资方,不是客户) | 中——投资方评论只能作为需求的二手证据,不是已确认客户 |
| 没有其他具名付费客户 | N/A | 截至 May 2026,未公开宣布其他合作 | N/A | 确认不存在——对新闻稿、CB Insights、Crunchbase 做穷尽搜索,未发现更多名称 |
覆盖范围穷尽所有公开具名关系。未具名早期访问合作伙伴作为一行计入,代表一个未知群体。Lilly 的财务条款未披露。
[CU001, CU002, CU003, CU004, CU007, CU008]六阶段旅程,从初次认知平台到主动多项目部署,说明药企或生物技术客户如何从 Chai-1 开源发现推进到正式 Chai-2 合作协议。Eli Lilly 是唯一确认走完全部六阶段的案例;其他潜在合作方估计处在第二至第四阶段。
[CU002, CU003, CU006, CU007, CU009, CU013]五阶段漏斗,从开源发现到主动生产合作,展示 Chai-1 开发者 / 研究用户向 Chai-2 企业商业合作伙伴的转化级联。可量化阶段尽量使用已确认数据;估计阶段另行标注。
第 2 阶段(访问申请方)和第 3 阶段(早期访问合作伙伴)为估计值;Chai 未披露实际数量。第 1 阶段(GitHub star)为截至 2026 年 5 月 GitHub API 调用的精确值。第 4–5 阶段已确认。
[CU001, CU003, CU007, CU008, CU010, CU017]6.2 开发者采用与开源生态信号
Chai-1 于 2024 年 9 月开源后,已经获得有意义的第三方开发者采用;即使这种采用不直接转化为收入,也能作为商业兴趣的领先指标。截至 2026 年 5 月,chaidiscovery/chai-lab GitHub 仓库累计 1,938 个 star 和 274 个 fork,87 个开放 issue 说明社区互动活跃。该仓库创建于 2024 年 9 月,最近一次 push 在 2026 年 4 月,约二十个月持续开发。这条轨迹与同成熟阶段的其他计算生物学开源项目相比并不逊色。 PyPI 上的 chai_lab Python 包——注意这里是下划线版本,不同于更早的连字符包名——已达到 0.6.1 版本,发布日期为 2025 年 3 月,显示仍在积极维护。Chai Discovery 的 HuggingFace 组织页托管模型权重和文档,为研究社区提供额外分发渠道。PyPI 和 HuggingFace 的采用信号,都是研究者到商业转化漏斗的代理指标:在学术工作流中使用 Chai-1 的开发者,可能成为其机构内部推动企业采用的支持者。 Chai-1 的 biorxiv 预印本在 2024 年 10 月更新到第 2 版,Chai-2 技术报告于 2025 年 7 月作为 biorxiv 预印本发布,显示 Chai 对科学透明度的承诺,也把它与完全封闭的竞争者区分开。但开源模型的非商业许可证限制了创收型部署,意味着庞大的开发者采用基础不会直接变现,除非转成付费合作协议。lab.chaidiscovery.com 的商业网页界面仍向个人研究者免费开放,这进一步限制了非药企用户基础的直接变现。 [CU016, CU017, CU018, CU019, CU020, CU021]
| 信号 | 指标 / 数值 | 日期 | 解读 | 来源 |
|---|---|---|---|---|
| GitHub 星标(chai-lab) | 1,938 stars | May 2026 | 可作为开发者和研究社区认知度的代理指标;显示有活跃关注者 | GitHub API 来源(api.github.com/repos/chaidiscovery/chai-lab) |
| GitHub 分叉(chai-lab) | 274 forks | May 2026 | 显示研究者和开发者正在基于 Chai-1 代码构建或评估 | GitHub API 来源(api.github.com/repos/chaidiscovery/chai-lab) |
| GitHub 未关闭 Issues | 87 open issues | May 2026 | 社区互动活跃;说明用户在提交 bug 和功能请求 | GitHub API 来源(api.github.com/repos/chaidiscovery/chai-lab) |
| PyPI 包版本(chai_lab) | v0.6.1 | March 2025 | 可通过 pip 安装的包仍在维护,显示持续开发节奏 | PyPI 页面(pypi.org/project/chai_lab/) |
GitHub 指标来自 GitHub REST API,时间为 May 22, 2026。PyPI 版本日期来自 PyPI 包页面。没有公开的付费客户或 ARR 增长指标。
[CU017, CU018, CU021, CU022]矩阵从五个证据维度评估四类已识别客户或用户群体:来源独立性、财务确认、部署阶段、结果可见性和总体证据质量。确认用户结构极度偏向非商业用户;只有 Eli Lilly 达到已验证付费合作伙伴门槛。
[CU001, CU007, CU010, CU015, CU017, CU024]6.3 客户留存、集中度风险与证据缺口
Chai 的客户画像存在显著集中度风险。截至 2026 年 5 月,Eli Lilly 是唯一公开具名的付费合作伙伴。公司没有公开披露 ARR、logo 数、留存率、NPS 分数或任何其他客户健康指标。鉴于合作范围包括多个生物制剂靶点和定制模型训练,Lilly 合作很可能是多年期;但合同期限或续约条款没有宣布。第二个具名付费客户缺席,意味着 Lilly 关系若出现恶化,将实质性损害 Chai 的商业轨迹。 留存韧性也被临床转化风险复杂化。截至 2025 年,没有任何 AI 设计药物候选物获得 FDA 批准。Exscientia、BenevolentAI、Recursion 等同行公司经历了知名临床失败,削弱了药企对 AI 药物发现平台的信心。反向来源 loonbio.com 记录称,AI 药物发现累计投入 $60 billion,却没有 FDA 批准;这一动态会让人质疑,缺乏已证明临床证据时,药企客户是否愿意长期为发现阶段 AI 工具付费。从发现到 IND 申报的典型时间线是四到七年,意味着即便最早的 Chai-Lilly 合作产出,最早也不太可能在 2029 年之前产生临床证据。 证据缺口很重要:没有独立第三方审计 Chai 的命中率基准,“负责任部署”访问政策没有公开,早期访问生物技术合作伙伴的名称也没有披露。单一具名客户、无披露收入指标、无临床证据三者叠加,形成了监控义务——投资人应要求在 Series B 资本完全部署之前,按季度报告 logo 数、Chai-2 合作伙伴转化率和 Lilly 合作里程碑进展。 [CU026, CU027, CU028, CU029, CU030, CU031]
| 指标 | 数值 / 状态 | 证据基础 | 风险影响 |
|---|---|---|---|
| Lilly 合作期限 / 续约 | 未披露;按合作范围看,可能为多年期 | 新闻稿提到多个靶点 + 定制模型建设;多年期为推断 | 中——看不到合同续约条款;存在单一客户风险 |
| ARR / 收入 | 未公开披露 | 没有监管文件、新闻稿或投资方披露给出收入数字 | 高——无法评估收入可持续性或增长轨迹 |
| NPS / 客户满意度 | 未公开披露 | 没有任何 Chai 客户的公开调查、案例研究或推荐语 | 中——无法评估续约风险或扩张意愿 |
标为“未公开披露”的所有值表示确认没有公开披露,并不表示指标本身确认不存在。Lilly 多年期时长是推断,不是已披露事实。
[CU027, CU028, CU034]| 风险因素 | 严重性 | 细节 | 缓释因素(如有) |
|---|---|---|---|
| 单一具名付费客户(Lilly) | 高 | 截至 May 2026,已知商业收入 100% 归因于一个合作伙伴 | Chai-2 早期访问管线可能转为付费合作;未披露时间表 |
| 临床转化缺口 | 高 | 尚无 Chai 设计分子进入临床试验;从发现到 IND 的典型周期为 4–7 年 | Lilly 临床开发基础设施深,可能加快候选物推进 |
| 全行业 AI 药物失败记录 | 中 | Exscientia、BenevolentAI、Recursion 经历过临床失败;loonbio 反向分析称已投入 $60B,FDA 批准数为零 | Chai 切入更早的发现阶段,降低近期临床失败风险 |
| 访问政策瓶颈 | 中 | 负责任部署准入可能拖慢获客;SLA 或转化时间表未公开 | 选择性准入可能保护合作质量;Oak HC/FT 和 General Catalyst 的投资方董事会监督提供制衡 |
严重性评级是基于公开证据的定性评估。缓释因素只能部分对冲,无法完全抵消所列风险。
[CU026, CU029, CU030, CU031, CU032, CU033]以三个队列、四个时间窗口观察 Chai 客户管线留存。第 1 行模拟 Chai-2 申请方向正式付费合作伙伴的估计转化漏斗。第 2 行将 Chai-1(2024 年 9 月队列)已确认 GitHub star 增长标准化为 0–100,显示开发者社区建设。第 3 行估计学术用户留存衰减。所有数值均为估计或标准化结果——Chai 未披露任何分部的实际留存百分比。
第 1 行:第 3 个月的 100% 代表早期访问管线中的全部申请方;估计第 6 个月有 25% 获得早期访问(根据 Menlo Ventures 评论和典型制药 BD 周期推断);约 8% 在第 12 个月转为试点;约 3% 在第 18 个月成为正式付费合作伙伴。公司未披露实际转化率。第 2 行:归一化 GitHub star 数(M+3 估计 500、M+6 为 1,000、M+12 为 1,500、M+18 确认为 1,938),以第 18 个月重新索引为 100。第 3 行:学术用户留存根据典型研究软件采用衰减估算;Chai 未披露实际学术留存数据。
[CU027, CU028, CU034, CU036]6.4 重点观察
07风险
7.1 技术与科学风险
Chai 的技术承诺很大胆:公司把 Chai-2 定位为具备原子级精度的从头抗体设计,同时把 Chai-1 定位为开放、领先的多模态结构模型。风险不在于模型没意思,而在于基准质量和治疗质量不是一回事。Chai 的公开证据仍主要来自一组预印本和公司发布材料,而不是独立、同行评议的转化研究。本章最强的外部评论都指向同一个失败模式:即便模型提出了结合体或看似合理的结构,抗体后续仍会因为聚集、表达差、不稳定、剪切、脱靶相互作用或免疫原性而失败。Chai 尚未公布足够的原始湿实验室数据或第三方复现,外部无法独立验证其最具商业意义的主张——接近 20% 的 Chai-2 命中率。这个缺口并不否定公司,但意味着投资人承销的仍是一个部分黑箱的科学到产品转化问题,而不是一个已完全验证的平台。[CR001, CR002, CR003, CR005, CR012, CR013]
| 失败模式 | 可能性 | 严重性 | 缓释成熟度 | 剩余暴露 | 未解决缺口 |
|---|---|---|---|---|---|
| 湿实验命中率尚未被独立复现 | 高 | 高 | 低 | 高——投资逻辑仍依赖公司发布的数据 | 需要盲法第三方复现,以及支撑 Chai-2 性能主张的原始检测表 |
| 结合后的可开发性失败(聚集 / 免疫原性 / 表达差) | 高 | 高 | 中 | 高——许多候选物在初始结合或结构前景之后失败 | 需要项目级失败率数据,覆盖表达、纯度、稳定性和免疫原性筛选 |
| CMC / 无菌 / 杂质 / 稳定性在 IND 前失败 | 中-高 | 高 | 低-中 | 高——FDA 质量要求可能主导进入临床的时间 | 需要具名制造商、放行规格、稳定性研究和杂质控制 |
| 托管实验室访问门槛拖慢自助采用 | 中 | 中 | 中 | 中——登录和负责任部署步骤可能削弱漏斗顶部使用 | 需要从免费用户和研究者转向付费企业或合作伙伴项目的转化漏斗 |
| 开放分发扩大滥用和 IP 泄露暴露面 | 中 | 高 | 低-中 | 中-高——多个软件渠道扩大攻击面和复制风险 | 需要覆盖 GitHub、HF 和 PyPI 的分发政策、滥用监控和制品治理控制 |
运营风险按其在首次临床证明前打断科学到商业化链条的直接程度排序。
[CR011, CR014, CR015, CR016, CR018, CR021]Chai 的核心风险集中在高概率 / 高严重度象限:科学可复现性、可开发性淘汰、客户集中度,与监管质量负担叠加。
发生概率和严重程度是基于已引用公开证据形成的定性投资判断;公司未公开披露概率。
[CR018, CR021, CR026, CR031, CR034, CR041]7.2 监管与质量风险
监管风险不在于 FDA 敌视 AI,而在于 Chai 仍需满足任何治疗开发公司都要承担的质量和临床就绪负担,同时还叠加模型特有的不确定性。FDA 发布了面向软件的 AI 指引,也更广泛讨论了药物开发中的 AI,但两个来源都没有为 AI 设计抗体提供定制审批路径。因此,Chai 应被视为一家恰好使用先进 AI 的生物制剂发现公司,而不是一家可以绕过 CMC、检测验证或 IND 级证据的公司。FDA 自身 CMC 材料强调批次数据、工艺描述、纯度、稳定性,以及无菌或内毒素控制,这些都是安全关键项。Chai 的公开材料没有披露任何先导资产的生产合作伙伴、放行规格集或稳定性包。Lilly 公告证明了市场对发现工作流有商业兴趣,但尚未证明 Chai 能推动一个抗体完成 IND 支持性工作、监管沟通或生产放大。结果是典型的生物技术时点风险:发现主张可能成熟得远快于质量体系和临床证据。[CR019, CR020, CR021, CR022, CR023, CR024]
| 风险 / 规则 / 约束 | 司法辖区 | 状态(2026) | 可能性 | 严重性 | 缓释措施 | 剩余暴露 | 尽调路径 |
|---|---|---|---|---|---|---|---|
| AI 设计抗体没有专门 FDA 路径 | 美国 / FDA / CDER | 已有 AI 指南,但未找到 AI 设计抗体的路径专属公开先例 | 高 | 高 | 把 AI 作为发现基础设施处理;围绕实验验证、模型使用和可比性尽早与 FDA 沟通 | 高——定制化证据要求可能拉长时间线并抬高成本 | 索取关于模型使用、验证和监管定位的任何 pre-IND 或科学建议往来文件 |
| 生物制剂 CMC 准备缺口 | 美国 / FDA / CDER | 未公开披露制造合作伙伴、放行规格或稳定性资料包 | 高 | 高 | 尽早锁定制造对手方、分析方法和批次谱系 | 高——缺少批次或质量数据会卡住 IND 准备 | 审查牵头项目的制造合同、批记录、稳定性计划和杂质控制 |
| 开源 / 许可 / IP 边界风险 | 全球 | Chai-1 使用 Apache 2.0 软件条款,而竞品模型采用限制性权重条款 | 中 | 高 | 把专有价值留在数据、湿实验循环、客户合同和后续闭源模型里 | 中-高——开源代码会压缩差异化,也让所有权边界更复杂 | 审查专利申请、贡献者协议、模型卡披露和商业许可姿态 |
| 生物安全治理收紧 | 美国 / 全球政策 | 独立政策论文主张评估并可能限制先进生物 AI 模型 | 中 | 高 | 落地访问分层、日志记录、红队测试和负责任部署审核 | 中-高——未来治理规则可能拖慢分发或增加合规成本 | 审查负责任部署政策、滥用升级流程,以及任何出口管制或生物安全审查 |
各行按 2026 年投资人视角下的剩余严重性排序。表中同时放入公开监管证据,以及未找到公司特定质量资料包时明确披露的公开缺口。
[CR019, CR020, CR021, CR022, CR023, CR032]7.3 市场与竞争风险
Chai 的公开商业故事仍然很窄。在检索到的材料中,Eli Lilly 是唯一具名药企合作伙伴;Chai 不披露收入、客户数或留存指标,外部无法判断平台是否正在从单一锚点合作故事变成多元化业务。这种集中度很重要,因为市场不会等待。AlphaFold 3 仍是生物分子相互作用预测的基准参照点,Absci、Generate Biomedicines 和 EvolutionaryScale 也都在销售重叠的 AI 生物制剂或前沿蛋白模型能力。竞争会压缩定价、拉长采购周期,并抬高买方证明要求——尤其当药企客户可以追问,为什么 Chai 优于 AlphaFold 类工具、既有生物学团队和竞争 AI 生物技术供应商的组合。战略上的转折点在许可:竞争者的访问条款本身也可能变成护城河或摩擦。简言之,Chai 同时在科学可信度、平台可用性和商业条款上竞争,而公开证据中合作伙伴多元化仍然稀疏。[CR006, CR007, CR008, CR009, CR026, CR027]
| 依赖 | 对手方 / 平台 | 角色 | 集中度 | 失败情景 | 严重性 | 缓释措施 | 剩余暴露 |
|---|---|---|---|---|---|---|---|
| 锚定药企验证伙伴 | Eli Lilly | 旗舰外部验证和潜在收入来源 | 高——唯一具名伙伴 | 项目交付不达预期,或无法扩展为重复合作 | 高 | 拿下更多具名药企客户,避免商业故事只靠单一背书 | 高 |
| 资金提供方 | 当前及未来成长轮投资方 | 在临床证明出现前,为湿实验、算力和招聘供血 | 高 | 复现或 IND 准备就绪前,融资窗口关闭 | 高 | 按里程碑花钱,并争取更多伙伴出资的发现项目 | 高 |
| 分发平台 | GitHub / Hugging Face / PyPI | 托管代码、包和公开模型制品 | 中 | 平台政策、滥用担忧或 IP 争议中断分发 | 中 | 镜像关键制品,并收紧许可 / 合规流程 | 中 |
| 监管守门人 | FDA / 未来审评方 | 定义 pre-IND、IND 和质量体系的证据负担 | 高 | 定制化证据要求延误时间线,或需要更多实验 | 高 | 尽早与监管方沟通,并准备可追踪检测 / CMC 证据 | 高 |
对 Chai 来说,依赖风险不只来自供应商;公司还依赖一个具名合作伙伴、投资人信心、开发者渠道,以及监管方对证据质量的接受。
[CR011, CR024, CR026, CR027, CR028, CR029]Chai 同时依赖一个具名药企合作伙伴、监管接受度、开放软件渠道、私人资本,以及一支创始人权重高的小型技术团队。
[CR011, CR023, CR026, CR037, CR041, CR045]7.4 运营、资本与治理风险
从运营上看,Chai 卡在软件创业公司和治疗公司之间那个昂贵的中间地带。公司已融资不少,并被报道拥有十亿美元以上估值,但公开材料仍未披露现金消耗、现金跑道或单位经济。外部无法判断最近一轮融资是否足以抵达下一个真正降风险的里程碑——独立湿实验室复现、第二个具名药企伙伴,或可见的 IND 支持性进展。通过 GitHub、Hugging Face、PyPI 和托管实验室产品公开分发,确实扩大了触达,但不会自动产生持久企业收入。治理也较集中。公开材料反复把公司聚焦在小规模创始和研究团队上,Mikael Dolsten 加入董事会提高了可信度,却不能解决继任风险。如果一两位核心创始人离开,Chai 可能同时遭遇科学、招聘、融资和合作伙伴信心冲击。因此,投资人不仅要跟踪科学结果,也要跟踪资本效率、组织纵深,以及业务能否超越创始人能量实现扩张的证据。[CR010, CR011, CR028, CR029, CR030, CR035]
| 角色 / 职能 | 依赖或缺口 | 可能性 | 严重性 | 缓释措施 | 尽调路径 |
|---|---|---|---|---|---|
| Joshua Meier / CEO 和联合创始人 | 科学愿景、融资叙事和外部可信度高度集中在一位领导者身上 | 中 | 高 | 记录继任安排,并下放科学 / 商业决策权 | 审查继任覆盖、关键人保险和决策权地图 |
| Jack Dent / 联合创始人 | 公开材料显示,产品、工程和转化优先级较依赖创始人 | 中 | 中-高 | 在创始人以下搭起 VP 梯队和运营节奏 | 审查组织架构、高管招聘计划和跨职能项目负责人归属 |
| Matthew McPartlon 与 Jacques Boitreaud / 研究负责人 | 前沿模型经验似乎集中在少数核心研究人员身上 | 中 | 高 | 加强文档、可复现训练 / 评估流程和留才计划 | 访谈第二梯队研究员,并审查留任 / 归属安排 |
| 董事会 / 风险监督 | 除创始人和 Mikael Dolsten 外,独立治理的公开证据有限 | 中 | 中高 | 增补具备生物制品 CMC 和安全经验的独立董事 | 审查董事会构成、委员会、风险归属和事件升级流程 |
人员风险偏高,因为 Chai 同时做前沿模型研发、生物技术落地和融资,而公司公开认知仍集中在少数具名领导者身上。
[CR035, CR041, CR042]| 风险 | 可监测触发项 | 阈值 / 事件 | 行动含义 |
|---|---|---|---|
| 湿实验室可复现性缺口 | 独立盲法复现 Chai-2 命中率 | 下一次重大融资或商业扩张前仍无第三方复现 | 折价看待模型性能主张,并扩大下行情景 |
| 合作伙伴集中 | 第二个具名药企合作伙伴,或 Lilly 合作范围扩大 | 到 2026 年底仍无新增具名合作伙伴 | 把商业牵引视为集中且脆弱 |
| CMC / IND 准备度 | 具名生产商,以及批次、稳定性或 IND 支持里程碑 | 未见 CMC 对手方或准备资料包 | 假设入临床时间更长、资金需求更高 |
| 资本效率不透明 | 与融资里程碑绑定的烧钱速度或现金跑道披露 | 持续扩张却仍未量化披露烧钱速度或现金跑道 | 更激进地计入稀释和下轮降价融资风险 |
| 治理 / 生物安全控制 | 负责任部署政策、日志记录和滥用审查流程 | 软件开放分发却没有正式治理控制 | 在承认可广泛分发的上行空间前,要求先补上治理 |
这些触发项是投资者监测框架,不是公司指引。未来公司披露、合作伙伴公告或监管证据包都能观察到这些触发项。
[CR018, CR023, CR026, CR029, CR034, CR042]主导失败路径从未经复现的模型主张开始,经可开发性和 CMC 摩擦,传导到合作伙伴证明延迟、收入放慢,并再次带来融资压力。
[CR018, CR021, CR023, CR024, CR026, CR030]7.5 IP、法律与生物安全风险
本章最明确的负面证据集中在 IP 边界设定和生物安全。Chai 对 Chai-1 选择了相对开放的软件姿态,按 Apache 2.0 发布代码和权重,并通过多个开发者渠道分发制品。这个选择加速生态采用,但也削弱了单靠软件封装形成的产品护城河。与此同时,周边市场显示政策环境可能多么不稳定:AlphaFold 3 发布后引发了关于代码访问的公开争议,DeepMind 发布的条款也把权重限定为非商业使用。独立生物安全文献更进一步,认为先进生物 AI 模型可能具备令人担忧的双重用途能力,并可能需要保护措施或限制。对 Chai 而言,这形成双重束缚。访问太开放,误用和泄漏风险上升;访问太严格,商业转化和社区采用放慢。公司需要明确的治理、日志记录和部署控制,才能让这组权衡仍具可投资性。[CR032, CR033, CR037, CR038, CR039, CR040]
7.6 重点观察
08估值
8.1 投资建议与入场纪律
Chai Discovery 不是一个容易做空的投资论点:公司融资金额可观,招到的创始团队具备真实的前沿模型信誉,并说服 Eli Lilly 用专有生物制剂数据开展定制模型合作。公司 2024 年才成立,这些证明点并不常见。但这些证明点不会自动让最近披露的价格变得有吸引力。在 $1.3 billion 的 Series B 估值上,Chai 已经按严肃 AI 生物技术平台定价,尽管公开材料仍未披露收入、ARR、利润率或任何临床阶段资产。因此,合适的建议是跟踪,而不是买入,也不是回避。跟踪意味着技术故事真实,并已有一个高质量战略客户;同时也意味着当前估值给执行失误留下的余地很小。投资人应把 Chai 当作一份押注未来证明的期权,而不是一个已降风险的软件或生物技术运营模型。公司能够证明重复客户需求、更清晰的商业经济性,或 Lilly 之外的转化证据之前,入场纪律应保持严格,并对价格敏感。[CV001, CV004, CV007, CV010, CV011, CV012]
| 维度 | 评估 | 置信度 | 决策含义 |
|---|---|---|---|
| 建议 | 观察 | 中 | 公司质量信号不错,但当前估值支撑还不足以给出买入结论。 |
| 风险评级 | 高 | 高 | 临床前状态、单一具名标杆客户和行业证据缺口让下行风险仍然不小。 |
| 估值立场 | 上一轮 $1.3B 估值偏高 | 中 | 价格已计入比公开来源目前显示更多的商业与转化证据。 |
| 当前锚点 | 2025 年 12 月 Series B,估值 $1.3B | 高 | 任何新一轮融资都应对照该定价点以来的进展来判断,而不是只看技术热度。 |
| 上调触发项 | 第二个标杆伙伴,加上更清晰的商业经济性 | 中 | 更多证据可能支持从观察转向正面结论。 |
| 下调触发项 | 没有新证据却提价 | 中 | 在客户多元化或临床进展前再次重估,会让风险收益变差。 |
截至 2026-05-22,判断明确取决于价格和证据,而不是泛泛的公司质量分。
[CV001, CV010, CV012, CV056, CV057, CV058]决策流把 Chai 的技术前景、有限商业证明、行业风险和当前定价串起来,导向跟踪建议。
本图面向决策而非穷尽所有因素;重点是当前估值尽调中最关键的变量。
[CV001, CV007, CV010, CV011, CV012, CV050]8.2 投资论点与反论点
Chai 的投资论点建立在技术野心和机构验证的罕见组合上。Chai-2 报告的命中率、底层零样本抗体设计框架,以及创始团队来自 OpenAI、Meta 和 Absci 的履历,共同构成一个可信案例:Chai 正在生成式生物制剂设计前沿工作。Lilly 愿意评估 Chai 输出并资助定制模型合作,是最强的外部证明,说明公司模型不只是学术新奇物。反论点同样重要。今天支撑估值的几乎每一个头条事实,都仍位于临床和商业证明的上游:性能主张由公司自报,Lilly 交易经济性未披露,公司在保留来源中只有一个具名大客户,也没有任何 Chai 设计资产进入临床试验的公开证据。换句话说,当前估值承销了大量未来转化:从基准成功转为持久收入,并最终转为治疗证明。这种转化可能发生,但公开来源尚未证明。[CV005, CV006, CV007, CV008, CV009, CV010]
| 维度 | 投资逻辑 | 反向逻辑 | 改变观点的因素 |
|---|---|---|---|
| 技术证据 | Chai-2 声称在 de novo 抗体设计中命中率接近 20%,较此前方法提升超过 100 倍。 | 最强的性能证据仍来自公司自己,尚未转化为临床证据。 | 独立湿实验室复现或更广泛下游数据会强化投资逻辑。 |
| 客户验证 | Lilly 评估了 Chai 设计,并出资开展定制模型合作。 | 保留来源中只看到一个具名标杆客户。 | 第二个大型药企客户会降低集中风险。 |
| 团队质量 | 创始人来自 OpenAI、Meta 和 Absci,有蛋白语言模型和设计模型经验。 | 创始人履历强,不等于一定能转成商业成果。 | 对客户和管线里程碑的持续兑现,比简历更重要。 |
| 资本位置 | 累计融资超过 $225M,给了 Chai 迭代时间。 | 融资规模仍远小于资金最充裕的私营 AI 生物技术同行。 | 可持续证据比单纯多融钱更重要。 |
| 估值 | 从表面市场价值看,Chai 低于 Recursion 和 Generate。 | 这些上市可比公司披露了 Chai 缺少的收入、现金或临床证据。 | 只有 Chai 弥合证据缺口的速度快于估值上升,观点才应转正。 |
反向逻辑聚焦公开来源尚未证明的部分,而不是否定底层技术努力。
[CV005, CV007, CV009, CV012, CV048, CV049]针对 Chai 当前估值点,从证明、商业化、估值支撑和证据质量给出的 IC 式记分卡。
分数是 0-10 的启发式评估,用于衡量当前披露估值下的可投性,不是公司质量的绝对分。
[CV007, CV008, CV010, CV011, CV012, CV043]8.3 估值方法与可比公司
对今天的 Chai 做传统 DCF 是虚假精确,因为公司没有披露收入、利润率、合同期限或产品层面的商业化模型。更干净的方法,是把最近一轮估值与公开和私营参照点交叉核对;这些参照点至少披露收入、现金、情绪或临床阶段中的若干项。在这个基础上,Chai 的 $1.3 billion 估值要求很高。Recursion 市值约 $1.65 billion,且披露收入、现金并具备公开市场透明度。Generate 市值约 $1.79 billion,同时已经拥有一个 Phase 3 先导项目,并披露现金和季度收入。Schrödinger 拥有数亿美元收入和公开财务报表,市值约 $0.99 billion,低于 Chai;Absci 更低,接近 $0.79 billion。私募资本形成也有两面性:Isomorphic 和 Xaira 表明投资人仍会积极资助前沿 AI 生物技术平台,但它们更大的资本基础并不能改变一个事实:Chai 的估值已接近或高于证据深度明显更强的上市公司。[CV013, CV014, CV015, CV016, CV021, CV022]
| 可比对象 | 状态 / 阶段 | 当前价值标记 | 相关原因 | 主要限制 |
|---|---|---|---|---|
| Chai Discovery | 私营,2024 年成立,临床前平台 | 上一轮估值 $1.3B | 被评估的资产。 | 私营公司经济性、股权结构表和收入仍未披露。 |
| Recursion Pharmaceuticals | 上市 AI 药物发现平台 | ~$1.65B 市值;~$1.07B EV | 最接近的上市 AI 平台参照,且披露现金和收入。 | 公开市场波动和大得多的股本让直接比较变复杂。 |
| Schrödinger | 上市软件 + 药物发现平台 | ~$0.99B 市值;~$0.70B EV | 显示在该行业中,已披露收入和软件经济性可以获得什么估值水平。 | 业务组合更宽,也比 Chai 更偏软件。 |
| Absci | 上市 AI 生物制品平台 | ~$0.79B 市值;~$0.67B EV | 生物制品设计叙事更接近,商业证据也仍处早期。 | 收入基数小,且有公开市场风险,作为可比对象证据强度低。 |
| Generate:Biomedicines | 上市科技生物公司,拥有 Phase 3 领先资产 | ~$1.79B 市值 | 有用的上限质量基准,因为它披露现金、收入和 Phase 3 证据。 | 临床阶段公司,资产画像与 Chai 很不同。 |
| Isomorphic Labs | 私营 AI 药物设计平台 | 2025 年融资 $600M,2026 年融资 $2.1B | 显示顶级 AI 生物技术平台能吸引多少资本。 | 保留来源中没有披露估值锚点。 |
| Xaira | 私营 AI 药物发现初创 | 启动时承诺资金 $1B | 说明私营市场对前沿 AI 生物技术公司的组建仍有兴趣。 | 融资规模不等于已定价估值。 |
可比组有意保持部分覆盖:它横跨本估值讨论实际使用的公开和私营参照点,而不是市场上所有 AI 生物学公司。
[CV001, CV013, CV014, CV021, CV022, CV028]条形图对比 Chai 最近披露的估值、部分可比上市公司市值和情景参考点。
数值四舍五入至百万美元,反映 2026-05-21/22 左右的公开市场快照,以及 Chai 最近披露的私募轮次。
[CV001, CV013, CV021, CV028, CV034, CV048]8.4 乐观、基准与悲观情景分析
Chai 的悲观情景不是技术崩盘,而是在商业转化跑通前先遭遇估值压缩。如果 Lilly 仍是唯一有分量的验证点,后续可开发性数据无法走出公司自述材料, 或者 AI 药物发现板块继续带着验证折价交易,Chai 的估值可能更接近 $0.4 billion 至 $0.8 billion。基准情景约 $0.9 billion 至 $1.4 billion,前提是 Lilly 需求可复制而非一次性交易,Chai 守住技术可信度,并在短期内不进临床的情况下, 把基准测试热度转成更持久的客户价值。乐观情景约 $1.8 billion 至 $2.6 billion,不能只靠又一轮亮眼融资。它需要反复出现的大药企需求、 证明 Chai-2 或后续模型能稳定交出达到湿实验室标准输出的证据,以及从设计成功案例走向临床收益或软件式经常性收入的更清晰路径。这些里程碑出现之前, Chai 当前估值更像落在基准情景上半区,而不是明显便宜。[CV043, CV044, CV045, CV052, CV053, CV054]
| 情景 | 估值区间(USD B) | 核心假设 | 概率信号 | 行动含义 |
|---|---|---|---|---|
| 悲观 | 0.4-0.8 | Lilly 仍是孤例,基准测试证据不能顺畅转化为持久客户价值,行业证据折价持续存在。 | 25% | 避免追高;要求价格重置或强得多的证据。 |
| 基准 | 0.9-1.4 | Lilly 转化为可重复需求,技术可信度守住,但近期没有临床资产,也未披露软件经济性。 | 50% | 当前轮次接近该区间上半段;观察而非买入。 |
| 乐观 | 1.8-2.6 | 第二个标杆伙伴落地,Chai-2 级别输出在湿实验室场景泛化,并出现更清晰的经常性经济性路径。 | 25% | 有上行空间,但需要比当前公开来源更多的证据支撑。 |
这些区间是基于公开可比价值和按阶段调整证据搭出的启发式情景带,不是电子表格 DCF。
[CV052, CV053, CV054]相对于 Chai 最近披露的 $1.3B 轮次,展示悲观、基准和乐观估值结果区间。
区间是基于已保留公开证据的情景启发式估计,未建模期权条款或清算优先权。
[CV001, CV052, CV053, CV054]8.5 投资逻辑失效触发点与最后尽调问题
下一轮尽调应盯住四个公开信息仍答不清的问题。第一,Lilly 合作的真实经济条款是什么:首付款、里程碑、排他边界和续约逻辑。第二, 种子轮、Series A 和 Series B 之后,Chai 的股权结构表到底怎样,包括任何会影响普通股结果的优先股堆叠或投资人保护条款。第三, 除 Lilly 之外的客户集中风险多高,是否已有另一家具名设计伙伴接近转化。第四,Chai 的技术基准优势有什么证据能延伸到更广的可开发性、 可制造性并最终进入临床。如果估值继续上行,而这些答案让人失望,投资逻辑就会失效。具体说,缺少第二个标杆客户、无法独立验证 Chai-2 的下游实用性, 或在缺少增量证据时又以高得多价格融资,都应成为投资人后退的理由。现阶段,上行空间真实存在,但为今天的证据支付明天的价格,风险同样真实。[CV012, CV046, CV057, CV058, CV059, CV060]
| 触发项 | 阈值 / 信号 | 重要性 | 行动含义 |
|---|---|---|---|
| 没有第二个标杆客户 | 到下一轮融资周期,Lilly 仍是唯一具名大型伙伴。 | 客户集中度仍然极高,商业故事的可重复性被削弱。 | 维持观察建议;若价格仍上涨则下调。 |
| 基准到产品转化滑坡 | 独立或下游数据未能支持 Chai-2 级别的可开发性或可制造性主张。 | 高溢价技术叙事会失去可信度。 | 估值转向悲观情景区间。 |
| 没有证据却更高估值 | 新融资明显高于上一轮,但没有新增客户或临床证据。 | 未来上行已被提前定价,风险收益变差。 | 在更贵条款下放弃进入。 |
| 行业证据不及预期 | AI 设计药物继续拿不出 Phase II 差异化,或仍没有获批案例。 | 行业证据折价会维持甚至扩大。 | 提高所需安全边际。 |
| Lilly 经济性不及预期 | 条款显示采用浅、范围窄,或续约潜力有限。 | 最强外部验证点会明显变弱。 | 重切商业假设和基准情景区间。 |
触发项按它们对估值支撑的直接损害程度排序,而不是按公关影响排序。
[CV012, CV043, CV044, CV058, CV060]| 主题 | 缺失证据 | 重要性 | 尽调路径 |
|---|---|---|---|
| Lilly 交易经济性 | 首付款、里程碑、排他性、续约权和收入确认处理。 | 没有这些信息,投资者无法判断 Lilly 到底是有意义的商业锚点,还是主要只是信号事件。 | 要求管理层披露,或在 NDA 下进入资料室查看。 |
| 股权结构表与优先股堆叠 | 清算优先权、按比例跟投权,以及任何影响普通股结果的结构。 | 不了解收益如何在不同证券之间分配,私营估值意义有限。 | 审查投资条款清单和最新股权结构分配瀑布。 |
| 客户多元化 | Lilly 之外新增付费或已承诺合作伙伴的证据。 | 单客户故事应拿低于可重复平台的倍数。 | 询问商业账户储备和转化指标。 |
| 技术转化 | 独立湿实验室复现,以及下游可开发性 / 制造证据。 | 基准测试热度必须转化为可用于治疗的输出。 | 索取合作伙伴案例研究或第三方验证资料包。 |
| 商业模式 | 定价、使用模式、毛利率画像和预期合同期限。 | 任何可信的软件式或服务式估值框架,都离不开这些输入。 | 获取带队列经济性的产品和财务模型。 |
这些问题都不是装饰:每一项都会实质影响当前估值有多少可以靠证据而不是乐观情绪来支撑。
[CV010, CV011, CV012, CV046, CV057, CV060]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
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | Chai Discovery was founded in early 2024 in San Francisco, California. | 高 | SO001, SO006, SO018 |
| CO002 | Chai Discovery's stated mission is to transform biology from science into engineering using frontier AI. | 高 | SO001, SO003, SO010 |
| CO003 | The four co-founders began Chai working out of OpenAI's San Francisco offices in the Mission neighborhood in 2024. | 中 | SO006 |
| CO004 | Joshua Meier and Jack Dent originally met in computer science classes at Harvard University. | 高 | SO006, SO007 |
| CO005 | Chai Discovery's core business is building AI foundation models for molecular prediction and de novo therapeutic design, operated as a platform company serving biopharma partners. | 高 | SO002, SO018, SO010 |
| CO006 | Chai describes its vision as building a 'computer-aided design suite' for molecules, analogous to CAD software in mechanical engineering. | 高 | SO002, SO004, SO012 |
| CO007 | Chai Discovery was reported to have approximately 29 employees as of early 2026. | 低 | SO008 |
| CO008 | Chai does not maintain an internal drug pipeline; the company operates as a platform and partner-facing R&D organization. | 高 | SO018, SO004 |
| CO009 | OpenAI CEO Sam Altman messaged Jack Dent to ask whether Meier would be open to collaborating on a proteomics startup, which seeded the founding discussion. | 中 | SO006 |
| CO010 | OpenAI became one of Chai Discovery's first seed investors after the founders started the company in 2024. | 高 | SO001, SO006, SO016 |
| CO011 | Joshua Meier served on OpenAI's research and engineering team in 2018 before pursuing further AI and biotech roles. | 高 | SO005, SO006, SO007 |
| CO012 | At Meta FAIR, Joshua Meier co-led development of ESM1, the first transformer protein-language model, a foundational precursor to modern protein AI. | 高 | SO006, SO007, SO019 |
| CO013 | Meier and co-founder Matthew McPartlon led the AI division at Absci, where they pioneered early research on de novo antibody design contributing to multiple drug candidates in clinical trials. | 高 | SO007, SO016, SO018 |
| CO014 | Jack Dent is co-founder and president of Chai Discovery, bringing experience from Stripe where he was an engineering and product leader building large-scale machine learning systems. | 高 | SO001, SO006, SO018 |
| CO015 | Matthew McPartlon is co-founder and CTO of Chai Discovery, with hands-on expertise in AI for de novo antibody design from his time at Absci. | 高 | SO001, SO018, SO025 |
| CO016 | Jacques Boitreaud is a co-founder of Chai Discovery and served as AI lead at Aqemia, productionizing machine learning tools for small molecule discovery. | 高 | SO001, SO007, SO018 |
| CO017 | Mikael Dolsten, former Chief Scientific Officer of Pfizer where he oversaw more than 150 molecules advancing to clinical trials and 36 approved medicines, joined Chai's board of directors in 2025. | 高 | SO001, SO009, SO014 |
| CO018 | Following the December 2025 Series B, Annie Lamont (co-founder and managing partner of Oak HC/FT) and Hemant Taneja (managing director of General Catalyst) joined Chai's board of directors. | 高 | SO002, SO005, SO012 |
| CO019 | Chai's team background spans OpenAI, Meta FAIR, Stripe, Google X, and Absci; the founding team's collaboration traces to joint Harvard AI research. | 高 | SO002, SO007, SO018 |
| CO020 | Chai Discovery closed a $30 million seed round in approximately September 2024, led by Thrive Capital, OpenAI, and Dimension Capital. | 高 | SO001, SO016, SO020 |
| CO021 | The 2024 seed round valued Chai Discovery at approximately $150 million. | 中 | SO020 |
| CO022 | Chai raised a $70 million Series A in August 2025, led by Menlo Ventures through its Anthology Fund, a joint investment partnership with Anthropic. | 高 | SO001, SO009, SO014 |
| CO023 | The Series A brought Chai's total funding to approximately $100 million and valued the company at approximately $550 million. | 中 | SO014, SO016, SO020 |
| CO024 | Chai Discovery closed a $130 million Series B in December 2025, co-led by Oak HC/FT and General Catalyst, at a post-money valuation of $1.3 billion. | 高 | SO002, SO005, SO013 |
| CO025 | The Series B included participation from Thrive Capital, OpenAI, Dimension, Menlo Ventures, Lachy Groom, Yosemite, Neo, SV Angel, Emerson Collective, and Glade Brook Capital. | 高 | SO002, SO005, SO008 |
| CO026 | The Series B brought Chai's total funding to more than $225 million; Chai's January 2026 press release stated the company had raised 'nearly $230M to date'—a minor rounding difference. | 高 | SO002, SO004, SO017 |
| CO027 | Yosemite, the venture fund co-founded by Reed Jobs (Steve Jobs' son) with an oncology focus, participated in both the Chai Series A and Series B rounds. | 中 | SO008, SO005 |
| CO028 | Emerson Collective, the investment firm of Laurene Powell Jobs, joined the Chai Series B as a new investor. | 高 | SO008, SO005, SO002 |
| CO029 | Chai-1, released in late 2024 as an open-source foundation model for biomolecular structure prediction, established Chai's reputation in the research community at state-of-the-art benchmark performance. | 中 | SO001, SO016, SO020 |
| CO030 | Chai-2, announced June 30, 2025, claims a fully de novo antibody design hit rate of approximately 16–20%—100 times higher than the prior computational state-of-the-art of below 0.1%. | 高 | SO003, SO022, SO005 |
| CO031 | Prior to Chai-2, traditional laboratory antibody discovery methods required screening millions to billions of candidates; prior computational methods achieved hit rates below 0.1%. | 高 | SO003, SO022, SO012 |
| CO032 | Chai-2 accepts only the target antigen and epitope as input and generates all CDRs from scratch in a zero-shot setting, without templates, MSAs, or prior experimental examples. | 高 | SO003, SO022, SO024 |
| CO033 | In Chai's June 2025 preprint, Chai-2 was tested on 52 diverse antibody targets with ≤20 designs per target; approximately 50% of targets (26/52) yielded at least one validated hit. | 中 | SO022, SO003, SO009 |
| CO034 | Chai-2 reportedly solved in a few hours an antibody discovery challenge that had previously consumed more than $5 million in R&D spend, with wet-lab validation completed in under two weeks. | 中 | SO001, SO003, SO025 |
| CO035 | In Chai's November 2025 preprint, more than 86% of Chai-2-designed full-length monoclonal antibodies showed strong developability profiles comparable to approved therapeutics across eight biophysical criteria. | 中 | SO023, SO007, SO009 |
| CO036 | Chai-2's November 2025 preprint demonstrated functional GPCR agonism and highly specific binding of tumor-specific neoepitopes—two historically challenging drug target categories. | 中 | SO023, SO007 |
| CO037 | Chai enforces a 'Responsible Deployment' policy, offering selective partner access to the Chai-2 platform rather than open commercial availability. | 中 | SO020, SO001 |
| CO038 | Chai Discovery and Eli Lilly announced a collaboration on January 8, 2026, under which Lilly will deploy Chai's AI platform to design novel biologic therapeutics across multiple drug targets. | 高 | SO004, SO006, SO015 |
| CO039 | Under the Chai-Lilly collaboration, Chai will develop a purpose-built AI model trained exclusively on Lilly's large-scale proprietary datasets and tailored to Lilly's discovery workflows. | 高 | SO004, SO019, SO015 |
| CO040 | As of May 2026, no Chai-designed therapeutic molecule has entered human clinical trials. | 高 | SO011, SO012, SO018 |
| CO041 | The AI drug discovery sector attracted more than $60 billion in venture capital since 2015 yet had not produced a single FDA-approved AI-designed drug as of early 2025, with high-profile clinical failures across first-generation AI biotechs. | 中 | SO011, SO021 |
| CO042 | Competitor Isomorphic Labs, an Alphabet subsidiary led by Nobel laureate Demis Hassabis, raised $600 million in March 2025 in its first external funding round. | 中 | SO008, SO021 |
| CO043 | General Catalyst publicly projected that early pharma adopters of AI drug design tools such as Chai's may see first-in-class biologics entering clinical trials by end of 2027. | 中 | SO006, SO007 |
| CO044 | The Chai-Lilly collaboration was described by at least one analyst as among the largest AI software deals in biotech; its financial terms were not publicly disclosed. | 中 | SO019, SO006 |
| CO045 | In Chai's June 2025 preprint, Chai-2 achieved a 68% wet-lab success rate in miniprotein binder design, routinely yielding picomolar binders. | 中 | SO022, SO003 |
| CM001 | Global pharmaceutical R&D spending reached approximately $300 billion in 2025, nearly double the 2016 level, with annual growth decelerating to approximately 1.7%. | 高 | SM014, SM003 |
| CM002 | The AI drug discovery software market represents approximately 1.5% of the $194 billion global pharma R&D spend, making it a small but fast-growing slice of total industry investment. | 中 | SM019 |
| CM003 | The narrowest market definition for AI drug discovery covers only AI-enabled software and services materially supporting target identification, hit/lead generation, lead optimization, de novo design, and preclinical candidate selection before IND; it explicitly excludes clinical trial AI, manufacturing AI, pharmacovigilance, and commercial analytics. | 高 | SM019, SM009, SM001 |
| CM004 | Grand View Research estimates the global AI in drug discovery market at $2.35 billion in 2025, projected to reach $13.77 billion by 2033 at a CAGR of 24.8%; North America holds 52.85% market share. | 高 | SM001, SM023 |
| CM005 | Mordor Intelligence estimates the AI in pharmaceutical R&D market at $3.30 billion in 2025, growing to $4.36 billion in 2026 and reaching $17.66 billion by 2031 at a CAGR of 32.25%; software led with 57.34% share in 2025. | 高 | SM003, SM011 |
| CM006 | Fortune Business Insights estimates the AI drug discovery market at $4.46 billion in 2025, $5.00 billion in 2026, and $12.56 billion by 2034 at a CAGR of 12.20%; North America held 65.93% market share in 2025. | 中 | SM011 |
| CM007 | Towards Healthcare / Global Market Insights estimates the AI in drug discovery market at $24.51 billion in 2026, growing to $160.49 billion by 2035 at a CAGR of 23.22%; this broad estimate includes clinical trial AI and manufacturing AI, making it non-comparable to narrow platform estimates. | 中 | SM021, SM018 |
| CM008 | Business Research Insights estimates the AI for pharma and biotech market at $2.68 billion in 2026, growing to $8.67 billion by 2035 at a CAGR of 13.95%. | 中 | SM024 |
| CM009 | Axis Intelligence estimates the narrow AI drug discovery software market at $1.94 billion in 2025 and $2.6–2.8 billion in 2026, with a CAGR of approximately 27% to reach $16.49 billion by 2034; this estimate is based on verified AI program involvement rather than broad self-reported categories. | 中 | SM009 |
| CM010 | Analyst estimates for the 2026 AI drug discovery market range from $1.94 billion to $24.51 billion—a 9x spread—driven primarily by definitional scope differences; narrow platform-only estimates cluster between $2.3 billion and $5.0 billion, while broad ecosystem estimates exceed $10 billion. | 高 | SM004, SM009, SM001, SM011, SM021 |
| CM011 | The global antibody discovery market was valued at $9.78 billion in 2025 and is projected to reach $10.75 billion in 2026 at a CAGR of 9.8%, and $15.79 billion by 2030 at a CAGR of 10.1%. | 高 | SM025, SM012, SM002 |
| CM012 | Mordor Intelligence estimates the antibody discovery market at $9.09 billion in 2025, growing to $15.45 billion by 2030 at an 11.3% CAGR; pharma and biopharmaceutical companies held 48.3% market share in 2024 and North America commanded 41.5%. | 高 | SM002, SM016 |
| CM013 | AI/ML-enabled antibody discovery platforms are forecast to grow at a 22.4% CAGR between 2025 and 2030, more than twice the rate of the overall antibody discovery market (10.1–11.3%); contract and outsourced discovery models are growing fastest at 17.3% CAGR. | 高 | SM002, SM016 |
| CM014 | The AI protein structure prediction market is estimated at approximately $1.80 billion in 2025, growing to $2.33 billion in 2026 at roughly 30% CAGR, and reaching $6.62 billion by 2030. | 中 | SM017, SM013 |
| CM015 | The global pharmaceutical pipeline approached 23,000 drug candidates in development in 2026, supported by over 7,000 companies with active pipelines, up nearly fourfold since 2001. | 高 | SM014, SM007 |
| CM016 | Average clinical trial duration extended to over 100 months in 2024 for the first time, up from 93 months four years earlier, indicating worsening industry-wide development efficiency despite rising R&D budgets. | 中 | SM014 |
| CM017 | Developing a new molecular entity costs an average $2.8 billion (capital-adjusted, including failures) and takes 12–15 years from target hypothesis to regulatory approval, with roughly 90% of Phase I candidates never reaching patients. | 高 | SM010, SM001 |
| CM018 | Branded drugs generating more than $180 billion in annual U.S. revenues face loss of exclusivity between 2024 and 2030, creating urgent board-level pressure to replenish pipelines faster than traditional timelines allow. | 高 | SM010, SM003 |
| CM019 | The Phase I success rate for AI-enabled emerging biopharma programs was 75% for the most recent three-year window per IQVIA Global R&D Trends 2026, compared to 40–65% for traditional programs—a substantial advantage concentrated within the EBP segment. | 高 | SM007, SM005 |
| CM020 | Phase II success rates for AI-enabled EBP programs track on par with non-AI-enabled peers, indicating that AI is not simply accelerating poorly-validated candidates into Phase I; the benefit is visible in Phase I but does not yet translate to improved overall industry success rates. | 中 | SM007 |
| CM021 | No AI-designed drug had received FDA approval as of early 2026; over 200 AI-enabled drug candidates were in clinical development globally, with the most advanced (Insilico Medicine's rentosertib) completing Phase IIa in June 2025. | 高 | SM009, SM005, SM007 |
| CM022 | 69% of pharmaceutical companies were investing in AI as of 2026, surpassing cloud computing and other digital initiatives, with 30% of all new 2025 drug discoveries incorporating AI technologies. | 中 | SM005, SM020 |
| CM023 | AI adoption can reduce preclinical R&D costs by 25–50% and accelerate development timelines by up to 60%; multiple analysts cite a compression of preclinical candidate development from 5–6 years to 12–18 months. | 中 | SM005, SM008, SM004 |
| CM024 | The FDA reviewed 170 AI-related drug development submissions in 2022, up from only 14 in 2020, reflecting rapid AI adoption across pharma R&D workflows and growing regulatory familiarity. | 高 | SM003, SM009 |
| CM025 | In January 2026, the FDA and EMA jointly issued ten guiding principles for AI practices in drug development, providing the first substantive joint regulatory clarity for AI-enabled drug discovery and reducing investment uncertainty. | 高 | SM003, SM022 |
| CM026 | Large pharmaceutical companies are the highest-spending buyers of AI drug discovery tools, allocating $100 million to $500+ million annually to AI initiatives including platform partnerships, internal build, and equity investments; Eli Lilly alone signed over $3.75 billion in AI drug discovery deals in Q1 2026. | 中 | SM024, SM019 |
| CM027 | Contract research organizations are the fastest-growing end-user segment for AI pharmaceutical R&D at a 33.15% CAGR through 2031, as they integrate AI capabilities to expand discovery service offerings to pharma clients. | 中 | SM003, SM015 |
| CM028 | Pharmaceutical and biotechnology companies held 59.45% of AI pharma R&D spend in 2025, with pharma/biotech companies also accounting for 59.19% of total segment revenue per Grand View Research; the remaining share goes to CROs and academic/research institutes. | 中 | SM003, SM023 |
| CM029 | Partnership deal structures between pharma buyers and AI discovery platforms include research collaborations, equity investments, milestone-based licensing agreements, and joint research programs; milestone and licensing models are most common for pre-commercial AI platforms. | 中 | SM015, SM019 |
| CM030 | AI drug discovery venture capital funding surged to $5.7 billion in 2025, up 78% from $3.2 billion in 2024, with annualized 2026 projections of $7.2–8.8 billion; twelve AI deals exceeded $200 million in 2024 alone. | 中 | SM009, SM003 |
| CM031 | Data quality, not data volume, is the primary implementation barrier for AI in drug discovery; organizations routinely have sufficient data but struggle with curation, contextualization, and alignment to specific discovery questions, causing expensive AI initiatives to stumble on foundational data issues. | 高 | SM006, SM024 |
| CM032 | Only 22% of life sciences leaders have successfully scaled AI as of 2025–2026 despite high investment, with fewer than 10% reporting significant returns, indicating that the majority of pharma organizations remain at the pilot or evaluation stage. | 中 | SM022, SM024 |
| CM033 | Main adoption barriers cited in industry surveys include data privacy concerns (40%), integration challenges (36%), talent shortage (32%), and algorithmic/regulatory concerns (28%); talent shortage—requiring professionals who understand both AI and drug development—is particularly acute. | 中 | SM024, SM006 |
| CM034 | Switching costs for AI drug discovery platforms are high, involving deep data integration, staff retraining, regulatory validation of model outputs, and change management across chemistry and biology departments, often requiring 2–4 year organizational commitments. | 中 | SM006, SM022 |
| CM035 | AlphaFold 2's mapping of more than 200 million protein structures served as a foundational enabler for AI drug discovery, boosting hit identification rates by approximately 50% and making AI protein design tools mainstream in pharma R&D workflows by 2024. | 高 | SM008, SM013 |
| CM036 | Software platforms led AI pharma R&D market share at 57.34% in 2025, while deep and generative learning technologies represented the fastest-growing technology subsegment at 32.79% CAGR through 2031. | 中 | SM003 |
| CM037 | The 9x spread in AI drug discovery market estimates ($1.94B–$24.51B in 2026) reflects definitional differences: narrow analysts count only pre-IND AI software; broad analysts absorb clinical trial operations, manufacturing AI, and pharmacovigilance into the same figure, creating incomparable data points. | 高 | SM004, SM009, SM019, SM001 |
| CM038 | Despite high Phase I success rates for AI-enabled EBP programs (75%), overall industry-wide drug development success rates were unchanged from the prior year per IQVIA 2026; the AI benefit is detectable within the emerging biopharma segment but has not yet improved the broader pharmaceutical industry's clinical performance metrics. | 高 | SM007, SM005 |
| CP001 | The AI drug discovery competitive landscape as of May 2026 contains five functionally distinct competitor segments: (1) AlphaFold-lineage structure prediction and generalist AI platforms, (2) full-stack AI drug discovery orchestration platforms, (3) generative biology companies targeting protein and antibody design, (4) physics-plus-ML hybrid incumbents, and (5) open-source and academic alternatives. | 中 | SP023, SP024 |
| CP002 | Isomorphic Labs raised approximately $600 million in a Series A funding round in November 2024, backed by Alphabet, making it the best-capitalized pure-play AI drug discovery company. Disclosed pharma partners include Eli Lilly and Novartis. | 中 | SP001, SP023 |
| CP003 | Generate:Biomedicines' lead candidate GB-0895, an AI de novo designed anti-TSLP antibody for severe asthma, has entered Phase 3 clinical trials, making Generate:Biomedicines the most clinically advanced generative protein design company as of May 2026. | 高 | SP005, SP006 |
| CP004 | Recursion Pharmaceuticals, having completed its merger with Exscientia in 2025, holds over 50 petabytes of proprietary experimental data generated via BioHive-2 robotic infrastructure in partnership with NVIDIA, processing 2.2 million biological samples per week. | 中 | SP011, SP012 |
| CP005 | AbSci's ABS-201, designed de novo using AI, became the first AI de novo antibody to enter human Phase 1 clinical trials as of 2025, establishing clinical proof-of-concept ahead of Chai Discovery and setting industry benchmarks for AI-designed antibody candidates. | 高 | SP014, SP015 |
| CP006 | Insilico Medicine's ISM001-055, an AI-designed small molecule drug for idiopathic pulmonary fibrosis, has reached Phase 2 clinical trials—the furthest any AI-designed drug candidate has advanced in clinical development as of May 2026. | 中 | SP008, SP009 |
| CP007 | Schrödinger has operated for over 35 years, holds licensing relationships with more than 1,750 pharmaceutical, biotechnology, and materials science customers, and is publicly listed on NASDAQ (SDGR), giving it the deepest enterprise switching-cost moat in the computational drug discovery sector. | 中 | SP002, SP003 |
| CP008 | Isomorphic Labs' platform is primarily optimized for small molecule drug design, applying AlphaFold-lineage models to molecular generation—a modality focus that differs from Chai's biologics-first (antibody, nanobody, miniprotein) approach. | 中 | SP001 |
| CP009 | Generate:Biomedicines has experimentally generated and tested over 42,000 designed proteins using its Generative Biology™ platform, and operates over 140,000 square feet of physical wet-lab space for integrated design-make-test capability. | 中 | SP005, SP006 |
| CP010 | AbSci claims a 6-week design-to-characterization cycle through its ACE Assay and SoluPro expression system, directly competing with Chai-2's rapid de novo antibody design workflow on cycle time as a buying criterion. | 中 | SP014, SP015 |
| CP011 | The Recursion–Exscientia merger, completed in 2025, created a combined platform spanning phenomics-based target discovery (Recursion) and AI-chemistry candidate design (Exscientia), covering more of the drug discovery workflow end-to-end than any other AI drug company. | 中 | SP011, SP022 |
| CP012 | Insilico Medicine operates the Pharma.AI platform comprising three integrated modules: Biology42 for target identification, Chemistry42 for molecular design, and Medicine42 for clinical trial optimization—covering the full drug discovery pipeline end-to-end. | 中 | SP008, SP009 |
| CP013 | Schrödinger's platform differentiates from pure deep-learning competitors by combining physics-based computational methods (FEP+ binding free energy, Glide docking, BioLuminate biologics modeling) with ML layers, providing interpretability and validated accuracy on small molecules that ML-only platforms cannot replicate. | 中 | SP002, SP003, SP026 |
| CP014 | Generate:Biomedicines has reached IPO-filing status and disclosed Phase 3 pipeline data, positioning it as the first generative biology company approaching commercial-stage clinical validation. | 中 | SP005, SP007 |
| CP015 | The AlphaFold Database (alphafold.ebi.ac.uk), maintained by EMBL-EBI with Google DeepMind and NVIDIA, provides structure predictions for over 200 million proteins under a CC-BY-4.0 license, and added protein complex structures in a March 2026 update—permanently commoditizing protein structure prediction as a distinct commercial value layer. | 中 | SP017 |
| CP016 | AlphaFold3, released by Google DeepMind, predicts protein complexes including protein-small molecule and protein-nucleic acid interactions, but restricts model weight access to non-commercial use only (CC-BY-NC-4.0), requiring explicit Google DeepMind approval for download—limiting its deployability for pharma self-hosting. | 中 | SP019 |
| CP017 | Boltz-2 (github.com/jwohlwend/boltz), released under the MIT License, explicitly benchmarks its structure prediction performance against Chai-1 and additionally predicts binding affinities—a capability Chai-1 does not provide—representing a direct commoditization risk to Chai's freely available structure prediction value proposition. | 中 | SP018 |
| CP018 | ESMFold (github.com/facebookresearch/esm), developed by Meta AI under MIT license, enables protein structure prediction from a single amino acid sequence without requiring multiple sequence alignment, providing rapid inference useful for high-throughput early-stage screening. | 中 | SP020 |
| CP019 | OpenFold (github.com/aqlaboratory/openfold), available under Apache 2.0 license from Columbia University's AQ Laboratory, provides a trainable reimplementation of AlphaFold2 that enables academic groups to fine-tune structure prediction models on proprietary data. | 中 | SP021 |
| CP020 | Internal pharma AI teams at major pharmaceutical companies represent a major status-quo substitute for external AI design platforms; large pharma investment in internal computational biology infrastructure reduces dependence on vendors and compresses the addressable market for platforms like Chai Discovery. | 中 | SP023, SP024 |
| CP021 | Chai Discovery's primary structural competitive advantages as of May 2026 are: (1) zero-shot de novo antibody design capability via Chai-2 validated on challenging targets; (2) the open-weight Chai-1 model driving community adoption; and (3) the Eli Lilly partnership validating commercial readiness—but all three remain early-stage and subject to displacement. | 中 | SP023, SP025 |
| CP022 | The Eli Lilly partnership announced in January 2026 provides Chai Discovery with commercial validation and access to real-world biologics design contexts, but deal economics and data sharing terms are not publicly disclosed, limiting independent assessment of its strategic value. | 中 | SP023, SP024 |
| CP023 | Open-source commoditization is a material competitive risk for Chai Discovery: Boltz-2's MIT license and direct benchmarking against Chai-1 means that a production-deployable open-source alternative to Chai-1 exists without licensing fees, undermining Chai-1's role as an adoption driver and competitive differentiator. | 中 | SP018, SP025 |
| CP024 | Recursion's 50+ PB proprietary experimental dataset, generated via BioHive-2 robotic infrastructure co-developed with NVIDIA, represents a data moat that is structurally difficult for smaller competitors to replicate due to the capital intensity of building equivalent physical screening infrastructure. | 中 | SP011, SP012 |
| CP025 | Schrödinger's switching cost advantage stems from deep integration of its LiveDesign collaborative platform into pharmaceutical computational chemistry workflows over 35+ years, creating organizational inertia and workflow dependencies that are difficult to displace. | 中 | SP002, SP003 |
| CP026 | AbSci's first-mover position in clinical-stage AI de novo antibodies—ABS-201 in Phase 1— may establish pharma partner expectations, regulatory pathway precedents, and commercial benchmarks before Chai files its first IND, creating a meaningful first-mover advantage in the de novo antibody design market. | 中 | SP014, SP016 |
| CP027 | The AlphaFold Database's CC-BY-4.0 license and coverage of 200M+ protein structures means that structural information for essentially all known proteins is now freely accessible, permanently commoditizing protein structure prediction as a discrete commercial value proposition independent of any particular platform. | 中 | SP017, SP018 |
| CP028 | Generate:Biomedicines operates over 140,000 square feet of physical wet-lab infrastructure, providing an integrated design-make-test capability that Chai Discovery's capital-light, partner-dependent validation model does not replicate. | 中 | SP005 |
| CP029 | Isomorphic Labs' Alphabet-backed capital position (approximately $600 million raised) exceeds Chai Discovery's cumulative capital (~$200 million across seed, Series A, and Series B) by a factor of approximately 3x, enabling substantially greater investment in model development, pharma business development capacity, and co-development program scale. | 中 | SP001, SP023 |
| CP030 | The Recursion–Exscientia merged entity has created a combined platform covering phenomics and AI-chemistry, but remains predominantly focused on small molecules; its direct threat to Chai's biologics design niche is currently limited by modality focus. | 中 | SP011, SP022 |
| CP031 | Isomorphic Labs has disclosed pharma drug discovery co-development partnerships with both Eli Lilly and Novartis, establishing a two-partner commercial validation portfolio that directly competes with Chai's single Eli Lilly partnership for signaling commercial credibility. | 中 | SP001 |
| CP032 | Schrödinger is publicly listed on NASDAQ (SDGR), providing it with access to public equity markets for funding, analyst coverage, and transparent revenue reporting—advantages that make its financial position and competitive trajectory more readily assessable than pre-commercial peers including Chai Discovery. | 中 | SP002 |
| CP033 | Insilico Medicine has obtained 13 IND approvals across more than 40 discovery programs, spanning oncology, fibrosis, immunology, and infectious disease—establishing the broadest IND track record of any AI-native drug discovery company as of May 2026. | 中 | SP009 |
| CP034 | No pure-play AI drug discovery company has received FDA drug approval as of May 2026, making clinical validation the key differentiating milestone that the entire AI drug discovery sector is racing toward. | 中 | SP023, SP024 |
| CP035 | Insilico Medicine's ISM001-055 for idiopathic pulmonary fibrosis is the furthest-progressed AI-designed drug in clinical development as of May 2026, having reached Phase 2—demonstrating that AI drug design can produce viable clinical candidates across modalities even if no drug has achieved final regulatory approval. | 中 | SP008, SP010 |
| CP036 | Three independent, high-capability open-source protein structure and design tools are available under permissive licenses as of May 2026: Boltz-2 (MIT), ESMFold (MIT), and OpenFold (Apache 2.0)—all functional alternatives to commercial structure prediction at zero licensing cost. | 中 | SP018, SP019, SP020, SP021 |
| CP037 | Chai-2's key claimed differentiator is zero-shot de novo antibody design validated on challenging targets as documented in its published technical report—a capability that AbSci and Generate:Biomedicines also claim but through structurally different approaches (ACE Assay-based experimental iteration vs. Generative Biology™ directed evolution). | 中 | SP024 |
| CP038 | Recursion's BioHive-2 supercomputer, co-developed with NVIDIA, enables processing of 2.2 million biological samples per week through automated robotic lab infrastructure, providing a scale of experimental data generation that pure AI-design platforms without physical infrastructure cannot match. | 中 | SP011, SP012 |
| CP039 | Chai Discovery's competitive position requires advancing IND candidates to close the first-mover gap versus AbSci (Phase 1) and Generate:Biomedicines (Phase 3); without clinical data, commercial pharma partnerships will remain limited in scope and credibility relative to clinically validated peers. | 中 | SP023, SP025 |
| CP040 | Exscientia, before its merger into Recursion, had three AI-designed drug candidates reach Phase 1 clinical trials, all of which were subsequently discontinued—illustrating that AI-generated clinical candidates can face significant efficacy and safety attrition even when the AI design process is technically successful. | 中 | SP022, SP023 |
| CI001 | Chai-1 is available for free for all users—including commercial users—through the Chai Discovery web interface; model weights are available for non-commercial use under the Apache 2.0 open-source license. | 高 | SI001, SI017 |
| CI002 | Chai-2, the company's flagship de novo antibody design system, is not publicly available and is governed by a "Responsible Deployment" policy under which only selected pharma partners receive access to the platform. | 高 | SI019, SI008 |
| CI003 | The Eli Lilly collaboration announced January 8, 2026 is Chai Discovery's first and only publicly confirmed commercial revenue-generating partnership as of May 2026. | 高 | SI007, SI008, SI009 |
| CI004 | Under the Lilly collaboration, Chai will develop a custom AI model trained exclusively on Lilly's large-scale proprietary datasets and tailored to Lilly's discovery workflows, and Lilly will also receive access to Chai's core platform models. | 高 | SI007, SI009 |
| CI005 | The financial terms of the Eli Lilly collaboration—including any upfront fee, annual license fee, milestone payments, and royalty provisions—have not been publicly disclosed by either Chai Discovery or Eli Lilly. | 高 | SI007, SI008, SI009 |
| CI006 | General Catalyst publicly projected that early pharma adopters of AI drug design platforms such as Chai's may see first-in-class biologics entering clinical trials by end of 2027, implying investor expectation of revenue model advancement within the Series B horizon. | 中 | SI020, SI008 |
| CI007 | Chai's stated commercial vision—a "computer-aided design suite" for molecules—signals ambitions toward broad platform access, but as of May 2026 commercial execution remains selective, partner-gated, and opaque in financial terms. | 中 | SI006, SI008 |
| CI008 | Industry-standard AI drug discovery platform deals typically involve four economic layers: upfront platform licensing fees, custom model development fees, milestone payments on candidate advancement, and long-horizon royalty or co-development participation. | 中 | SI004, SI014 |
| CI009 | No ARR figure, contract count, paying-customer number, or revenue metric for Chai Discovery has been publicly confirmed by the company or any independent third party as of May 2026. | 中 | SI005, SI011 |
| CI010 | Chai Discovery was reported to have approximately 29 employees as of early 2026, per indirect reporting from BuiltInSF; this figure has not been officially confirmed by the company in any public disclosure. | 低 | SI021, SI008 |
| CI011 | At SF AI talent rates of $250,000–$400,000 fully loaded per employee, Chai's estimated ~29-person team implies annual payroll burn of approximately $7–12 million, representing the largest single cost line. | 低 | SI021, SI015 |
| CI012 | Chai's codebase is described as entirely homegrown with no off-the-shelf large language models, suggesting a senior AI research team skewing toward the upper range of compensation brackets. | 中 | SI008, SI019 |
| CI013 | Chai Discovery operates no owned wet-lab or experimental infrastructure; all validation of computationally designed molecules is conducted by pharma partners, making Chai's model structurally capital-light compared to competitors with owned laboratory space. | 高 | SI007, SI016 |
| CI014 | The absence of wet-lab infrastructure enables Chai to operate at a fraction of the capital intensity of Generate:Biomedicines (140,000 sq ft of wet lab) and AbSci (77,000 sq ft of wet lab), creating a potential high gross-margin software model. | 高 | SI016, SI013 |
| CI015 | Training frontier AI biology models comparable to Chai-2's capabilities requires substantial GPU compute infrastructure with estimated training run costs ranging from $1 to $15 million per major training run, plus ongoing inference costs. | 低 | SI014, SI015 |
| CI016 | The Lilly custom model development agreement will generate additional compute costs for Chai—custom model training on Lilly's proprietary data represents incremental compute expenditure beyond Chai's base platform development. | 中 | SI007, SI009 |
| CI017 | Big pharma companies collectively spend over $200 billion annually on R&D, with AI software tools for discovery representing a small but rapidly growing share; individual platform licensing deals are typically valued in the $1–20 million range per engagement. | 中 | SI015, SI025 |
| CI018 | Chai Discovery has raised approximately $230 million in total capital: $30 million seed (September 2024), $70 million Series A (August 2025), and $130 million Series B (December 2025), per official press releases. | 高 | SI006, SI010, SI026 |
| CI019 | The $130 million Series B press release stated proceeds would be used to "accelerate research and product development, and expand commercialization efforts"—language consistent with transitioning from R&D-only to commercial infrastructure build-out. | 高 | SI006, SI022 |
| CI020 | At an estimated burn rate of $20–35 million per year, the $130 million Series B provides approximately three to six years of runway from the December 2025 close, before accounting for any revenue from partnerships. | 低 | SI006, SI015 |
| CI021 | Chai's investor syndicate—including Oak HC/FT, General Catalyst, Thrive Capital, OpenAI, and Menlo Ventures—represents tier-one institutional capital with significant reserves and several investors having participated in multiple rounds, materially reducing near-term refinancing risk. | 高 | SI006, SI020, SI022 |
| CI022 | S&P Global Market Intelligence data for 2025–2026 indicates biopharma venture capital continues to concentrate in AI-enabled preclinical platforms, with Series B+ deals increasingly requiring credible commercialization evidence—a gate Chai has partially cleared through the Lilly partnership. | 中 | SI012, SI013 |
| CI023 | Loon Bio's sector analysis argues that more than $60 billion in AI drug discovery venture capital since 2015 had produced zero FDA-approved AI-designed drugs as of early 2025, raising questions about whether investor patience for the preclinical-to-clinical transition may be approaching its limits. | 中 | SI011, SI017 |
| CI024 | Chai must demonstrate a credible path toward either clinical milestones or meaningful platform revenue within the current financing cycle to avoid a valuation reset at its next fundraise, given no clinical-stage molecules as of May 2026. | 中 | SI011, SI005 |
| CI025 | Drug discovery news analysis of the AI hit-to-clinical-candidate gap identifies the transition from in silico design to validated clinical candidate as the most capital- intensive and failure-prone phase—a gap Chai has not yet navigated with any publicly disclosed program. | 中 | SI005, SI014 |
| CI026 | Chai's $130M Series B and $1.3B valuation lags Isomorphic Labs' $600M external raise by approximately 3× in absolute capital but exceeds most comparable AI drug discovery companies in valuation efficiency relative to years from founding. | 中 | SI010, SI013 |
| CI027 | Chai Discovery's financial disclosures are limited to funding amounts, valuation, and investor names; revenue, ARR, burn rate, deal economics, compute spend, margin structure, and balance sheet are entirely undisclosed as of May 2026. | 中 | SI009, SI011 |
| CI028 | As the sole publicly confirmed commercial deal, the Lilly contract's financial structure defines Chai's current revenue trajectory; without its terms, diligence cannot estimate Chai's actual commercial traction or validate the implied $1.3B valuation. | 高 | SI007, SI008, SI010 |
| CI029 | Eli Lilly, as a public company subject to SEC reporting, may reference the Chai collaboration in financial filings if it meets materiality thresholds, but no Lilly SEC disclosure has confirmed specific Chai deal economics as of May 2026. | 中 | SI007, SI009, SI031 |
| CI030 | The pda.org AI drug discovery analysis indicates that AI platform deals typically involve milestone-based payment structures tied to discovery outcomes, making early-stage collaboration economic value highly contingent on candidate advancement. | 中 | SI014, SI004 |
| CI031 | Chai has not disclosed an official headcount, compute spend, sales pipeline, or contract count; the ~29-employee figure from BuiltInSF is an indirect estimate with low confidence. | 中 | SI021, SI005 |
| CI032 | Maven Bio's analysis of big pharma R&D capital allocation shows AI software tools for discovery typically command annual platform fees of $1–20 million per large pharma engagement, providing context for the likely range of Chai's Lilly deal economics. | 中 | SI015, SI025 |
| CI033 | The drugdiscoverynews.com analysis of AI-transformed drug discovery economics highlights that AI platform companies must negotiate favorable milestone and royalty terms early because pharma partners become more sophisticated about computational contribution valuations as the sector matures. | 中 | SI004, SI014 |
| CI034 | The transition from an AI-generated molecular hit to a clinical candidate requires two to three orders of magnitude more capital than the discovery phase alone; Chai's model defers this cost to partners, limiting current burn but also capping near-term economic upside from candidate advancement. | 中 | SI005, SI016 |
| CI035 | If pharma partners do not advance Chai-designed programs into clinical development, Chai's revenue ceiling remains bounded by platform fees and annual licensing, with milestone and royalty income unrealized—a structural revenue risk given zero current clinical-stage molecules. | 中 | SI011, SI005, SI014 |
| CI036 | S&P Global Market Intelligence's 2026 biopharma VC outlook confirms that AI-enabled preclinical platforms continue to attract disproportionate venture capital in 2026, supporting Chai's fundraising environment for a potential Series C within a two-to- three-year horizon. | 中 | SI012, SI013 |
| CI037 | IQVIA's 2026 global R&D trends report finds credible early signal that AI-enabled programs influence productivity metrics at the preclinical stage, but clinical-stage evidence for AI-specific economic advantages remains limited—a headwind for Chai's revenue multiple at the next financing round. | 中 | SI025, SI016 |
| CI038 | Chai's BusinessWire Chai-2 announcement (June 30, 2025) and subsequent Series B (December 2025) established a three-round fundraising sequence at rapidly escalating valuations ($150M → $550M → $1.3B), implying investor conviction has advanced faster than commercial milestone delivery. | 中 | SI006, SI019, SI010 |
| CE001 | Chai-1 is a multimodal foundation model for molecular structure prediction released as an open-source biorxiv preprint in October 2024, achieving state-of-the-art performance across drug discovery benchmarks. | 中 | SE009, SE001 |
| CE002 | Chai-1 supports simultaneous prediction of proteins, small molecules, DNA, RNA, glycosylations, and mixed-modality molecular complexes in a single architecture. | 中 | SE009, SE001 |
| CE003 | Chai-1 model weights and inference code are released under Apache 2.0 License, permitting both academic and commercial use including drug discovery applications. | 高 | SE014, SE013 |
| CE004 | Chai-1 is distributed as a Python package (chai_lab, version 0.6.1) on PyPI and requires a Linux environment with a CUDA-capable GPU for local inference. | 中 | SE002, SE001 |
| CE005 | Chai-1 can be run in single-sequence mode without MSAs while preserving most of its structure prediction performance, reducing computational dependency. | 中 | SE009, SE001 |
| CE006 | Chai-1 can be prompted with experimental restraints such as crosslinking mass spectrometry data, which boosts structure prediction performance by double-digit percentage points over baseline. | 中 | SE009 |
| CE007 | Chai-2 achieves a 16% hit rate in fully de novo antibody design—over 100-fold higher than the sub-0.1% hit rates reported for prior computational antibody methods. | 中 | SE010, SE012 |
| CE008 | Chai-2's antibody design validation was performed on 52 diverse antigen targets, none of which had a preexisting antibody or nanobody binder recorded in the RCSB Protein Data Bank. | 中 | SE010, SE003 |
| CE009 | Testing ≤20 designs per antigen target, Chai-2 produced at least one experimentally validated binder for 50% of the 52 targets in a single round of wet-lab assay. | 中 | SE010, SE012 |
| CE010 | Chai-2-designed antibodies exhibit nanomolar-range binding affinities, specificity for their intended targets, and strong developability profiles suitable for rapid therapeutic translation. | 中 | SE010, SE011 |
| CE011 | Chai-2 supports both nanobody (VHH single-domain) and full-length VH-VL antibody formats in its de novo generative design output. | 中 | SE010, SE012 |
| CE012 | Chai-2 designs all six complementarity-determining regions (CDRs) of an antibody entirely from scratch using only the target antigen identity and epitope location as input. | 中 | SE012, SE010 |
| CE013 | In the November 2025 challenging-targets preprint, over 86% of full-length monoclonal antibodies designed by Chai-2 showed developability profiles on par with approved therapeutic antibodies. | 中 | SE011, SE017 |
| CE014 | Chai-2 successfully designed functional antibodies mediating GPCR agonism and highly specific antibodies selectively binding tumor-specific neoepitopes—both considered challenging targets for conventional approaches. | 中 | SE011, SE013 |
| CE015 | Chai-2 achieved a 68% wet-lab success rate in miniprotein binder design, routinely yielding picomolar-affinity binders. | 中 | SE010, SE012 |
| CE016 | As of May 2026, all Chai-2 antibody design performance claims have been published exclusively as company-authored biorxiv preprints; none has undergone external peer review. | 中 | SE017, SE010, SE011 |
| CE017 | Traditional computational antibody discovery approaches—including immunization, directed evolution, and yeast-surface display—consistently report experimental hit rates below 0.1%. | 中 | SE012, SE010 |
| CE018 | AlphaFold 3, developed by Google DeepMind, was published in Nature in May 2024 as a peer-reviewed paper demonstrating accurate structure prediction of diverse biomolecular interactions. | 高 | SE004, SE005 |
| CE019 | ESMFold, developed by Meta FAIR, predicts atomic-level protein structure from a single protein sequence using a large protein language model, as published in Science in January 2023. | 中 | SE004, SE009 |
| CE020 | Chai-1's preprint benchmarks AlphaFold3 primarily using AF3's publicly released predictions rather than running AlphaFold3 natively, limiting direct head-to-head comparability. | 中 | SE009 |
| CE021 | Chai-2 model weights and training data are not publicly disclosed; access is available only through Chai Discovery's invitation-based early-access partner program. | 中 | SE013, SE012 |
| CE022 | Chai-1's GitHub repository (chaidiscovery/chai-lab) provides full source code, model weights, dev-container setup for reproducible environments, and citation metadata for both Chai-1 and Chai-2. | 中 | SE014, SE001 |
| CE023 | Chai-1 inference is recommended on NVIDIA A100 80 GB or H100 80 GB GPUs; A10s and A30s support smaller complexes; consumer-grade RTX 4090 has also been reported to work. | 中 | SE001, SE009 |
| CE024 | Chai-2 designs are benchmarked against the RCSB Protein Data Bank to confirm that none of the 52 test targets had a preexisting antibody binder in the PDB, establishing the novelty of the validation set. | 中 | SE010, SE003 |
| CE025 | Experimentally determined crystal structures of Chai-2-designed antibodies closely matched their in silico predictions, demonstrating atomic-level structural accuracy of the generative model. | 中 | SE011 |
| CE026 | lab.chaidiscovery.com provides a browser-based interface for Chai-1 structure predictions, free of charge and including for commercial drug discovery, requiring only email-based authentication. | 中 | SE015, SE013 |
| CE027 | Chai-1 accepts FASTA format inputs and generates five sample predictions by default; it can connect to an external MSA server for improved accuracy using the chai_lab.chai1.run_inference function. | 中 | SE001, SE009 |
| CE028 | Chai Discovery operates a Responsible Deployment Framework for Chai-2 early access, focused on health-positive and low-risk applications, biosafety, and alignment with societal goals. | 中 | SE019, SE013 |
| CE029 | The FDA published draft guidance titled 'Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products' in January 2025, and has reviewed over 500 AI-incorporating drug submissions since 2016. | 中 | SE022 |
| CE030 | Chai Discovery co-founder Jack Dent stated that the company's codebase is fully homegrown and uses highly custom model architectures, not fine-tuned open-source LLMs. | 中 | SE018 |
| CE031 | Chai-2 enables the full workflow from AI-generated antibody sequences to wet-lab experimental validation in under two weeks for most tested targets. | 中 | SE010, SE012 |
| CE032 | In one documented case study, Chai-2 solved an antibody challenge that had previously consumed over $5 million and multiple years of traditional R&D in a few hours, with lab validation within two weeks. | 中 | SE012, SE021 |
| CE033 | Chai-2's November 2025 challenging-targets preprint demonstrated de novo antibody design against GPCR targets—a class considered very difficult due to their transmembrane architecture—as well as tumor-specific neoepitopes. | 中 | SE011, SE013 |
| CE034 | Chai-1's benchmarks for AlphaFold3 were sourced from AF3's publicly released predictions; Chai Discovery explicitly states it did not run AlphaFold3 natively for this comparison. | 中 | SE009 |
| CE035 | CASP (Critical Assessment of Protein Structure Prediction) is the international blind benchmarking competition for protein structure prediction, representing the gold standard for independent model evaluation. | 中 | SE007, SE004 |
| CE036 | The LAP (Liability Antibody Profiler) tool, published in PLOS Computational Biology, provides sequence and structural mapping of antibody liabilities against natural and therapeutic antibody repertoires for developability assessment. | 中 | SE006 |
| CE037 | The 52 antigen targets used in Chai-2's July 2025 validation study were confirmed as lacking any preexisting antibody or nanobody binder in the Protein Data Bank, making the validation set genuinely novel. | 中 | SE010, SE003 |
| CE038 | Chai Discovery has publicly indicated plans to expand Chai-2 to additional molecular modalities including peptides, enzymes, and small molecules, and to develop bespoke models for pharma partners using proprietary data. | 中 | SE019, SE020, SE018 |
| CU001 | Eli Lilly and Company is Chai Discovery's sole publicly named commercial paying partner as of May 2026, announced January 8, 2026 via joint BusinessWire press release. | 高 | SU001, SU002 |
| CU002 | The Chai-Lilly collaboration deploys Chai's frontier AI models across multiple biologic targets at Lilly's internal drug discovery programs. | 高 | SU001, SU003 |
| CU003 | The Chai-Lilly collaboration includes training a custom Chai model on Lilly's proprietary compound and sequence data. | 高 | SU001, SU002 |
| CU004 | No financial terms, contract duration, or milestone structure for the Eli Lilly collaboration have been publicly disclosed. | 高 | SU001, SU002, SU003 |
| CU005 | Lilly's internal TuneLab program, led by Aliza Apple, is the organizational unit that sponsored the Chai collaboration. | 中 | SU002 |
| CU006 | The Lilly collaboration followed a disclosed 'period of evaluation,' indicating that large pharma customers conduct due diligence before formalizing Chai partnerships. | 中 | SU002, SU004 |
| CU007 | Chai-2 early access was opened in July 2025 to 'select partners,' with biotech and pharmaceutical companies invited to apply through Chai's Responsible Deployment program. | 高 | SU005, SU015 |
| CU008 | Menlo Ventures, Chai's Series A lead investor, stated publicly that 'a meaningful fraction of the biotech industry' had already applied for Chai-2 early access. | 中 | SU006, SU029 |
| CU009 | Chai's commercial model is a partnership structure, not self-service SaaS; access requires application review under the Responsible Deployment policy. | 中 | SU005, SU015, SU002 |
| CU010 | No commercial customers beyond Eli Lilly are publicly named by Chai Discovery, in press releases, CB Insights, Crunchbase, or independent news coverage as of May 2026. | 高 | SU001, SU020, SU021 |
| CU011 | CB Insights named Chai Discovery in its AI 100 2026 list under the Healthcare and Life Sciences category. | 中 | SU020 |
| CU012 | Chai grew from a $150 million valuation at seed in September 2024 to a $1.3 billion valuation at Series B in December 2025, a roughly 9× increase in approximately fifteen months. | 高 | SU007, SU020, SU027 |
| CU013 | Chai's Responsible Deployment policy gates partner access selectively; the policy page returned a 404 error during research, making specific screening criteria unavailable publicly. | 中 | SU015, SU002 |
| CU014 | Chai-2 claims a ~16–20% de novo antibody design hit rate versus less than 0.1–0.2% for traditional computational methods, based on company-authored biorxiv preprints. | 中 | SU023, SU030, SU004 |
| CU015 | Chai's customer base comprises three operationally distinct segments: large pharma payers (e.g., Lilly), biotech early-access pilots (unnamed), and non-commercial academic or research users of Chai-1. | 中 | SU001, SU005, SU018 |
| CU016 | Chai-1 is distributed as an open-source model under a non-commercial license on GitHub, PyPI, and HuggingFace, and is available for free inference at lab.chaidiscovery.com. | 高 | SU018, SU024, SU025 |
| CU017 | The chaidiscovery/chai-lab GitHub repository has 1,938 stars and 274 forks as of May 2026, indicating significant developer and research community interest. | 中 | SU018 |
| CU018 | The chai_lab Python package (underscore variant) on PyPI reached version 0.6.1 with a March 2025 release date, reflecting active package maintenance. | 中 | SU025 |
| CU019 | Chai Discovery's HuggingFace organization page (huggingface.co/chaidiscovery) hosts model weights and documentation, providing an additional distribution channel for research users. | 中 | SU024 |
| CU020 | Chai-1 web server at lab.chaidiscovery.com provides free structure prediction to individual researchers subject to account registration. | 中 | SU015, SU002 |
| CU021 | The chaidiscovery/chai-lab GitHub repository has 87 open issues as of May 2026, indicating active user community engagement through bug reports and feature requests. | 中 | SU018 |
| CU022 | The chaidiscovery/chai-lab GitHub repository was created September 2024 and received its most recent push in April 2026, demonstrating approximately twenty months of continuous open-source development. | 高 | SU018, SU019 |
| CU023 | The Chai-2 main technical report was posted as a biorxiv preprint in July 2025, and the Chai-1 preprint reached version 2 by October 2024, demonstrating scientific transparency. | 高 | SU022, SU023 |
| CU024 | Academic and independent research institutions are the primary Chai-1 adopters via GitHub and PyPI; these users are non-paying unless they enter a formal commercial agreement. | 中 | SU018, SU025, SU024 |
| CU025 | Chai-1's non-commercial license prohibits revenue-generating deployments without a paid agreement; the open-source user base does not directly monetize at its current scale. | 中 | SU022, SU015 |
| CU026 | Chai Discovery's customer revenue is highly concentrated in a single disclosed partner (Eli Lilly), creating material dependency risk with no disclosed diversification timeline. | 高 | SU001, SU010, SU020 |
| CU027 | No NPS score, customer satisfaction rating, or renewal intent signal has been publicly disclosed for any Chai commercial partnership. | 中 | SU002, SU009, SU021 |
| CU028 | No ARR, quarterly revenue, or total contract value figure has been publicly disclosed by Chai Discovery as of May 2026. | 高 | SU007, SU027, SU020 |
| CU029 | Chai's selective Responsible Deployment access policy, while preserving partnership quality, may slow customer acquisition by creating a gating bottleneck with no disclosed conversion timeline. | 中 | SU005, SU015 |
| CU030 | Peer AI drug discovery companies including Exscientia, BenevolentAI, and Recursion have experienced high-profile clinical failures that have eroded broad pharma confidence in AI discovery platforms. | 中 | SU014 |
| CU031 | No Chai Discovery-designed molecule has entered clinical trials (IND filing) as of May 2026. | 高 | SU014, SU002, SU023 |
| CU032 | The typical timeline from AI-assisted antibody discovery to IND filing is four to seven years, meaning that even the earliest Chai-Lilly outputs are unlikely to generate clinical proof before 2029. | 中 | SU014, SU030 |
| CU033 | As of 2025, zero drugs designed end-to-end by AI have received FDA approval; the $60 billion invested in AI drug discovery has yet to produce a single approved product. | 中 | SU014 |
| CU034 | The Chai-Lilly collaboration is likely multi-year in duration given the scope—multiple biologic targets and custom model training—though no contract term has been disclosed. | 中 | SU001, SU004 |
| CU035 | Chai-2's application-based early-access gating may slow new customer acquisition; no SLA or conversion timeline for moving applicants to paying partner status has been disclosed. | 中 | SU005, SU008 |
| CU036 | Menlo Ventures' 'meaningful fraction of biotech industry' applied statement suggests substantial inbound demand, but the conversion rate from applicant to paying partner is undisclosed and likely low. | 中 | SU006, SU008 |
| CU037 | The absence of named customers beyond Lilly is consistent with standard pharma industry practice of confidential partnership agreements; however, investor diligence should demand disclosure of logo counts without partner names. | 中 | SU002, SU009, SU016 |
| CU038 | Chai-2's hit-rate benchmarks are based on company-authored preprints evaluated on PDB-novel targets; no independent third-party replication across customer-specified targets has been published. | 中 | SU023, SU030, SU022 |
| CU039 | Series B board additions from Oak HC/FT (Annie Lamont) and General Catalyst (Hemant Taneja) bring healthcare network and commercial relationships that may accelerate future customer acquisition. | 低 | SU007, SU008, SU027 |
| CU040 | Industry coverage of Chai by FierceBiotech, BioPharma Trend, and Biotech Industry Examiner positions Chai as a new category player, increasing pharma awareness of the platform beyond the Lilly deal. | 中 | SU010, SU011, SU017 |
| CR001 | Chai's homepage frames Chai-2 as de novo antibody design with atomic precision against challenging targets. | 中 | SR001 |
| CR002 | Chai's official Chai-1 launch post says the model is available through a free web interface and that the code and weights were released under an Apache 2.0 license. | 高 | SR002, SR013, SR015 |
| CR003 | As of 2026-05-22, Chai-1 remains documented in fetched public materials as a bioRxiv preprint rather than a peer-reviewed journal article. | 中 | SR004, SR005 |
| CR004 | The Chai-1 preprint describes a multimodal foundation model spanning proteins, small molecules, DNA, RNA, covalent modifications, and related molecular inputs. | 中 | SR004, SR005 |
| CR005 | Chai's launch post reports 77% PoseBusters success and 69.8% acceptable multimer predictions in a single-sequence setting, but those benchmarks are company-selected evaluation claims rather than clinical proof. | 中 | SR002, SR004 |
| CR006 | AlphaFold 3 publicly claims accurate prediction of complexes involving proteins, nucleic acids, small molecules, ions, and modified residues, making it a powerful benchmark competitor for any structure-first platform. | 中 | SR003 |
| CR007 | Absci publicly positions itself as an AI biologics company with internal and partnered programs, showing that Chai competes in a crowded AI-biologics category rather than a greenfield niche. | 中 | SR007 |
| CR008 | Generate Biomedicines publicly markets generative biology for therapeutic creation, reinforcing that pharma buyers can compare Chai against other vertically integrated AI-therapeutics vendors. | 中 | SR008 |
| CR009 | EvolutionaryScale markets frontier protein sequence models for life sciences, adding another well-funded foundation-model competitor for talent, partnerships, and benchmark mindshare. | 中 | SR009 |
| CR010 | Chai's public software artifacts remained actively distributed through at least release v0.6.1 in February–March 2025 across GitHub Releases and PyPI. | 中 | SR014, SR030 |
| CR011 | As of runDate, lab.chaidiscovery.com redirects users to an email magic-link login flow rather than an anonymously usable dashboard, so hosted access is gated. | 中 | SR028 |
| CR012 | A 2026 PMC study on antibody developability concluded that protein language model performance and reliability in real-world industrial antibody discovery pipelines remain underexplored. | 中 | SR018 |
| CR013 | That same PMC study relied on retrospective measurements from 33 historical therapeutic programs, underscoring that much of the best available validation evidence is still proprietary and backward-looking. | 中 | SR018 |
| CR014 | The mAbs developability review says antibody lead selection still requires production, analytical and biophysical characterization, stability work, and process and formulation assessment beyond in silico design. | 中 | SR027 |
| CR015 | The same review identifies aggregation, self-interaction, hydrophobicity, deamidation, oxidation, clipping, and poor purity as recurring liabilities in antibody developability. | 中 | SR027 |
| CR016 | The mAbs review explicitly notes that aggregation can reduce efficacy and create immunogenicity risk after administration. | 中 | SR027 |
| CR017 | A 2024 therapeutic-AI review says one of the field's central challenges is bridging computational molecular design to experimental validation and downstream therapeutic translation. | 中 | SR019 |
| CR018 | Fetched official and developer materials do not publish raw assay-level wet-lab datasets or a third-party replication package sufficient to independently verify Chai-2 performance claims. | 中 | SR001, SR002, SR013, SR014 |
| CR019 | FDA's AI/ML medical-device framework is written for software as a medical device, not for therapeutic antibody approval. | 中 | SR006 |
| CR020 | FDA's drug-development AI page discusses use of AI in development and review but does not publish a dedicated approval pathway for AI-designed antibodies or biologics. | 中 | SR010 |
| CR021 | FDA CMC materials say IND-stage programs must provide manufacturing-process, quality, purity, strength, stability, and batch information, which means discovery-stage model quality alone is insufficient for clinical progression. | 高 | SR026, SR027 |
| CR022 | FDA CMC materials identify insufficient batch data, stability problems, impurity concerns, and sterility or endotoxin control failures as potential safety concerns during development. | 中 | SR026 |
| CR023 | Chai's fetched public materials do not disclose a manufacturing partner, process scale, release specification set, or stability package for any antibody asset. | 中 | SR001, SR002, SR016, SR025 |
| CR024 | The Lilly collaboration announcement is discovery-stage evidence—multiple biologic targets plus custom model training on Lilly data—not evidence of an IND, clinical asset, or approved therapy. | 高 | SR016, SR025 |
| CR025 | No fetched official or major-news source shows a public IND-stage, clinical-stage, or FDA-reviewed Chai therapeutic program as of 2026-05-22. | 中 | SR001, SR016, SR025, SR010 |
| CR026 | Among fetched official and major-news materials, Eli Lilly is the only publicly named pharma collaboration attached to Chai at runDate. | 高 | SR016, SR025 |
| CR027 | Fetched public materials do not disclose ARR, revenue, or customer-count metrics, leaving monetization and concentration risk unquantified. | 中 | SR001, SR024, SR025, SR012 |
| CR028 | Chai's August 2025 Series A press release states that the company raised $70M and reached $100M total funding, which provides capital but also creates expectations for rapid proof of commercial and scientific progress. | 中 | SR024 |
| CR029 | TechCrunch reported a $130M Series B at a $1.3B valuation in December 2025, and Forbes echoed the same valuation context. | 高 | SR023, SR025 |
| CR030 | Because Chai is already priced as a high-growth frontier-biology company, any delay in reproducible wet-lab proof, partner diversification, or pre-IND progress would increase down-round risk. | 中 | SR023, SR024, SR025 |
| CR031 | AlphaFold 3, Absci, Generate Biomedicines, and EvolutionaryScale each market overlapping AI-biologics or biomolecular-model capabilities, increasing competition for pharma attention and pricing leverage. | 中 | SR003, SR007, SR008, SR009 |
| CR032 | AlphaFold 3's published weight terms restrict use to non-commercial research, showing that model-access terms themselves can become a strategic moat and a procurement constraint in this category. | 中 | SR017 |
| CR033 | Nature's criticism of the AlphaFold 3 publication without open code shows that openness, reproducibility, and licensing choices are already contested governance issues in frontier biology AI. | 中 | SR011, SR017 |
| CR034 | Chai's Series A press release claims Chai-2 delivered a near-20% de novo antibody hit rate versus prior computational approaches around 0.1%, but that comparison currently rests on company-issued evidence. | 中 | SR024 |
| CR035 | Chai's Series A press release says former Pfizer Chief Scientific Officer Mikael Dolsten joined the board in 2025, adding scientific oversight without eliminating founder dependence. | 中 | SR024 |
| CR036 | Forbes summarizes Chai as an AI-antibody company that had already landed Lilly and reached a $1.3B valuation by December 2025. | 中 | SR025 |
| CR037 | Chai distributes artifacts through GitHub, Hugging Face, and PyPI, which broadens reach and developer adoption but also expands IP-leakage and misuse surface area. | 中 | SR013, SR029, SR030 |
| CR038 | The Center for Health Security argues that governments should evaluate advanced biological AI models and impose safeguards or limits when biosecurity risk is high. | 中 | SR022 |
| CR039 | A 2025 Frontiers review says AI protein design creates significant biosecurity concerns that must be balanced against innovation benefits. | 中 | SR020 |
| CR040 | A 2025 PLOS Computational Biology paper characterizes biological AI models as having dual-use capabilities of concern. | 中 | SR021 |
| CR041 | Public financing and profile materials center Chai around Joshua Meier, Jack Dent, Matthew McPartlon, and Jacques Boitreaud, indicating key-person concentration around a small founding and research core. | 高 | SR024, SR025 |
| CR042 | Fetched official and financing materials do not disclose burn, runway, gross margin, or unit economics, so capital efficiency cannot be independently verified. | 中 | SR001, SR024, SR025 |
| CR043 | Chai's official launch post and repository together show an Apache 2.0 software release for Chai-1 even though long-term economic value likely depends on proprietary data, wet-lab loops, and future closed models or services. | 高 | SR002, SR013, SR015 |
| CR044 | Chai's Hugging Face organization page shows one public model and only limited visibly named team representation, implying that public governance surface is still thin relative to the sensitivity of protein-model deployment. | 中 | SR029 |
| CR045 | The chai_lab package was publicly uploadable on PyPI in March 2025, confirming that Chai distributes developer tooling through mainstream software channels rather than a closed research enclave. | 中 | SR030 |
| CR046 | Given the current disclosure gaps, the most useful external monitorables before underwriting more upside are independent wet-lab replication, a second named pharma partner, visible CMC progress, and any quantified burn or runway disclosure. | 中 | SR001, SR024, SR025, SR026 |
| CV001 | Chai Discovery closed a $130 million Series B in December 2025 at a $1.3 billion valuation. | 中 | SV001, SV002 |
| CV002 | The Series B brought Chai’s total funding to more than $225 million. | 中 | SV001, SV002 |
| CV003 | Chai’s January 2026 Lilly collaboration said the company had raised nearly $230 million to date, indicating only a minor rounding difference versus the more-than-$225 million formulation from the Series B release. | 中 | SV001, SV004 |
| CV004 | Chai was founded in 2024, so the company reached a $1.3 billion valuation within roughly its first year of operation. | 中 | SV002, SV005 |
| CV005 | Josh Meier previously worked at OpenAI and Meta, and General Catalyst says he co-led development of ESM1 before starting Chai. | 中 | SV005, SV006 |
| CV006 | Chai positions itself as a computer-aided design suite for molecules rather than as a company with a disclosed internal clinical pipeline. | 中 | SV001, SV004, SV005 |
| CV007 | The Lilly collaboration deploys Chai’s frontier AI across multiple biologic targets and includes a custom model trained on Lilly’s proprietary data. | 中 | SV004, SV005 |
| CV008 | Chai-2 is described in company materials as a zero-shot antibody design platform with double-digit hit rates and drug-like property design. | 中 | SV004, SV008 |
| CV009 | Chai’s June 2025 release says Chai-2 achieved an antibody-design hit rate close to 20% while the direct technical report describes over 100-fold improvement versus previous computational methods. | 中 | SV007, SV008 |
| CV010 | Public Chai materials reviewed for this chapter do not disclose revenue, ARR, gross margin, or unit economics. | 中 | SV001, SV004, SV005, SV030 |
| CV011 | Public Chai materials reviewed for this chapter do not identify any Chai-designed asset in Phase I, Phase II, Phase III, or approved-commercial status. | 中 | SV005, SV030 |
| CV012 | The Lilly deal is the only named large-pharma commercial validation point surfaced in retained Chai sources for this chapter. | 中 | SV004, SV005, SV030 |
| CV013 | Recursion’s market cap was about $1.65 billion as of May 2026. | 中 | SV011, SV012 |
| CV014 | Recursion’s enterprise value was about $1.07 billion as of May 2026. | 中 | SV011 |
| CV015 | Recursion had trailing-twelve-month revenue of about $66.41 million. | 中 | SV011 |
| CV016 | Recursion held about $654.47 million of cash and cash equivalents with $72.38 million of debt, implying about $582.10 million of net cash. | 中 | SV011 |
| CV017 | Recursion reported first-quarter 2026 revenue of $6.5 million, primarily from collaboration agreements rather than product sales. | 中 | SV009, SV010 |
| CV018 | Recursion’s net loss for the first quarter of 2026 was $117.5 million. | 中 | SV009, SV010 |
| CV019 | Recursion had 530.76 million shares outstanding and its share count had increased by 51.66% year over year. | 中 | SV011 |
| CV020 | Stock Analysis showed a Hold consensus on Recursion with an average price target of $6.64, about 113.5% above the then-current price. | 中 | SV011 |
| CV021 | Schrödinger’s market cap was about $0.99 billion as of May 2026. | 中 | SV013, SV014 |
| CV022 | Schrödinger’s enterprise value was about $696.6 million as of May 2026. | 中 | SV013 |
| CV023 | Schrödinger had trailing-twelve-month revenue of about $254.91 million and about $398.96 million of cash as of May 2026 market-data snapshots. | 中 | SV013 |
| CV024 | Schrödinger reported first-quarter 2026 total revenue of $58.6 million, including $22.9 million of drug-discovery revenue and $35.6 million of software revenue. | 中 | SV015 |
| CV025 | Schrödinger reported $406 million of cash, cash equivalents, restricted cash, and marketable securities at the end of the first quarter of 2026. | 中 | SV015 |
| CV026 | Schrödinger’s first-quarter 2026 ACV was $28.4 million and trailing four-quarter ACV was $201 million. | 中 | SV015 |
| CV027 | Stock Analysis showed a Buy consensus on Schrödinger with an average price target of $20.88, about 57.8% above the then-current price. | 中 | SV013 |
| CV028 | Absci’s market cap was about $793.6 million as of May 2026. | 中 | SV018, SV019 |
| CV029 | Absci’s enterprise value was about $672.27 million as of May 2026. | 中 | SV018 |
| CV030 | Absci had trailing-twelve-month revenue of about $1.84 million and cash of about $125.71 million as of May 2026 market-data snapshots. | 中 | SV018 |
| CV031 | Absci reported first-quarter 2026 revenue of $0.2 million and said existing cash should fund operations into the first half of 2028. | 中 | SV016, SV017 |
| CV032 | Absci’s first-quarter 2026 net loss was $29.6 million and net cash used in operating activities was $26.3 million. | 中 | SV017 |
| CV033 | Stock Analysis showed a Strong Buy consensus on Absci with an average price target of $8.76, about 72.1% above the then-current price. | 中 | SV018 |
| CV034 | Generate:Biomedicines’ market cap was about $1.79 billion as of May 2026. | 中 | SV021 |
| CV035 | Generate’s shares closed around $13.98 on May 21, 2026, and open-source market-data pages showed an average price target of about $25.40 with Strong Buy consensus. | 中 | SV028, SV029 |
| CV036 | Generate had trailing-twelve-month revenue of about $30.30 million. | 中 | SV029 |
| CV037 | Generate reported $516.6 million of cash and marketable securities at March 31, 2026, quarter revenue of $7.2 million, and runway into the first half of 2028 while still expecting to require additional capital long term. | 中 | SV020 |
| CV038 | Generate’s lead GB-0895 program was already in Phase 3 severe asthma in first-quarter 2026 disclosures. | 中 | SV020 |
| CV039 | Isomorphic Labs raised $600 million in its first external funding round in 2025 to advance programs into clinical development. | 中 | SV023 |
| CV040 | Isomorphic Labs announced a $2.1 billion Series B in 2026. | 中 | SV022 |
| CV041 | Isomorphic Labs said its 2024 Lilly and Novartis collaborations together could be worth nearly $3 billion excluding royalties. | 中 | SV024 |
| CV042 | Xaira emerged in 2024 with $1 billion in committed funding. | 中 | SV025 |
| CV043 | No AI-designed drug had received regulatory approval in the retained 2026 sector review. | 中 | SV026 |
| CV044 | The same 2026 sector review said more than 173 AI drug programs were in clinical trials while Phase II success rates still matched traditional drugs despite early-stage gains. | 中 | SV026 |
| CV045 | DrugPatentWatch framed AI drug discovery as attacking a real structural problem: average NME cost around $2.8 billion, 12 to 15 years from hypothesis to approval, and roughly 90% Phase I attrition. | 中 | SV027 |
| CV046 | Because Chai has no disclosed revenue, ARR, or margin profile, public evidence does not support a conventional DCF or EV-revenue model with false precision. | 中 | SV001, SV004, SV030 |
| CV047 | A stage-appropriate approach for Chai is to triangulate the last-round mark against public-comp market values, disclosed cash positions, partnership proof, and clinical-stage distance. | 中 | SV011, SV013, SV018, SV020 |
| CV048 | Chai’s $1.3 billion last-round valuation sits below Recursion’s and Generate’s current public market caps but above Schrödinger’s and Absci’s, despite Chai lacking the revenue, cash disclosure, or clinical-stage proof those public comps provide. | 中 | SV001, SV012, SV014, SV019, SV021 |
| CV049 | Relative to the best-capitalized private peers, Chai’s capital base is meaningful but smaller than Isomorphic’s $600 million then $2.1 billion rounds and Xaira’s $1 billion launch financing. | 中 | SV002, SV022, SV023, SV025 |
| CV050 | The strongest bull-case support for Chai is the combination of elite AI founding talent, aggressive funding support, company-reported Chai-2 benchmark outperformance, and Lilly’s willingness to evaluate Chai designs and sponsor a custom-model collaboration. | 中 | SV004, SV006, SV007, SV008 |
| CV051 | The strongest bear-case challenge is that Chai reached a $1.3 billion valuation before publicly disclosing revenue, a second named customer, or any clinical-stage asset. | 中 | SV001, SV004, SV005, SV030 |
| CV052 | A reasonable bear case for Chai is roughly $0.4 billion to $0.8 billion if Lilly remains an isolated proof point and investors re-rate the company toward lower-proof platform comps. | 中 | SV018, SV019, SV026, SV027 |
| CV053 | A reasonable base case for Chai is roughly $0.9 billion to $1.4 billion if Lilly converts into repeatable demand, Chai sustains technical credibility, and the company still lacks near-term clinical proof or disclosed software economics. | 中 | SV001, SV012, SV014, SV021 |
| CV054 | A reasonable bull case for Chai is roughly $1.8 billion to $2.6 billion if benchmark claims hold up, a second marquee pharma partner lands, and the company shows a path from design wins toward clinical programs or recurring commercial use. | 中 | SV004, SV008, SV022, SV024 |
| CV055 | Compared with Generate, which already has Phase 3 data-bearing assets and disclosed cash plus revenue, Chai’s current $1.3 billion mark offers less evidence support per dollar of valuation. | 中 | SV001, SV020, SV021, SV029 |
| CV056 | Recommendation should remain track rather than buy because technology and partner validation are real, but the valuation leaves limited margin of safety relative to disclosed public comps. | 中 | SV001, SV012, SV014, SV019, SV021 |
| CV057 | Confidence should be medium because the company is private and public evidence leaves major gaps around revenue, Lilly economics, cap table, and next-round terms. | 中 | SV004, SV017, SV030 |
| CV058 | Risk rating should be high because Chai is preclinical, commercially concentrated, and exposed to a sector that still has zero approved AI-designed drugs. | 中 | SV004, SV026, SV030 |
| CV059 | Valuation stance is stretched because the company reached unicorn status within about a year of founding without publicly disclosed revenue or clinical-stage assets. | 中 | SV001, SV002, SV005 |
| CV060 | The most plausible thesis-break triggers are failure to add a second marquee pharma partner, inability to show independent wet-lab or downstream developability proof, and any sign that Lilly remains a one-off experiment rather than a repeatable customer motion. | 中 | SV004, SV007, SV027, SV030 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | BusinessWire / Chai Discovery | Chai Discovery Announces $70 million Series A To Transform Molecular Design | "Progress towards game-changing drugs and treatments is far too slow, stymied by costly trial-and-error experiments. Chai Discovery exists to push the boundaries of what's possible in this field, applying frontier AI to transform biology from science to engineering." |
| SO002 | BusinessWire / Chai Discovery | Chai Discovery Announces $130 Million Series B To Transform Molecular Discovery | "The Series B round brings Chai's total funding to more than $225 million. As part of the fundraise, Annie Lamont from Oak HC/FT and Hemant Taneja from General Catalyst will be joining the board." |
| SO003 | BusinessWire / Chai Discovery | Chai Discovery Unveils Chai-2 Breakthrough, Achieving Fully De Novo Antibody Design With AI | "Chai-2 demonstrates a remarkable antibody design hit rate close to 20%. Prompted with just the target and epitope, Chai-2 successfully designs all complementarity-determining regions (CDRs) entirely from scratch." |
| SO004 | BusinessWire / Chai Discovery | Chai Discovery Announces Collaboration with Eli Lilly and Company to Accelerate Biologics Discovery | "Chai has raised nearly $230M to date. Chai-2 is the first zero-shot antibody design platform to achieve double-digit experimental hit rates and design molecules with drug-like properties." |
| SO005 | TechCrunch | OpenAI-backed biotech firm Chai Discovery raises $130M Series B at $1.3B valuation | |
| SO006 | TechCrunch | From OpenAI's offices to a deal with Eli Lilly — how Chai Discovery became one of the flashiest names in AI drug development | "Meier and Dent had originally met in computer science classes at Harvard but, at the time, Dent was a Stripe engineer. Altman asked him if he thought Meier would be open to collaborating on a proteomics startup." |
| SO007 | General Catalyst | Our Investment in Chai Discovery | "Josh Meier helped pioneer frontier AI in the early days of OpenAI and at Meta where he co-led development of ESM1, the first transformer protein-language model." |
| SO008 | Observer | This Startup Backed By OpenAI and the Jobs Family Is the Latest A.I. Drug Discovery Unicorn | |
| SO009 | FierceBiotech | Chai infuses AI drug discovery efforts with $130M series B | |
| SO010 | Chai Discovery | Chai Discovery — Official Homepage | |
| SO011 | LoonBio | AI Drug Discovery's $60 Billion Reality Check: Hype, Failures, and the Market Access Blindspot | "Despite over $60 billion in venture capital pouring into AI-driven drug discovery since 2015 and myriad computational breakthroughs, as of 2025 not a single AI-designed drug has achieved FDA approval." |
| SO012 | Biotech Industry Examiner | Chai Discovery's $130m bet: can 'CAD for molecules' make biologics faster and cheaper? | |
| SO013 | Bloomberg | OpenAI-Backed Chai Discovery Raises $130 Million for AI-Designed Molecules | |
| SO014 | Built In San Francisco | AI Molecular Design Startup Chai Discovery Secures $70M Series A | |
| SO015 | HIT Consultant | Eli Lilly Taps Chai Discovery's Frontier AI to Design 'Computer-Aided' Biologics | |
| SO016 | TechFundingNews | Chai Discovery snaps up $70M to slash drug development timelines with AI | |
| SO017 | Longevity Technology | AI drug discovery startup valued at $1.3b in huge funding round | |
| SO018 | The Pharmaletter | Chai Discovery | |
| SO019 | IntuitionLabs | AI Biologics Design: Chai Discovery & Eli Lilly Partnership | |
| SO020 | BioPharma Trend | Chai Discovery Raises $70M to Expand Zero-Shot Biologics Design AI Platform | |
| SO021 | Analytics India Magazine | How is Chai Discovery revolutionizing biotech with AI funding led by Anthology | |
| SO022 | bioRxiv (Chai Discovery) | Zero-shot antibody design in a 24-well plate (Chai-2 preprint) | "We introduce Chai-2, a multimodal generative model that achieves a 16% hit rate in fully de novo antibody design, representing an over 100-fold improvement compared to previous computational methods." |
| SO023 | bioRxiv (Chai Discovery) | Drug-like antibody design against challenging targets with atomic precision (Chai-2 challenging targets preprint) | "We find that >86% of these full-length mAbs have strong developability profiles on par with therapeutic antibodies. We further show that experimentally determined structures of Chai-2 designs closely match their in silico predictions." |
| SO024 | MedVolt AI | Chai-2 Redefines Antibody Discovery: AI-Driven De Novo Design with Industry-Leading Hit Rates | |
| SO025 | The AI Insider | Chai Discovery Announces $70M Series A To Transform Molecular Design | |
| SM001 | Grand View Research | Artificial Intelligence In Drug Discovery Market Report, 2033 | "The global artificial intelligence in drug discovery market size was estimated at USD 2.35 billion in 2025 and is projected to reach USD 13.77 billion by 2033, growing at a CAGR of 24.8% from 2026 to 2033." |
| SM002 | Mordor Intelligence | Antibody Discovery Market Size, Share & 2030 Growth Trends Report | "The antibody discovery market size stands at USD 9.09 billion in 2025 and is projected to reach USD 15.45 billion by 2030, translating into an 11.3% CAGR across the forecast period." |
| SM003 | Mordor Intelligence | AI In Pharmaceutical R&D Market Size, Share & 2031 Growth Trends Report | "The AI In Pharmaceutical R&D Market size was valued at USD 3.30 billion in 2025 and is estimated to grow from USD 4.36 billion in 2026 to reach USD 17.66 billion by 2031, at a CAGR of 32.25%." |
| SM004 | Bio-in-Tech | AI Drug Discovery Market Size & Growth 2026: Forecasts, CAGR & Reality Check | "GrandView Research places the market at USD 2.35 billion in 2025 [...] GlobalMarketInsights reports as high as USD 24.5 billion in 2026 growing toward USD 160.49 billion by 2035 at 23.22% CAGR. The variation reflects genuine differences in scope." |
| SM005 | AllAboutAI | AI in Drug Development Statistics 2026: The $60 Billion Reality vs. Hype Analysis | "AI Pharma Adoption Rate: 69% of pharmaceutical companies are now investing in AI, surpassing cloud computing and other digital initiatives." |
| SM006 | CAS | AI in drug discovery: Moving from potential to practical | "The challenge is not data volume but data readiness. The data must be curated, contextualized, and aligned with specific discovery questions to produce more relevant, actionable results." |
| SM007 | IQVIA Institute for Human Data Science | IQVIA Institute's Global R&D Trends 2026 Report Finds Credible Signal on AI-Enabled Programs | "For the most recent three-year window, the Phase I success rate for AI-enabled emerging biopharma programs was 75%. That is a substantial advantage over comparable non-AI-enabled programs." |
| SM008 | WorldMetrics | AI Drug Discovery Statistics — 2026 Sourced Report | "AI reduces drug discovery time from 5-6 years to 12-18 months on average. AI can cut drug development costs by up to 30-50% through better target identification." |
| SM009 | Axis Intelligence | AI Drug Discovery 2026: 173 Programs, FDA Framework & Market | "2025: $1.94B (28% YoY growth). 2026: $2.6-2.8B projected [...] Conservative: $2.6B → $8.2B (CAGR 25.8%)." |
| SM010 | DrugPatentWatch | AI in Drug Discovery 2026: What Actually Works, What Remains Hype, and Where the IP Value Sits | "Bringing a New Molecular Entity to market costs, on average, US $2.8 billion when accounting for the cost of capital and the failures that subsidize every approval." |
| SM011 | 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. The market is expected to rise from USD 5.00 billion in 2026 to USD 12.56 billion by 2034, expanding at a CAGR of 12.20% from 2026 to 2034." |
| SM012 | Healthcare Research Reports | Global Antibody Discovery Market Set for Strong Expansion, Reaching $15.79 Billion With 10.1% CAGR by 2030 | "The antibody discovery market has experienced robust expansion over recent years. It is projected to expand from $9.78 billion in 2025 to $10.75 billion in 2026, demonstrating a CAGR of 9.8%." |
| SM013 | Drug Discovery Trends | 2024: The year AI drug discovery and protein structure prediction took center stage—2025 set to amplify growth | "The global AI drug discovery market, valued around $1 to $1.7 billion in 2023, will be worth a multiple of that by the decade's end. Analysts project the sector could be worth $9 billion or more." |
| SM014 | Statista | Topic: Pharmaceutical research and development (R&D) | "Global pharmaceutical R&D spending has nearly doubled since 2016, reaching around 300 billion U.S. dollars in 2025, and projections point to continued but modest expansion through 2030." |
| SM015 | LifeSciVoice | AI Drug Discovery Investment in 2026: How Much Is Being Spent | "The question in 2026 is no longer whether AI will play a role in drug discovery. The strategic issue is how much capital should be allocated, where it is flowing, and which AI platforms are demonstrating measurable R&D productivity gains." |
| SM016 | Research and Markets | Antibody Discovery Market Report 2026 | |
| SM017 | Research and Markets | AI Protein Structure Prediction Global Market Report 2026 | |
| SM018 | Global Market Insights | Artificial Intelligence in Drug Discovery Market Size, Share — 2035 | |
| SM019 | New Market Pitch | AI In Drug Discovery Market Update (Q1 2026) | "The AI in drug discovery market is estimated at $2.9 billion in 2026, which is roughly 1.5% of the $194 billion global pharma R&D spend, a small but fast-growing slice." |
| SM020 | Code Brew | AI in Pharma and Biotech: Key Market Trends for 2026 | |
| SM021 | Towards Healthcare / Global Market Insights | AI in Drug Discovery Market Rises USD 160.49 Billion by 2035 | "The global AI in drug discovery market size was evaluated at USD 19.89 billion in 2025 and is expected to attain around USD 160.49 billion by 2035, growing at a CAGR of 23.22% from 2026 to 2035." |
| SM022 | Ardigen | AI in Biotech: 2026 Drug Discovery Trends | |
| SM023 | Grand View Research | Artificial Intelligence In Drug Discovery Market To Reach $13.7Bn By 2033 | "The pharmaceutical & biotechnology companies segment led the market with the largest revenue share of 59.19% in 2025." |
| SM024 | Business Research Insights | AI for Pharma and Biotech Market Size | Industry Trends [2026-2035] | "The global ai for pharma and biotech market size stood at USD 2.68 Billion in 2026 growing further to USD 8.67 Billion by 2035 at an estimated CAGR of 13.95% from 2026 to 2035." |
| SM025 | The Business Research Company | Antibody Discovery Global Market Report 2026 | "The antibody discovery market size has grown strongly in recent years. It will grow from $9.78 billion in 2025 to $10.75 billion in 2026 at a compound annual growth rate (CAGR) of 9.8%." |
| SP001 | Isomorphic Labs | Isomorphic Labs — Company Homepage | "Isomorphic Labs is an AI-first drug design company, using AI to unlock the secrets of biology and supercharge drug discovery." |
| SP002 | Schrödinger | Schrödinger — Company Overview | "Schrödinger has been developing innovative computational methods for more than 30 years to understand the physical world." |
| SP003 | Schrödinger | Schrödinger — Platform Overview | "The Schrödinger platform combines physics-based and machine learning approaches for drug discovery." |
| SP004 | Schrödinger | Schrödinger — Life Science Solutions | |
| SP005 | Generate:Biomedicines | Generate:Biomedicines — Company Homepage | "Generate:Biomedicines is a generative biology company designing proteins that couldn't be discovered through traditional means." |
| SP006 | Generate:Biomedicines | Generate:Biomedicines — Pipeline | "GB-0895 is a first-in-class, AI-designed anti-TSLP antibody currently in Phase 3 clinical development for severe asthma." |
| SP007 | Generate:Biomedicines | Generate:Biomedicines — News | |
| SP008 | Insilico Medicine | Insilico Medicine — About | "Insilico Medicine is an end-to-end, clinical-stage AI-driven drug discovery company that uses AI to discover, design, and develop drugs." |
| SP009 | Insilico Medicine | Insilico Medicine — Pipeline | "Insilico has generated more than 40 drug candidates with 13 IND approvals spanning oncology, fibrosis, immunology, and infectious disease." |
| SP010 | Insilico Medicine | Insilico Medicine — Blog | |
| SP011 | Recursion Pharmaceuticals | Recursion — Company Homepage | "Recursion is a clinical-stage TechBio company combining technology and wet-lab capabilities to industrialize drug discovery." |
| SP012 | Recursion Pharmaceuticals | Recursion — Technology Platform | "Recursion generates more than 50 petabytes of proprietary biological and chemical data using its automated labs and BioHive-2 supercomputer." |
| SP013 | Recursion Pharmaceuticals | Recursion — Pipeline | |
| SP014 | AbSci | AbSci — Company Homepage | "AbSci is unlocking the potential of artificial intelligence to design breakthrough medicines that were previously out of reach." |
| SP015 | AbSci | AbSci — Technology Platform | "AbSci's ACE (Antibody Creation and Evaluation) Assay enables ultra-high-throughput experimental characterization of AI-designed antibodies." |
| SP016 | AbSci | AbSci — News | |
| SP017 | EMBL-EBI / Google DeepMind | AlphaFold Database — Protein Structure Predictions | "The AlphaFold Protein Structure Database provides open access to protein structure predictions for the human proteome and 47 other organisms, covering over 200 million protein structures." |
| SP018 | Massachusetts Institute of Technology / jwohlwend (GitHub) | Boltz-2: Open-Source Biomolecular Structure and Affinity Prediction | "Boltz-2 is the state-of-the-art open-source model for biomolecular structure and affinity prediction. We benchmark against Chai-1 and achieve competitive results while also predicting binding affinities." |
| SP019 | Google DeepMind | AlphaFold3 — GitHub Repository | "The weights of AlphaFold 3 are made available for non-commercial use only, under the Creative Commons Attribution Non-Commercial 4.0 International licence." |
| SP020 | Meta AI (Facebook Research) | ESM — Evolutionary Scale Modeling (ESMFold) | "ESMFold enables rapid, accurate protein structure prediction from a single sequence using a large protein language model." |
| SP021 | Aqsa Laboratory (Columbia University) | OpenFold — Apache 2.0 AlphaFold2 Reimplementation | "OpenFold is a faithful, but trainable, PyTorch reproduction of DeepMind's AlphaFold 2, available under the Apache 2.0 License." |
| SP022 | Exscientia (now Recursion) | Exscientia — Company Homepage | |
| SP023 | Axis Intelligence | AI Drug Discovery Companies 2026 — Complete Analysis | "The AI drug discovery landscape in 2026 features five major categories: structure predictors, full-stack platforms, generative biology, physics-ML hybrids, and open-source alternatives—each targeting distinct aspects of the R&D workflow." |
| SP024 | Ardigen | AI in Biotech: Lessons from 2025 and Trends Shaping Drug Discovery in 2026 | "No AI-designed drug has crossed the FDA finish line yet, making clinical validation the key differentiating milestone the entire sector is racing toward in 2026." |
| SP025 | LoonBio | AI Drug Discovery's $60 Billion Reality Check: Hype, Failures, and the Market Access Blindspot | "The competitive moat in AI drug design is narrowing faster than most vendors admit: open-source models now benchmark within single-digit percentage points of commercial platforms on structure prediction tasks." |
| SP026 | Schrödinger | Schrödinger — Glide Docking Platform | |
| SP027 | Recursion Pharmaceuticals | Recursion — Team and Leadership | |
| SI001 | bioRxiv | Chai-1: Decoding the molecular interactions of life | "Chai-1 is available to all users—including for commercial purposes—through the Chai Discovery web interface; model weights are made available under Apache 2.0 for non-commercial use." |
| SI002 | Drug Discovery News | Chai Discovery Launches with $30M Seed Round | "Chai Discovery has launched with a $30 million seed round to build AI foundation models for drug discovery, with Joshua Meier and team aiming to make molecular design as accessible as computer-aided design tools." |
| SI003 | Drug Discovery News | Chai Discovery Series B: AI Drug Discovery Platform Raises $130M | "Chai Discovery has raised $130 million in Series B funding at a $1.3 billion valuation, co-led by Oak HC/FT and General Catalyst, to accelerate commercialization of its AI-powered molecular discovery platform." |
| SI004 | Drug Discovery News | AI Transforming Drug Discovery Economics | "AI platforms generating drug discovery hits must negotiate milestone and royalty terms early to capture long-term economic value, as pharma partners become more sophisticated about what computational contributions are worth in clinical-stage economics." |
| SI005 | Drug Discovery News | Bridging the Gap: From AI Hit to Clinical Candidate | "The transition from an AI-generated molecular hit to a validated clinical candidate involves substantially greater capital, time, and failure risk than the discovery phase itself— a gap that AI platforms must navigate either through internal capabilities or partner agreements." |
| SI006 | BusinessWire | Chai Discovery Announces $130 Million Series B To Transform Molecular Discovery | "Chai Discovery today announced the close of a $130 million Series B round to accelerate research and product development, and expand commercialization efforts." |
| SI007 | BusinessWire | Chai Discovery Announces Collaboration with Eli Lilly and Company to Accelerate Biologics Discovery | "Chai Discovery will develop an AI model trained on Lilly's large-scale proprietary datasets and tailored to Lilly's discovery workflows, alongside access to Chai's core platform models." |
| SI008 | TechCrunch | From OpenAI's Offices to a Deal with Eli Lilly: How Chai Discovery Became One of the Flashiest Names in AI Drug Development | "Every line of code in our codebase is homegrown—we use no off-the-shelf LLMs, which gives us full control over how we solve novel drug discovery challenges at scale." |
| SI009 | HIT Consultant | Eli Lilly, Chai Discovery Partner on Frontier AI for Drug Discovery | "Lilly's TuneLab frontier AI unit will deploy Chai's custom-trained model for biologics discovery across multiple drug targets, a collaboration designed to compress discovery timelines from months to weeks." |
| SI010 | TechCrunch | OpenAI-Backed Biotech Firm Chai Discovery Raises $130M Series B at $1.3B Valuation | "Chai Discovery has raised $130 million at a $1.3 billion valuation, achieving unicorn status in under two years—one of the fastest valuations in AI drug discovery." |
| SI011 | Loon Bio | AI Drug Discovery's $60B Reality Check: Hype, Failures, and the Market Access Blindspot | "Despite over $60 billion in venture capital pouring into AI-driven drug discovery since 2015 and myriad computational breakthroughs, as of 2025 not a single AI-designed drug has achieved FDA approval." |
| SI012 | S&P Global Market Intelligence | Where Is Venture Capital in Biopharma Going in 2026? | "Biopharma venture capital in 2026 continues to concentrate disproportionately in AI-enabled preclinical platforms, with Series B+ deals increasingly requiring credible commercialization evidence beyond purely technical achievement." |
| SI013 | S&P Global Market Intelligence | Biopharma Venture Capital 2025 Outlook | "The 2025 biopharma venture capital landscape reflects continued enthusiasm for AI-enabled drug discovery platforms, with capital increasingly favoring companies with at least one pharma partnership as commercial validation signal." |
| SI014 | PDA | The AI Revolution in Drug Discovery | "AI platforms in drug discovery typically generate value through milestone-based payment structures tied to discovery outcomes; the regulatory validation gap between computational hit and clinical proof-of-concept represents the highest-risk commercial transition." |
| SI015 | Maven Bio | Top Pharmaceutical R&D Spending in 2025: How Big Pharma Is Allocating Capital | "The top pharmaceutical companies collectively allocate more than $200 billion annually to R&D, with AI-software tools for discovery representing a small but rapidly growing share; individual platform licensing deals typically range from $1M to $20M+ per year per engagement." |
| SI016 | Genetic Engineering & Biotechnology News | Artificial Intelligence Topics — AI in Bioprocessing and Drug Discovery | "Artificial intelligence and outsourcing together are driving structural cost reductions in bioprocessing and discovery workflows, enabling capital-light platform companies to generate discovery data without maintaining costly internal laboratory infrastructure." |
| SI017 | BioSpace | BioSpace News — Biotech and Pharma Industry Updates | "The biotech industry in 2026 continues to grapple with regulatory and commercial headwinds, with AI-enabled discovery platforms facing increasing pressure to demonstrate clinical translation alongside computational achievement." |
| SI018 | Endpoints News | Endpoints News — Biopharma News and Analysis | "Biopharma regulatory dynamics in 2026 continue to shape the timeline and economics of drug discovery programs, with increasing scrutiny of data integrity and evidence standards across both traditional and AI-enabled development pathways." |
| SI019 | BusinessWire | Chai Discovery Unveils Chai-2 Breakthrough: Achieving Fully De Novo Antibody Design With AI | "Chai-2 enables fully de novo antibody design with no experimental templates or prior examples required; access is governed by a Responsible Deployment policy with selective partner access." |
| SI020 | General Catalyst | Our Investment in Chai Discovery | "We believe early pharma adopters of AI-driven drug design tools, such as Chai's platform, may see first-in-class biologics entering clinical trials as early as the end of 2027." |
| SI021 | BuiltIn SF | Chai Discovery Secures $70M Series A | "Chai Discovery, the San Francisco-based AI drug discovery startup, has secured $70 million in Series A funding; the company currently employs approximately 29 people." |
| SI022 | FierceBiotech | Chai Infuses AI Drug Discovery Efforts With $130M Series B | "Chai Discovery has closed a $130 million Series B round as it enters the commercialization phase of its AI-powered molecular discovery platform." |
| SI023 | Mordor Intelligence | Antibody Discovery Market Report | "The global antibody discovery market is valued at approximately $10.75 billion in 2026 and growing at 22.4% CAGR, driven by AI/ML-enabled platforms accelerating discovery throughput." |
| SI024 | Mordor Intelligence | AI in Pharmaceutical R&D Market Report | "AI in pharmaceutical R&D is a high-growth segment with significant commercial opportunity for platform companies that can demonstrate translation from computational design to clinically relevant candidates." |
| SI025 | IQVIA | IQVIA Institute's Global R&D Trends 2026: Credible Signal on AI-Enabled Programs | "IQVIA's 2026 global R&D trends report finds credible signal that AI-enabled programs are beginning to influence productivity metrics at the preclinical stage, though clinical-stage evidence for AI-specific advantages remains limited." |
| SI026 | Bloomberg | OpenAI-Backed Chai Discovery Raises $130 Million for AI-Designed Molecules | "Chai Discovery raised $130 million at a $1.3 billion valuation, backed by OpenAI, Oak HC/FT, and General Catalyst, to build a computer-aided design suite for molecular discovery." |
| SI027 | Chai Assets | Chai-2 Technical Report | "Chai-2 achieves a 16% hit rate across 52 diverse antibody targets in zero-shot de novo design, with wet-lab validation completed in under two weeks per target." |
| SI028 | Intuition Labs | AI Biologics Design: Chai Discovery and Eli Lilly Collaboration | "The Chai-Lilly collaboration represents a new model for AI drug discovery deal-making: a custom foundation model trained on proprietary pharma data, deployed within the pharma partner's internal AI unit, with financial terms structured around discovery milestones." |
| SI029 | Observer | Chai Discovery Is the Latest AI Unicorn | "Chai Discovery's $1.3 billion valuation makes it the latest AI unicorn in the drug discovery space, reflecting investor conviction that AI biologics design can deliver commercial value ahead of clinical validation." |
| SI030 | Drug Discovery Trends | 2024: The Year AI Drug Discovery and Protein Structure Prediction Took Center Stage | "2024 marked the year AI drug discovery matured from research curiosity to commercial platform, with structure prediction models achieving pharma-grade accuracy and the first commercial AI drug discovery deals being announced across the sector." |
| SI031 | Eli Lilly and Company (Investor Relations) | Eli Lilly SEC Filings – 8-K and Annual Reports (Fiscal 2025–2026) | Eli Lilly's investor relations SEC filings page provides access to 8-K current reports and annual 10-K filings; no Chai Discovery-specific material contract 8-K has been separately identified as of May 2026, indicating the Lilly-Chai collaboration may be below the materiality threshold for standalone 8-K disclosure, or is disclosed within a subsequent quarterly or annual filing. |
| SE001 | Hugging Face | chaidiscovery/chai-1 — Chai-1 Model Card on HuggingFace Hub | Chai-1 is a multi-modal foundation model for molecular structure prediction that performs at the state-of-the-art across a variety of benchmarks. |
| SE002 | Python Package Index (PyPI) | chai-lab — PyPI Python Package Distribution | |
| SE003 | RCSB Protein Data Bank | RCSB PDB — Research Collaboratory for Structural Bioinformatics Protein Data Bank | |
| SE004 | Nature (Springer Nature) | Accurate structure prediction of biomolecular interactions with AlphaFold 3 | Abramson, J., Adler, J., Dunger, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630, 493–500 (2024). |
| SE005 | Google DeepMind | AlphaFold — Transforming our understanding of biology | |
| SE006 | PLOS Computational Biology | LAP: Liability Antibody Profiler by Sequence & Structural Mapping of Natural and Therapeutic Antibodies | |
| SE007 | CASP — Critical Assessment of Protein Structure Prediction | CASP15 — Critical Assessment of Protein Structure Prediction, 15th Round | |
| SE008 | eLife Sciences Publications | eLife — Open Access Peer-Reviewed Life Sciences Journal (Background Reference) | |
| SE009 | bioRxiv (Cold Spring Harbor Laboratory) | Chai-1: Decoding the molecular interactions of life | We introduce Chai-1, a multi-modal foundation model for molecular structure prediction that performs at the state-of-the-art across a variety of tasks relevant to drug discovery. |
| SE010 | bioRxiv (Cold Spring Harbor Laboratory) | Zero-shot antibody design in a 24-well plate | Chai-2 achieves a 16% hit rate in fully de novo antibody design, representing an over 100-fold improvement compared to previous computational methods. |
| SE011 | bioRxiv (Cold Spring Harbor Laboratory) | Drug-like antibody design against challenging targets with atomic precision | >86% of these full-length mAbs have strong developability profiles on par with therapeutic antibodies. |
| SE012 | Business Wire (Chai Discovery) | Chai Discovery Unveils Chai-2 Breakthrough, Achieving Fully De Novo Antibody Design With AI | Chai-2 demonstrates a remarkable antibody design hit rate close to 20%. Prompted with just the target and epitope, Chai-2 successfully designs all complementarity-determining regions (CDRs) entirely from scratch. |
| SE013 | Chai Discovery | Chai Discovery — Official Homepage | Drug-like antibody design against challenging targets with atomic precision |
| SE014 | GitHub (Chai Discovery) | chaidiscovery/chai-lab — GitHub Repository | Chai-1 is released under an Apache 2.0 License (both code and model weights), which means it can be used for both academic and commercial purposes, including for drug discovery. |
| SE015 | Chai Discovery | Chai Discovery Lab — Web Interface for Chai-1 Predictions | |
| SE016 | Chai Discovery | Chai-2 Technical Report — Chai Discovery | |
| SE017 | Fierce Biotech | Chai infuses AI drug discovery efforts with $130M Series B | Earlier this month, the Chai team released a preprint—a scientific paper that has not yet been peer-reviewed by other experts— claiming that they had used Chai 2 to develop monoclonal antibodies aimed against tough-to-drug targets. |
| SE018 | TechCrunch | From OpenAI's offices to a deal with Eli Lilly — how Chai Discovery became one of the flashiest names in AI drug development | Every line of code in our codebase is homegrown. We're not taking LLMs off the shelf that are in the open source [ecosystem] and fine-tuning them. These are highly custom architectures. |
| SE019 | Medvolt AI | Chai-2: De Novo Antibody Design — AI Breakthrough Analysis | Chai Discovery is selectively offering early access to academic and biopharma partners under a Responsible Deployment Framework. The company is focused on: Supporting health-positive, low-risk applications. |
| SE020 | Intuition Labs | AI Biologics Design: Chai Discovery and Eli Lilly Collaboration Analysis | |
| SE021 | The AI Insider | Chai Discovery Announces $70M Series A to Transform Molecular Design | Before Chai-2, the process was not unlike searching a giant bunch of keys for the right fit for a lock — but there are millions of keys. Now, it's like having a master locksmith design exactly the right shape key, based only on your description of the lock. |
| SE022 | Loon Bio | AI Drug Discovery's $60 Billion Reality Check: Hype, Failures, and the Market Access Blindspot | The FDA has been proactive in addressing AI in drug development, releasing its first-ever draft guidance in January 2025 titled 'Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products.' |
| SE023 | HIT Consultant | Eli Lilly and Chai Discovery Frontier AI Partnership | |
| SE024 | Business Wire (Chai Discovery) | Chai Discovery Announces Collaboration with Eli Lilly and Company to Accelerate Biologics Discovery | |
| SE025 | BiopharmaTrend | Chai Discovery Raises $70M to Expand Zero-Shot Biologics Design AI Platform | |
| SU001 | BusinessWire | Chai Discovery Announces Collaboration with Eli Lilly and Company to Accelerate Biologics Discovery | Chai Discovery, the artificial intelligence company transforming biology from science into engineering, today announced a collaboration with Eli Lilly and Company to apply Chai's frontier AI to multiple biologic targets and train a custom Chai model on Lilly's proprietary data. |
| SU002 | TechCrunch | From OpenAI's offices to a deal with Eli Lilly — how Chai Discovery became one of the flashiest names in AI drug development | Lilly's TuneLab program, led by Aliza Apple, is the internal program that brought in Chai; the collaboration followed a period of evaluation. |
| SU003 | HITConsultant | Eli Lilly Taps Chai Discovery Frontier AI to Accelerate Biologics | Eli Lilly will use Chai's frontier AI across multiple biologic targets as part of its internal antibody discovery programs. |
| SU004 | IntuitionLabs | AI Biologics Design: Chai Discovery and Eli Lilly Collaboration Analysis | The Lilly deal represents a shift from pilot to production for AI-designed biologics, with Chai deploying across multiple active programs simultaneously. |
| SU005 | BusinessWire | Chai Discovery Unveils Chai-2 Breakthrough — Achieving Fully De Novo Antibody Design With AI | Chai-2 is now available for early access to select partners — Chai invites biotech and pharmaceutical companies to apply for access through its Responsible Deployment program. |
| SU006 | BusinessWire | Chai Discovery Announces $70 Million Series A To Transform Molecular Design | A meaningful fraction of the biotech industry has already applied for access to Chai-2. |
| SU007 | BusinessWire | Chai Discovery Announces $130 Million Series B To Transform Molecular Discovery | |
| SU008 | General Catalyst | Our Investment in Chai Discovery | |
| SU009 | Observer | Chai Discovery Is the Latest AI Unicorn Betting on Drug Discovery | |
| SU010 | FierceBiotech | Chai infuses AI drug discovery efforts with $130M Series B | |
| SU011 | BioPharma Trend | Chai Discovery Raises $70M to Expand Zero-Shot Biologics Design AI Platform | |
| SU012 | MedVolt | Chai-2: De Novo Antibody Design AI Breakthrough | |
| SU013 | The AI Insider | Chai Discovery Announces $70M Series A to Transform Molecular Design | |
| SU014 | LoonBio | AI Drug Discovery's $60 Billion Reality Check — Hype, Failures, and the Market-Access Blindspot | As of 2025, no drug designed end-to-end by artificial intelligence has received regulatory approval. The $60 billion invested in AI drug discovery has yet to produce a single FDA-approved drug. |
| SU015 | Chai Discovery | Chai Discovery — Official Website | |
| SU016 | Longevity Technology | AI drug discovery startup valued at $1.3B in huge funding round | |
| SU017 | Biotech Industry Examiner | OpenAI-backed Chai Discovery: $130M Series B and AI Antibody Design | |
| SU018 | GitHub API | chaidiscovery/chai-lab — Repository Metadata (GitHub REST API) | stargazers_count: 1938, forks_count: 274, open_issues_count: 87, created_at: 2024-09-06, pushed_at: 2026-04-15 |
| SU019 | GitHub API | chaidiscovery/chai-lab — Releases List (GitHub REST API) | |
| SU020 | CB Insights | Chai Discovery — Company Profile and AI 100 2026 Recognition | Chai Discovery named to CB Insights AI 100 2026 in the Healthcare and Life Sciences category; company grew from $150M to $1.3B valuation in approximately 15 months. |
| SU021 | Crunchbase | Chai Discovery — Crunchbase Company Profile | |
| SU022 | bioRxiv | Chai-1: Decoding the molecular grammar of proteins, nucleic acids, and small molecules — Version 2 | |
| SU023 | bioRxiv | Chai-2: Toward Fully De Novo Antibody Design with AI — Full PDF (v1) | |
| SU024 | Hugging Face | chaidiscovery — Chai Discovery Organization on Hugging Face | |
| SU025 | Python Package Index (PyPI) | chai_lab — Chai-1 Python Package (underscore variant) | chai_lab 0.6.1 — Released March 2025. Chai-1 inference package for protein structure prediction. |
| SU026 | Built In San Francisco | Chai Discovery Secures $70M Series A | |
| SU027 | Bloomberg | OpenAI-Backed Chai Discovery Raises $130 Million for AI-Designed Molecules | |
| SU028 | TechCrunch | OpenAI-backed biotech firm Chai Discovery raises $130M Series B at $1.3B valuation | |
| SU029 | Analytics India Magazine | OpenAI-Backed Biotech Startup Chai Discovery Raises $70 Million Led by Anthology | |
| SU030 | bioRxiv | Chai-2 Challenging Targets — Supplementary Technical Report (v2) | |
| SR001 | Chai Discovery | Chai Discovery | With Chai-2, we’re moving de novo antibody design past binding and closer than ever to real therapeutics. |
| SR002 | Chai Discovery | Introducing Chai-1 | The model is available for free via a web interface, including for commercial applications such as drug discovery. We are also releasing the model weights and inference code as a software library under an Apache 2.0 License. |
| SR003 | Nature | Accurate structure prediction of biomolecular interactions with AlphaFold 3 | Here we describe AlphaFold 3, a deep learning system that predicts the joint structure of complexes including proteins, nucleic acids, small molecules, ions and modified residues. |
| SR004 | bioRxiv | Chai-1: Decoding the molecular interactions of life (v1) | We introduce Chai-1, a multi-modal foundation model for molecular structure prediction that performs at the state-of-the-art across a variety of tasks relevant to drug discovery. |
| SR005 | bioRxiv | Chai-1: Decoding the molecular interactions of life (v2) | We introduce Chai-1, a multi-modal foundation model for molecular structure prediction that performs at the state-of-the-art across a variety of tasks relevant to drug discovery. |
| SR006 | U.S. Food and Drug Administration | Artificial Intelligence in Software as a Medical Device | Artificial intelligence (AI) and machine learning (ML) technologies have the potential to transform health care. |
| SR007 | Absci | Home | Absci | We're unlocking novel biology and creating better biologics with AI. |
| SR008 | Generate Biomedicines | Generate Biomedicines | Generate Biomedicines is a new kind of therapeutics company—existing at the intersection of machine learning, biological engineering, and medicine. |
| SR009 | EvolutionaryScale | EvolutionaryScale | Frontier AI for the life sciences. |
| SR010 | U.S. Food and Drug Administration | Artificial Intelligence for Drug Development | Artificial Intelligence (AI) refers to a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments. |
| SR011 | Nature | AlphaFold3 — why did Nature publish it without its code? | Criticism of our decision to publish AlphaFold3 raises important questions. |
| SR012 | Crunchbase (archived via Wayback) | Chai Discovery - Crunchbase Company Profile & Funding | Chai Discovery creates Chai-1, a multi-modal foundation model designed for predicting molecular structures relevant to drug discovery. |
| SR013 | GitHub | GitHub - chaidiscovery/chai-lab: Chai-1, SOTA model for biomolecular structure prediction | Chai-1 is a multi-modal foundation model for molecular structure prediction that performs at the state-of-the-art across a variety of benchmarks. |
| SR014 | GitHub | Releases · chaidiscovery/chai-lab | v0.6.1 |
| SR015 | GitHub | chai-lab/LICENSE at main · chaidiscovery/chai-lab | LICENSE |
| SR016 | BusinessWire | Chai Discovery Announces Collaboration with Eli Lilly and Company to Accelerate Biologics Discovery | Chai Discovery’s AI platform, including purpose-built custom models, will be deployed to accelerate the discovery of next-generation therapeutics. |
| SR017 | GitHub | alphafold3/WEIGHTS_TERMS_OF_USE.md at main · google-deepmind/alphafold3 | The Model Parameters are made available for Non-Commercial Use only. |
| SR018 | PubMed Central | Application of protein language models for antibody developability prediction | However, their performance and reliability in real-world industrial antibody discovery pipelines remain underexplored. |
| SR019 | MDPI Molecules | Revolutionizing Molecular Design for Innovative Therapeutic Applications through Artificial Intelligence | Key challenges remain in bridging computational design with experimental validation and clinical translation. |
| SR020 | Frontiers in Bioengineering and Biotechnology | Responsible AI in biotechnology: balancing discovery, innovation and biosecurity risks | The integration of artificial intelligence (AI) in protein design presents unparalleled opportunities for innovation in bioengineering and biotechnology. However, it also raises significant biosecurity concerns. |
| SR021 | PLOS Computational Biology | Dual-use capabilities of concern of biological AI models | Dual-use capabilities of concern of biological AI models. |
| SR022 | Center for Health Security | AI and biosecurity: The need for governance | Governments should evaluate advanced models and if needed impose safeguards. |
| SR023 | TechCrunch | OpenAI-backed biotech firm Chai Discovery raises $130M Series B at $1.3B valuation | Chai Discovery, a biotech startup with backing from OpenAI, announced a $130 million Series B round at a $1.3 billion valuation on Monday. |
| SR024 | Yahoo Finance / Business Wire | Chai Discovery Announces $70 million Series A To Transform Molecular Design | Chai Discovery ... today announced its $70 million Series A financing round. |
| SR025 | Forbes | Chai Discovery | Company Overview & News | The company has already landed a partnership with Eli Lilly to accelerate the discovery of new drugs before the clinical trial phase. In December, Chai raised $130 million at a $1.3 billion valuation. |
| SR026 | U.S. Food and Drug Administration | Chemistry, Manufacturing and Controls: Regulatory Considerations Through Clinical Development | FDA has 30 days to review IND submissions. |
| SR027 | mAbs / Taylor & Francis | Blueprint for antibody biologics developability | Common issues and risks encountered during developability assessments, such as aggregation, self-interaction, hydrophobicity, deamidation and oxidation, are explored. |
| SR028 | Chai Discovery | Chai Discovery Lab | We'll email you a magic link. |
| SR029 | Hugging Face | chaidiscovery (Chai Discovery) | chaidiscovery/chai-1 Updated Feb 18, 2025 |
| SR030 | PyPI | chai_lab | Details for the file chai_lab-0.6.1.tar.gz. Upload date: Mar 18, 2025. |
| SV001 | Business Wire | Chai Discovery Announces $130 Million Series B To Transform Molecular Discovery | This round of financing values the company at $1.3 billion. |
| SV002 | TechCrunch | OpenAI-backed biotech firm Chai Discovery raises $130M Series B at $1.3B valuation | The firm’s total funding now stands at over $225 million. |
| SV003 | Fierce Biotech | Chai infuses AI drug discovery efforts with $130M series B | The round brings Chai’s total valuation to $1.3 billion. |
| SV004 | Business Wire | Chai Discovery Announces Collaboration with Eli Lilly and Company to Accelerate Biologics Discovery | Chai has raised nearly $230M to date. |
| SV005 | TechCrunch | From OpenAI’s offices to a deal with Eli Lilly — how Chai Discovery became one of the flashiest names in AI drug development | In December, the company completed its Series B, bringing in an additional $130 million and a valuation of $1.3 billion. |
| SV006 | General Catalyst | Our Investment in Chai Discovery | Josh helped pioneer frontier AI in the early days of OpenAI and at Meta where he co-led development of ESM1. |
| SV007 | bioRxiv | Zero-shot antibody design in a 24-well plate | fully de novo antibody design, representing an over 100-fold improvement compared to previous computational methods |
| SV008 | Business Wire | Chai Discovery Unveils Chai-2 Breakthrough, Achieving Fully De Novo Antibody Design With AI | The company’s latest model, Chai-2, demonstrates a remarkable antibody design hit rate close to 20%. |
| SV009 | Recursion Pharmaceuticals Investor Relations | Recursion Reports First Quarter Financial Results and Provides Business Update | Total revenue, consisting primarily of revenue from collaboration agreements, was $6.5 million for the first quarter of 2026. |
| SV010 | U.S. Securities and Exchange Commission | Recursion Pharmaceuticals, Inc. Form 10-Q for quarter ended March 31, 2026 | We do not have any products approved for commercial sale and have not generated any revenues from product sales. |
| SV011 | Stock Analysis | Recursion Pharmaceuticals (RXRX) Statistics & Valuation | RXRX has a market cap or net worth of $1.65 billion. The enterprise value is $1.07 billion. |
| SV012 | CompaniesMarketCap | Recursion Pharmaceuticals (RXRX) - Market capitalization | As of May 2026 Recursion Pharmaceuticals has a market cap of $1.65 Billion USD. |
| SV013 | Stock Analysis | Schrödinger (SDGR) Statistics & Valuation | Schrödinger has a market cap or net worth of $988.56 million. The enterprise value is $696.60 million. |
| SV014 | CompaniesMarketCap | Schrödinger (SDGR) - Market capitalization | As of May 2026 Schrödinger has a market cap of $0.98 Billion USD. |
| SV015 | Business Wire | Schrödinger Reports First Quarter 2026 Financial Results | Total revenue was $58.6 million, a 2% decrease. |
| SV016 | Absci Investor Relations | Absci Reports Business Updates and First Quarter 2026 Financial and Operating Results | Absci believes its cash, cash equivalents, and marketable securities will be sufficient to fund its operating plans into the first half of 2028. |
| SV017 | U.S. Securities and Exchange Commission | Absci Corporation Form 10-Q for quarter ended March 31, 2026 | Revenue was $0.2 million for the three months ended March 31, 2026 compared to $1.2 million for the three months ended March 31, 2025. |
| SV018 | Stock Analysis | Absci (ABSI) Statistics & Valuation | Absci has a market cap or net worth of $793.56 million. The enterprise value is $672.27 million. |
| SV019 | CompaniesMarketCap | Absci (ABSI) - Market capitalization | As of May 2026 Absci has a market cap of $0.79 Billion USD. |
| SV020 | PR Newswire | Generate Biomedicines, Inc. Reports First Quarter 2026 Financial Results and Provides Business Update | Cash, cash equivalents, and marketable securities were $516.6 million as of March 31, 2026. |
| SV021 | CompaniesMarketCap | Generate Biomedicines (GENB) - Market capitalization | As of May 2026 Generate Biomedicines has a market cap of $1.79 Billion USD. |
| SV022 | Isomorphic Labs | Isomorphic Labs announces Series B investment round | Isomorphic Labs announces it has raised $2.1 Billion in Series B funding. |
| SV023 | PR Newswire | Isomorphic Labs announces $600 million funding to further develop its next-generation AI drug design engine and advance therapeutic programs into the clinic | Isomorphic Labs ... has raised $600 Million in its first external funding round. |
| SV024 | Isomorphic Labs | Isomorphic Labs kicks off 2024 with two pharmaceutical collaborations | These partnerships have the potential to be worth nearly $3 billion to Isomorphic Labs. |
| SV025 | Fierce Biotech | New AI drug discovery powerhouse Xaira rises with $1B in funding | The company emerged Tuesday with $1 billion in committed funding. |
| SV026 | Science Reader | 173 AI Drugs in Trials, Zero Approved: What 2026 Is Missing | No AI-designed drug has received regulatory approval yet. |
| SV027 | DrugPatentWatch | AI in Drug Discovery 2026: What Actually Works, What Remains Hype, and Where the IP Value Sits | Bringing a New Molecular Entity (NME) to market costs, on average, US $2.8 billion. |
| SV028 | MarketBeat | Generate Biomedicines (GENB) Stock Price, News & Analysis | Average Price Target for Generate Biomedicines $25.40 |
| SV029 | Stock Analysis | Generate Biomedicines (GENB) Stock Price & Overview | According to 6 analysts, the average rating for GENB stock is Strong Buy. |
| SV030 | Chai Discovery | Chai Discovery news page | The retained public news archive highlights funding, product, and Lilly partnership announcements. |