Periodic Labs
顶尖 AI 科学团队、异常强的融资能力,以及远超验证进度的估值
Periodic Labs 把顶级前沿 AI 与材料科学创始团队、激进的自主实验室投资逻辑和极强投资人背书捆在一起,但公司离商业化验证仍很远;据报道的 2026 年融资估值,在公开收入或客户证据足以支撑前,已经提前计入成功。
封面要素
公司概况
Periodic Labs 是一家旧金山初创公司,由前 OpenAI 后训练负责人 Liam Fedus 和前 Google DeepMind 材料科学负责人 Ekin Dogus Cubuk 于 2025 年创立。公司在打造「AI 科学家」:把前沿模型、仿真工作流和自主实验室基础设施接成闭环,用来生成假设、运行实验并发现新材料。公开描述的应用包括高温超导体和半导体散热问题;早期商业活动据称覆盖半导体、航天和国防客户。Periodic 于 2025 年 9 月完成创纪录的 $300M 种子轮,投后估值约 $1.3B;2026 年又被报道正在讨论约 $7.5B 估值的新一轮融资。核心尽调问题不是团队是否出色——而是公司能否足够快地把科学前景变成可重复的商业结果,以支撑这次估值跃迁。
- 成立时间
- 2025-03-01
- 创始人
- Liam Fedus, Ekin Dogus Cubuk
- 创立地点
- San Francisco, CA, USA
- 总部
- San Francisco, CA, USA
- 产品
- Periodic Labs 面向物理 R&D 销售 AI 科学工作流:模型提出假设、运行仿真、解读客户实验数据,并最终接入自主机器人实验室,执行实验后把结果回灌系统。近期商业切口是为先进工业 R&D 团队定制解题,包括一个已公开描述的半导体散热项目。
- 客户
- 先进工业 R&D 团队,尤其是半导体、材料、能源、航空航天、航天和国防机构里的工程师与研究人员;这些组织实验预算大,面对的是难优化的物理世界问题。
- 商业模式
- 目前业务看起来是面向企业 R&D 客户的合同制 AI 科学服务和定制智能体部署;长期上行来自自研数据授权、发现 IP 授权,以及可能把平台发现的突破性材料商业化。定价为定制化,未公开披露。
- 阶段
- Seed
- 融资情况
- 已确认融资包括 2025 年 9 月 30 日宣布的 $300M 种子轮,由 Andreessen Horowitz 领投;披露的支持者还包括 Felicis、Accel、NVIDIA 的风险投资部门、Jeff Bezos、Eric Schmidt、Jeff Dean 和 Elad Gil。2026 年 5 月的公开报道称,公司正深入洽谈以约 $7.5B 估值融资至少 $500M,据报道由 AMP 领投。
执行摘要
主要优势
- 创始人履历极强:Fedus 曾领导 OpenAI 后训练工作,支撑 ChatGPT 时代模型;Cubuk 曾主导 DeepMind 材料方向,并参与打造 GNoME 式 AI 发现流程。
- 产品逻辑把前沿模型、仿真和物理实验连成闭环;如果实验室栈能在生产中跑通,就可能沉淀专有数据并滚出复利优势。
- 融资质量在 2025 年成立的创业公司里很少见:a16z 领投 $300M 种子轮,Bezos、Schmidt、NVIDIA、Accel 和 Felicis 入局,带来资金、信号和招人杠杆。
- 公开材料至少显示公司已在半导体及相邻领域产生商业接触和收入,说明它不只是研究项目。
主要风险
- 据报道的 2026 年 $7.5B 估值很难支撑:公司未披露 ARR、收入运行率、客户数,也没有证明自主实验室能以商业规模运转。
- 自主材料发现尚未在生产可靠性上被验证;从有潜力的模型和试点流程,走到稳定的物理实验室产出,是重大执行风险。
- 客户证据偏薄:没有具名客户、独立案例研究和合同金额披露,企业 R&D 销售周期还可能长达 12-24 个月,复用性很难看清。
- 模式吃资本,昂贵算力、仿真、科学人才和实验室基础设施都要砸钱;即便完成种子轮,后续融资需求仍可能很大。
- 关键人集中度很高,投资逻辑严重依赖少数高度差异化的创始人和资深研究员。
未决问题
- 公司未公开收入、ARR、定价、毛利率或财务报表,单位经济模型无法承销。
- 公司未披露具名客户背书、合同期限、续约数据或客户数,商业牵引质量仍只能部分推断。
- 目前没有公开证据证明 Periodic 的自主实验室已在工业规模持续交付商业发现成果。
- 据报道的 2026 年 $500M 融资和 $7.5B 估值仍来自媒体报道,而非已完成并有官方文件支持的交易。
- 客户数据和 Periodic 自营实验产生的 IP,其所有权和变现条款未公开说明。
目录
01公司概况
1.1 公司身份、使命与产品策略
Periodic Labs 是一家总部位于旧金山的人工智能公司,2025 年 3 月成立,目标很明确:创造 AI 科学家,也就是能够形成假设、设计并执行物理实验室实验,并在物理科学发现中持续迭代的自主系统。公司仍是私营种子阶段实体,没有公开财务披露。它的核心论点是,前沿 AI 模型已经基本吃尽互联网上约 10 万亿个文本 token;要取得有意义的科学进展,AI 系统必须直接从物理世界生成新的实验数据。Periodic Labs 的产品策略围绕三根互相咬合的支柱:第一,自主机器人实验室,在全自动闭环中执行粉末合成和材料表征实验;第二,高保真 AI 驱动仿真模型,在高成本物理测试前缩小实验搜索空间;第三,大语言模型研究助手,分析结果、生成新假设并指挥后续实验周期。每次实验都可能产生数 GB 自研数据,包括传统科学很少发表的负结果,进而形成竞争对手拿不到的训练语料。公司第一个商业应用,是帮助一家半导体制造商解决芯片散热问题,用该制造商的实验数据训练定制 AI 智能体。更长期的科学登月目标,是发现更接近常温条件下工作的高温超导体;一旦成立,可能改变电网、交通和计算。公司也瞄准航天和国防客户,这些领域的材料 R&D 周期同样昂贵且缓慢。Periodic Labs 明确把自己同消费级 AI 创业公司区分开来:它不追求通用人工智能,也不做聊天用大语言模型,而是完全聚焦物理科学发现闭环,在那里,自然本身就是强化学习环境。[CO001, CO007, CO008, CO009, CO010, CO028]
| 指标 | 数值 / 状态 | 日期 | 置信度 | 缺口 / 备注 |
|---|---|---|---|---|
| 总部 | San Francisco CA | 2026-06 | 高 | 多个来源确认 |
| 阶段 | 种子轮 / Pre-Series A | 2026-06 | 高 | $300M 种子轮之后;后续轮待定 |
| 累计融资 | $300M(种子轮) | 2025-09-30 | 高 | Wilson Sonsini 确认;后续轮尚未关闭 |
| 投后估值(种子轮) | $1.3B | 2025-09-30 | 高 | Wilson Sonsini 和一线媒体确认 |
| 据报道后续轮估值 | ~$7.5B(深入谈判) | 2026-05 | 中 | 据 Forbes 和 Bloomberg;交易尚未确认关闭 |
| 员工数 | 32-48 名员工(估计) | 2026-Q1 | 低 | 来自商业目录的区间;公司未披露 |
| 收入 / ARR | 未公开披露 | 2026-06 | 低 | 私有公司;半导体客户已确认,但条款未知 |
| 创始人 | Liam Fedus(CEO)和 Ekin Dogus Cubuk | 2025-03 | 高 | 多个一手来源确认 |
快照数值截至运行日期 2026-06-10。种子轮投后估值由 Wilson Sonsini 法律顾问和一线媒体确认。后续轮估值来自 Forbes 和 Bloomberg 报道的交易谈判,可能并不代表已关闭交易。员工数依据第三方目录估计;公司未披露官方数据。这家私有公司的收入和 ARR 未公开。
[CO001, CO006, CO023, CO024, CO032, CO034]闭环价值链把 AI 假设引擎、自主机器人实验室、模拟层和商业产出连在一起,展示各组件如何相互供给并生成自有实验数据。
[CO007, CO008, CO028, CO030]截至 2026 年 6 月运行日,Periodic Labs 的关键业绩和身份指标凸显融资规模、估值轨迹和团队厚度, 同时也暴露财务披露缺口。
[CO003, CO006, CO023, CO024, CO034]1.2 创始团队、领导层与治理
Periodic Labs 由 Liam Fedus 和 Ekin Dogus Cubuk 共同创立,两人都是过去十年最有履历背书的 AI 与材料科学研究者。担任 CEO 的 Fedus 拥有 MIT 物理学 BS(2010)、UC San Diego 物理学 MS(2016),以及 Universite de Montreal 和 MILA 计算机科学 PhD(2020),导师包括 Yoshua Bengio 和 Hugo Larochelle。在 Google Brain,他是 Switch Transformer 论文(2021)的第一作者;该架构是稀疏混合专家语言模型扩展到万亿参数的参考方案。2022 年他加入 OpenAI,任高级研究科学家,随后成为数据飞轮负责人并共同创造 ChatGPT,领导 GPT-4o、o1-mini 和 o1-preview 的后训练,2024 年 10 月升任后训练研究副总裁,2025 年 3 月离职。Cubuk 获得 Harvard PhD,并在 Stanford 完成博士后,之后在 Google Brain 和 Google DeepMind 领导材料与化学研究团队;他共同署名 2023 年 GNoME 论文,识别出约 220 万个新的稳定晶体结构。同年,他还共同发表了一个里程碑演示:全自动机器人 A-Lab 按 AI 生成配方,在 17 天内合成 41 种新化合物。除两位联合创始人外,创始团队还包括 OpenAI o1 和 o3 推理模型创造者 Alexandre Passos;曾发现超导体的材料科学家 Eric Toberer;以及 Microsoft MatterGen 材料科学生成式 AI 系统创造者 Matt Horton。另有 20 多名研究者来自 Meta、OpenAI、DeepMind、Databricks 和 Samsung,其中许多人放弃了大量未归属股权加入。科学顾问委员会由 Stanford 诺贝尔化学奖得主 Carolyn Bertozzi 主持,并包括超导物理和材料科学权威。Wilson Sonsini Goodrich and Rosati 为种子轮交易提供法律顾问。关键人物集中风险很实质:两位联合创始人都是投资逻辑和投资人信心的中心,公司也没有公开披露接班计划。[CO011, CO012, CO013, CO014, CO015, CO016]
| 人物 | 职务 | 背景 | 创始人-市场匹配 | 关键人物依赖 |
|---|---|---|---|---|
| Liam Fedus | 联合创始人 CEO | MIT 物理学 BS;MILA 计算机科学 PhD(Bengio/Larochelle);OpenAI 训练后 VP;ChatGPT、GPT-4o、o1 共同创造者 | 深厚的 LLM 训练后专长和物理背景,正是 AI 科学家架构所需 | 关键——主要外部门面和融资人;离职会引发投资者担忧 |
| Ekin Dogus Cubuk | 联合创始人 | Harvard PhD;Stanford 博士后;曾领导 Google Brain 和 DeepMind 的 Materials/Chemistry;GNoME 和 A-Lab 论文共同作者 | 世界级材料科学 AI 专长;通过同行评审突破验证过自动化实验室范式 | 关键——掌握科学领域专长和既有实验室自动化智力资本 |
| Alexandre Passos | 高级研究员 | OpenAI o1 和 o3 推理模型共同创造者 | 前沿推理模型专长对假设生成管线至关重要 | 高——罕见的 o 系列模型架构经验 |
| Eric Toberer | 研究员(材料科学家) | 既有超导体发现;物理材料科学专家 | 实验室验证需要领域化学和物理经验 | 中——专长稀缺,但长期可替代 |
| Matt Horton | 研究员 | Microsoft MatterGen 和 MatterSim 生成式材料科学工具创造者 | 带来构建 Periodic 生成式材料管线最接近先例的直接经验 | 中——从既有先例转向 Periodic 平台的关键人物 |
这是截至 2026 年 6 月一线新闻报道中公开点名团队成员的部分列举。公司未发布完整团队名册。据报道还有 20 多名研究人员来自 Meta、OpenAI、DeepMind、Databricks 和 Samsung,但公开来源未逐一列名。科学顾问委员会由诺贝尔奖得主 Carolyn Bertozzi 主持。
[CO011, CO013, CO014, CO015, CO016, CO017]1.3 融资历史、估值与投资人版图
2025 年 9 月 30 日,Periodic Labs 走出隐身状态,宣布以 $1.3B 投后估值融资 $300M,完成了风险投资史上最大的种子轮之一。本轮由 Andreessen Horowitz a16z 领投;第一张机构支票来自 Felicis Ventures 合伙人 Peter Deng,这位前 OpenAI 高管在公司尚未注册、尚未命名前就已承诺投资。其他机构投资方包括 DST Global、NVentures(NVIDIA 风险投资部门)和 Accel。天使团包括 Amazon 创始人 Jeff Bezos、前 Google CEO Eric Schmidt、Google 首席科学家 Jeff Dean 和投资人 Elad Gil。尽管 Fedus 的离职推文早期信号让外界以为 OpenAI 会投资,创始人向 TechCrunch 确认 OpenAI 并非 Periodic Labs 的支持者;他们选择 a16z,是为了获得更广的战略资源。约 $1.0B 的投前估值,反映的是投资人愿意为一家发布时没有产品、没有收入、没有披露 IP 的公司支付的价格。Wilson Sonsini Goodrich and Rosati 担任该交易法律顾问。到 2026 年 3 月,Bloomberg 报道 Periodic Labs 正洽谈交易,目标估值约 $7B。到 2026 年 5 月 7 日,Forbes 确认公司正深入洽谈以 $7.5B 估值融资至少 $500M,本轮由 AMP 领投;AMP 是前 a16z GP Anjney Midha 创立的投资载体。报道称本轮显著超额认购,并讨论紧随其后的、更高估值追加轮。不到九个月内,估值从 $1.3B 到 $7.5B,约六倍升值,属于种子阶段公司最快的已记录估值跃迁之一,反映出投资人对 AI for science 赛道的广泛信心。没有公开报道显示公司有老股交易、债务或授信安排。[CO003, CO004, CO005, CO006, CO023, CO024]
| 利益相关方 | 类型 | 轮次 | 经济 / 控制角色 | 尽调问题 |
|---|---|---|---|---|
| Andreessen Horowitz (a16z,领投方) | 领投机构 VC | 种子轮 $300M | 领投方;可能持有最大股权,并拥有董事会观察员或董事权利 | 确认董事会构成、按比例跟投权和保护性条款 |
| Felicis Ventures (Peter Deng) | 机构 VC(首张支票) | 种子轮 $300M | 首家机构投资;种子阶段领投关系;Deng 是 Fedus 的前 OpenAI 同事 | 确认 Felicis 持股规模,以及 Deng 是否拥有观察员权利 |
| DST Global | 机构 VC | 种子轮 $300M | 参与产品前阶段的后期科技基金;可能为少数股权 | 确认参与金额和任何信息权 |
| NVentures (NVIDIA) | 战略 VC | 种子轮 $300M | NVIDIA VC 部门;除资本外,可能提供算力访问或供应链优势 | 评估投资是否附带任何独占性、算力额度或技术许可 |
| Accel | 机构 VC | 种子轮 $300M | 美国 / 全球一线基金;少数股权 | 预计为标准信息权 |
| Jeff Bezos | 个人天使 | 种子轮 $300M | 通过 Bezos Expeditions 的个人投资;未披露治理角色 | 确认支票规模;评估是否跟进 Amazon/AWS 关系 |
| Eric Schmidt | 个人天使 | 种子轮 $300M | 前 Google CEO;个人投资;带来战略网络和可信度信号 | 未披露治理角色;可能存在顾问关系 |
| AMP / Anjney Midha | 领投方(后续轮) | 约 $500M,估值 $7.5B(深入谈判) | 前 a16z GP 领衔新工具 AMP;据报道为 2026 年融资领投方 | 确认关闭日期、治理条款,以及既有投资者是否维持按比例跟投 |
个人投资者的投资金额和持股比例未公开披露。后续轮(AMP/Anjney Midha)由 Forbes 在 2026 年 5 月 7 日报道;截至运行日期,交易尚未确认关闭。这家私有公司的董事会构成、清算优先权和保护性条款均未公开。
[CO003, CO005, CO006, CO023, CO026]1.4 关键里程碑与商业进展
从概念到成为风险投资史上估值最高的种子阶段公司之一,Periodic Labs 只用了不到 18 个月。Fedus 和 Cubuk 的创始对话发生在 2025 年 9 月亮相前约 7 个月,也就是大约 2025 年 2 月;当时两人判断,机器人自动化、材料仿真和 LLM 推理已经同时成熟,足以搭建真正的 AI 科学平台。Fedus 于 2025 年 3 月 17 日宣布离开 OpenAI,引发 VC 反向路演狂潮;Felicis 在公司尚未注册时就已承诺投资。到 2025 年 9 月 30 日隐身发布时,团队已扩大到 20 多名顶尖研究者,在旧金山建立了初始实验室,早期实验数据和仿真已经跑起来,尽管机器人系统仍在训练。公司首次披露的商业项目来自一家未具名半导体制造商,对方面临芯片散热挑战;Periodic 用该制造商的实验数据训练定制 AI 智能体,帮助工程师更快迭代。Observer 还报道,客户基础包括航天和国防行业公司。2026 年 3 月,Bloomberg 报道 Periodic Labs 正以约 $7B 估值洽谈交易;到 2026 年 5 月,Forbes 报道本轮已进入深入阶段,并以 $7.5B 估值显著超额认购。公司入选 2026 年 Forbes AI 50 Brink 榜单。截至 2026 年 6 月报告日期,公司尚未公开宣布完成任何超导体发现;这符合材料研究通常较长的开发周期,也符合公司先搭建数据飞轮的表述。包括 Meta 的 Yann LeCun 在内的科学批评者认为,当前 AI 的模式匹配能力不足以支撑自主科学发现所需的真正假设形成;2025 年一项 Nature 研究也发现,AI 工具虽然放大个体研究者产出,却可能缩窄科学探索的多样性。这些担忧构成核心论点的实质风险,必须靠已展示的实验结果来消解。[CO030, CO033, CO034, CO035, CO036, CO037]
| 日期 | 事件 | 类型 | 金额 / 估值 / 状态 | 参与方 | 含义 |
|---|---|---|---|---|---|
| 2025-02(约) | Fedus-Cubuk 在 San Francisco 的创始对话 | 创立 | N/A | Liam Fedus 与 Ekin Dogus Cubuk | AI 科学家投资逻辑的概念起点;两人都看到机器人自动化、仿真和 LLM 推理正在汇合 |
| 2025-03-17 | Liam Fedus 宣布离开 OpenAI | 创立 | N/A | Liam Fedus;OpenAI | 公开释放新的 AI 科学初创公司信号;触发 VC 反向推介热潮 |
| 2025-03(约) | Periodic Labs 注册成立;Felicis 承诺首张支票 | 融资 | 未披露种子轮分批金额 | Felicis (Peter Deng) 与 Liam Fedus、Ekin Dogus Cubuk | 公司正式存在;在注册文件完成前就开始融资 |
| 2025-Q2/Q3 | 团队组建:20+ 名来自 Meta、OpenAI、DeepMind 的研究人员加入 | 扩张 | N/A | Alexandre Passos;Eric Toberer;Matt Horton;20+ 其他成员 | 世界级创始团队组建完成;多人放弃大量未归属股权 |
| 2025-09-30 | Periodic Labs 走出隐身;宣布 $300M 种子轮 | 融资 | $300M,投后估值 $1.3B | a16z、Felicis、DST、NVentures、Accel、Bezos、Schmidt、Dean 与 Gil | 公布时为 VC 史上披露规模最大的种子轮;立即获得全球媒体报道 |
| 2025-10(约) | San Francisco 实验室建立;实验工作开始 | 产品 | N/A | 内部团队 | 物理自动化实验室投入运行;实验数据和仿真在跑;机器人系统处于训练中 |
| 2026-03-25 | Bloomberg 报道约 $7B 估值的交易谈判 | 融资 | 约 $7B 估值(报道) | Bloomberg 来源;Periodic Labs 投资者 | 重大后续轮的首个公开信号;验证快速价值增值投资逻辑 |
| 2026-05-07 | Forbes 确认 $500M 融资,估值 $7.5B;显著超额认购 | 融资 | $500M,估值 $7.5B(深入谈判) | AMP (Anjney Midha);未具名共同投资者 | 估值在不到 9 个月内较种子轮提升约 5.8x;市场讨论快速跟投轮 |
| 2026(持续) | 登上 Forbes AI 50 Brink 榜单;半导体和国防客户获确认 | 扩张 | N/A | 半导体和国防 / 航天客户(名称未披露) | 早期商业牵引力证明价值;AI 科学赛道品牌认知提升 |
标注(约)的日期依据报道语境估计(例如 TechCrunch 2025 年 10 月文章中的「7 个月前」)。后续融资事件据报道处于深入谈判阶段;截至运行日期尚未确认关闭。内部产品或研发里程碑未公开。
[CO001, CO002, CO003, CO023, CO025, CO027]按时间梳理从 2025 年 2 月创始对话到 2026 年中后续融资之间的创立、融资、运营和商业里程碑。
[CO002, CO003, CO006, CO021, CO023, CO024]1.5 图表与证据
02市场分析
2.1 市场边界与竞争性支出
AI 驱动的材料发现,是更广泛实验室科学技术栈中的一个独立市场分支。核心产品类别包括软件平台和软硬一体系统,用生成式 AI、图神经网络和大语言模型提出、仿真并通过实验验证新的材料组成——把过去按十年计的发现周期压缩到数月或数年。它不同于一般实验室自动化(仪器控制、液体处理、LIMS):主要价值主张不是吞吐量管理,而是假设生成和实验设计。 纳入的支出包括 AI 平台授权、自主实验室 SaaS 订阅、AI 驱动的计算筛选工具,以及专为闭环发现设计的机器人系统。排除的支出包括通用分析仪器、临床试验自动化、标准 CRO/CDO 外包,以及主要用于数据管理而非发现的 AI 工具。Periodic Labs 的上游机会还包括若干相邻支出池:$8.83B 的整体实验室自动化市场、$300B 的全球制药 R&D 预算,以及规模可比的化学品和先进材料 R&D 板块。现状替代方案是手工组合合成、传统计算化学和合同 CRO 服务——这些方案相较 AI 指挥的闭环系统,都明显更慢,也更吃资本。[CM001, CM002, CM003, CM004, CM005]
| 细分市场 / 类别 | 纳入支出 | 排除支出 | 主要买方 / 付款方 | 与 Periodic Labs 的相关性 |
|---|---|---|---|---|
| AI 材料发现软件 | 平台许可、SaaS 发现订阅、AI 筛选工具 | 通用 LIMS、ELN、实验室管理 SaaS | 研发总监 / 材料科学 VP | 核心可服务市场($970M,2026) |
| 自动化化学实验室系统 | 用于闭环合成和测试的集成机器人 + AI | 不具备 AI 假设生成能力的标准液体处理机器人 | CTO / 研发 VP(制药、电池、化工) | 硬件赋能层;Periodic Labs 平台可运行其上($5.75B,2026) |
| 实验室自动化中的 AI(更宽口径) | 药物发现 AI、基因组自动化、材料 AI、临床实验室 AI | 非 AI 实验室设备;手工流程管理 | 跨行业研发 VP、实验室负责人 | 竞争语境的上位基准($4.19B,2026) |
| 实验室自动化总市场 | 实验室自动化的所有硬件和软件,包括非 AI | 研发人员成本、耗材、设施成本 | CFO / 采购、实验室运营 | 上游预算池和规模锚点($8.83B,2026) |
| 制药研发服务(相邻) | 面向药物和材料筛选的 CRO/CDO 合同;AI 赋能发现服务 | 监管申报、临床运营、制造 | 研发 VP / 外部创新负责人 | 既有替代者 + 潜在渠道伙伴(总研发约 $300B) |
| 特种化学品 / 先进材料研发 | 新配方、涂层、聚合物、陶瓷的内部实验室支出 | 大宗化学品生产、标准 QC 测试 | 研发总监、工艺工程师 | 潜在需求:研发预算高,但今天 AI 准备度低 |
市场规模为分析师报告中的 2026 年估计(Business Research Company、DimensionMarketResearch、TowardsHealthcare);AI 材料发现软件细分市场是 Periodic Labs 的主要 TAM。排除支出类别反映定义边界,而不是竞争威胁。
[CM001, CM002, CM003, CM005]2.2 TAM、SAM 与 SOM:多视角测算
测算 AI 驱动材料发现,至少需要三层嵌套视角,因为没有单一分析师报告覆盖从自主实验室硬件到 SaaS 发现层的全栈。最窄、也最直接相关的分支——专用 AI 材料发现平台——2025 年价值约 $740M,按 The Business Research Company 的 AI in Materials Discovery Global Market Report 2026,预计 2026 年达到 $970M,并以 30.3% CAGR 增长至 2030 年 $2.77B。 再上一层,实验室自动化中的 AI 市场(包括药物发现 AI、基因组学自动化和材料科学)2025 年为 $3.54B,预计 2026 年达到 $4.19B,CAGR 为 18.4%。最广的可比市场——覆盖硬件、软件和 AI 的整体实验室自动化市场——2025 年为 $8.03B,2026 年为 $8.83B。自主化学实验室市场把硬件机器人与 AI 控制层结合起来,2026 年估计为 $5.75B,并以 14.5% CAGR 增长至 2035 年。 面向自主 AI 科学家平台模型(Periodic Labs 的路径)的 SOM 还无法用公开数据验证;行业分析师尚未单独拆出这个子分支。一个保守代理——AI 材料发现软件分支的 12%——意味着 2026 年 SOM 约 $116M,但该估计高度不确定,且应标注为证据缺口。所有市场规模数字都带有分析师方法论风险,应视为数量级锚点,而不是精确预测。[CM006, CM007, CM008, CM009, CM010]
| 发布方 | 年份 | 地域 | 数值(USD) | CAGR | 方法 | 置信度 | 关键限制 |
|---|---|---|---|---|---|---|---|
| The Business Research Company | 2026 | 全球 | $970M | 30.3% | 自上而下的供应商调研 + 需求建模 | 中 | 付费墙;方法不透明;AI 材料发现定义可能与通用实验室 AI 重叠 |
| DimensionMarketResearch | 2026 | 全球 | $5.75B(自动化化学实验室) | 14.5% | 自上而下、多细分聚合 | 中 | 宽于纯 AI 软件;包含机器人硬件 |
| TowardsHealthcare | 2026 | 全球 | $4.19B(实验室自动化中的 AI) | 18.4% | 自上而下细分模型 | 中 | 偏制药;未单独拆出材料科学份额 |
| The Business Research Company(更宽口径) | 2026 | 全球 | $8.83B(实验室自动化总市场) | 9.9% | 聚合硬件 + 软件市场 | 中 | 付费墙;包含非 AI 硬件,抬高基数 |
| RealTimeDataStats | 2025 | 全球 | $1.8B(自动化实验室机器人) | 19.5% 至 2033 年 | 自下而上的机器人市场模型 | 低 | 仅硬件;未包含 AI 软件;发布方声誉低 |
| IQVIA Global Trends in R&D 2026 报告 | 2026 | 全球 | $300B(制药研发) | ~1.7% YoY | 生物制药公司一手调研 | 高 | 不特指 AI 材料发现;仅为制药研发总预算 |
| 估算 SOM(推导) | 2026 | 全球 | ~$116M(估计) | n/a | AI 材料发现细分市场的 12% 份额代理;没有公开一手来源 | 低 | 自动化 AI 科学家平台没有公开 SOM 数据;已标注证据缺口 |
所有第三方市场规模均来自分析师研究报告(部分有付费墙);CAGR 数字由发布方提供,可能反映乐观情景。SOM 行是一阶原理代理值,不是一手来源估计。之所以展示多个视角,是因为没有单一报告能用一致定义覆盖整个技术栈的所有层。
[CM006, CM007, CM008, CM009, CM010, CM011]从整体实验室自动化到 AI 材料发现软件细分市场,再到自主 AI 科学家平台的估计 SOM,逐层拆分 TAM/SAM/SOM。
所有数值均为分析师报告中的 2026 年估计。SOM(约 $116M)是第一性原理代理值(AI 材料发现 TAM 的 12%),没有一手来源佐证;应按数量级理解。
[CM006, CM007, CM008, CM040]2026 年四个关键市场规模层级的低 / 基准 / 高估计区间,单位均为百万美元,反映分析师不确定性和不同来源的定义差异。
低 / 高边界根据 CAGR 差异和多家发布方的分析师分歧估计;基准值对应主要引用的分析师估计。 所有数值均为 2026 年估计。
[CM006, CM007, CM008, CM005]2.3 买方分层与采用路径
可以识别出四类主要商业买方,它们的预算归属、采用触发点和准备度各不相同。制药和生物技术公司是当前 AI 实验室自动化的最大部署基础,驱动力来自董事会层面的 AI 药物发现要求,以及压缩 pre-IND 材料筛选周期的可能性。电池和储能公司面对 EV 与电网储能市场的竞争压力,近期准备度最高;SandboxAQ 的 AQVolt26 固态电池材料项目就是证据之一。半导体制造商正在成为增长分支,因为 AI 平台现在已经能够为下一代芯片供应链提出新的镓合金和二维材料。 特种化学品公司的即时准备度较低——数据资产碎片化,内部 AI 治理框架也不成熟——但随着 ROI 证据累积,它们代表着一个大型潜在市场。政府机构和学术研究联盟构成第二层买方,尤其是与国家竞争力任务绑定的超导体和先进半导体项目。亚太是增长最快的地区,受中国 16%+ 年度制药 R&D 增长和国家支持的先进材料投资计划驱动。北美仍拥有最大的绝对市场份额,并承载该行业多数私营 VC 投资。[CM014, CM015, CM016, CM017, CM018]
| 细分市场 | 买方 | 用户 | 付款方 | 工作流适配 | 预算负责人 | 主要采用触发因素 |
|---|---|---|---|---|---|---|
| 制药 / 生物技术 | 研发 VP / 发现负责人 | 材料科学家、药物化学家 | CFO / 研发 VP | Pre-IND 材料筛选、辅料发现、药物递送支架 | 研发部门 | AI 药物发现指令;竞争时间线压力 |
| 电池 / 储能 | CTO / 工程 VP | 材料工程师、电化学家 | CTO / 工程 VP | 固态电解质设计、电极筛选、循环寿命优化 | 技术部门 | EV 和电网储能竞争紧迫性;政府资助任务 |
| 半导体 / 电子 | 技术 VP / 首席材料科学家 | 工艺工程师、器件物理学家 | 中央研发 / 技术 VP | 用于逻辑 / 存储节点的新型半导体合金、2D 材料 | 中央 R&D | 下一代芯片密度要求;供应链多元化 |
| 特种化学品 | R&D 总监 | R&D 化学家、配方科学家 | R&D 总监 | 聚合物与涂层配方、催化剂发现 | R&D 部门 | 竞争差异化;压缩 R&D 周期成本 |
| 政府 / 学术 | 项目经理 / 首席研究员 | 研究科学家、博士后研究员 | 机构预算官 / 联盟主任 | 国家竞争力项目、超导体与先进半导体研究 | 联邦机构或联盟 | 国家竞争力任务;受资助的发现项目 |
买方 / 用户 / 付款方角色来自行业分析(Pangaea Ventures、Cypris.ai、Royal Society)和公开案例证据 (SandboxAQ)。预算归属为概括口径;实际采购还可能需要 IT 和采购部门签批。就绪度随公司规模和 AI 成熟度而变。
[CM014, CM015, CM016, CM018]2026 年 AI 材料发现平台五类主要买方细分的采纳准备度、预算归属和付款方特征。
[CM014, CM018]2.4 增长驱动与采用约束
最主要的增长驱动,是 AI 已展示出把材料发现到商业化周期从数十年压缩到约一到两年的能力;任何面临竞争时间压力的大型 R&D 组织都会因此被吸引。生成式 AI 能力提升——尤其是可预测晶体结构稳定性的图神经网络,以及 DeepMind 的 GNoME 这类大模型(识别出 220 万种稳定材料)——已经把叙事从猜想推向验证。美国(CHIPS Act、DOE 材料计划)、欧盟(电池 AI 联盟、MPIE 牵头、横跨 12 个国家的 33 方项目)和中国先进材料规划中的国家竞争力项目,形成政策顺风,补充了私营部门需求。 与这些驱动相对,四个结构性采用约束分量很重。第一,数据质量和自研数据所有权是企业端首要瓶颈:实验数据集被孤岛化,往往缺乏标准化,还受复杂 IP 协议约束。第二,截至 2026 年,只有 29% 的企业报告 AI 带来显著 ROI,意味着发现平台必须先给出具体、可衡量的结果,企业 R&D 预算才会大规模投入。第三,AI 治理不成熟普遍存在——按 Deloitte,2026 年只有约 25% 的组织拥有成熟 AI 治理框架——这会拉长销售周期并增加合规负担,尤其在 AI 监管已覆盖 68 个国家的背景下。第四,AI 人才缺口(到 2026 年全球约 350 万个岗位空缺)限制了潜在客户部署和维护复杂发现系统的内部能力。ComputeForecast 识别出的第五个结构性约束是,企业 AI 采用基础设施搭建速度慢于技术本身,形成反复出现的部署滞后,并持续跑输市场乐观预期。[CM019, CM020, CM021, CM023, CM025, CM026]
| 驱动因素 / 约束 | 方向 | 时间 | 对采用的影响 | 尽调问题 |
|---|---|---|---|---|
| AI 时间压缩(从几十年到几个月) | 驱动因素 | 当前 | 任何面临竞争进度压力的 R&D 组织,商业理由都会更有说服力 | 向客户推荐人验证:买方是否已基于这一主张签约? |
| 生成式 AI 能力提升(GNoME、MatterGen、自主实验室) | 驱动因素 | 当前–2027 | 平台能力不再只是猜想;规模化验证降低技术风险 | Periodic Labs 哪些自研模型能力能与开放模型拉开差距? |
| 国家竞争力任务(CHIPS Act、EU 电池联盟、中国先进材料) | 驱动因素 | 当前–2028 | 政府和学术渠道补充商业需求;形成非商业收入底座 | 量化资助管线和政府合同积压 |
| 数据质量与 IP 碎片化 | 约束 | 持续 | 企业销售周期被拉长;客户可能不愿把自有数据交给第三方平台 | Periodic Labs 如何处理数据驻留、IP 所有权和模型污染? |
| 企业 AI 治理不成熟(约 25% 拥有成熟框架) | 约束 | 持续至 2028 | 采购审批变慢;尤其在受监管行业,合规开销增加 | 平台具备哪些合规认证(SOC 2、GxP、ISO 27001)? |
| 企业 AI 整体 ROI 兑现率为 29% | 约束 | 近期(2026–2028) | 大额合同前,供应商必须拿出可验证的 ROI 指标;销售动作会更偏试点 | 现有客户从试点走到生产合同平均需要多久? |
| AI 人才缺口(2026 年全球 3.5M 个岗位空缺) | 约束 | 2026–2030 | 客户部署和维护 AI 发现系统的能力受限;对供应商托管服务的依赖上升 | Periodic Labs 是否提供托管服务或实验室运营商模式? |
| 集成式自主实验室的资本强度 | 约束 | 持续 | 前期成本高,限制可触达客户总数;更利好大型药企和资金充足的电池初创公司 | 最小可行部署成本是多少,有哪些融资选项? |
驱动因素和约束综合自多家分析机构与一手来源(Deloitte、Forbes、ComputeForecast、IBM、Royal Society)。时间判断为指示性口径,可能随监管变化或技术拐点而移动。尽调问题为建议项,面向后续尽调渠道, 不是公开数据。
[CM019, CM021, CM023, CM025, CM026, CM028]2.5 垂直商业化路径
AI 材料发现平台近期商业化路径集中在四个垂直领域,它们在准备度、买方紧迫性和收入模式上不同。电池与储能准备度最高:全球电池储能市场围绕固态电解质和新电极化学的商业紧迫性,正好对应 AI 平台生成并筛选新材料配方的能力。SandboxAQ 的 AQVolt26 计划和欧盟 33 方 AI 电池联盟,都验证了这一应用路径的商业成熟度。半导体紧随其后:AI 平台现在能够提出符合芯片属性要求的新型镓合金和 2D 材料,直接接入先进逻辑和存储供应链。 高温超导体是 Periodic Labs 明确表述的主要目标,也可以说是上行空间最高的垂直领域——其在量子计算和电网基础设施中的商业应用会带来转型级经济价值——但科学风险和时间线风险也最高,因为从预测结构走到规模化验证合成,室温超导体领域从未跑通。制药相邻发现工具(赋形剂、新型药物递送支架、生物相容材料)构成 $300B 制药 R&D 预算中的一个大型增长子分支;IQVIA 的 Global Trends in R&D 2026 报告确认,AI 在 pre-IND 材料筛选中的作用正在扩大。面向这些垂直领域的 SaaS 模式平台可实现 70–90% 毛利率,显著高于传统 CRO 服务,使商业模式在规模化后具备较强防御性。[CM032, CM033, CM034, CM035, CM036, CM037]
| 垂直领域 | 目标材料类别 | 商业机会 | 关键买方 | 采用阶段(2026) | 主要收入模式 |
|---|---|---|---|---|---|
| 电池 / 储能 | 固态电解质、电极材料、隔膜 | EV 和电网储能市场;下游价值链达数千亿美元 | 电池 OEM、储能开发商、汽车制造商 | 早期商业试点正转向生产 | 平台许可 + 数据合作 + IP 授权 |
| 半导体 / 电子 | 新型镓合金、2D 材料、介电材料、低 k 绝缘体 | 先进逻辑和存储节点;光电子;柔性电子 | 领先芯片晶圆厂、先进材料供应商 | 面向精选客户的试点 / 概念验证 | SaaS 发现订阅 + IP 授权 |
| 高温超导体 | 室温或近环境温度超导化合物 | 量子计算、电网基础设施、医学影像 | 国家实验室、量子硬件公司、电网运营商 | 研究 / 早期试点;尚无已验证商业部署 | 政府资助 + 战略 R&D 合作 + IP 授权 |
| 药物相邻发现 | 药物递送支架、辅料、生物相容涂层 | Pre-IND 材料筛选;上游药企 R&D 预算 $300B | 大型药企 R&D 团队、特种生物技术公司、CMO | 需求增长;已有部分 AI 工具部署,但全栈自主实验室并不常见 | SaaS 订阅 + 类 CRO 服务费 |
| 特种化学品 | 催化剂、涂层、聚合物、胶黏剂、功能材料 | 配方效率;在近似大宗品市场中做出竞争差异 | 特种化学品公司、消费品 R&D | 就绪度低;多数仍在评估阶段 | 试点合同;治理成熟后转 SaaS |
采用阶段评估综合自行业分析(Cypris.ai、Pangaea Ventures、ChemDive、Royal Society)和公开项目证据 (SandboxAQ AQVolt26、MPIE EU 电池项目)。除非已引用,商业机会评估为定性判断;收入模式基于可比 AI 平台先例作指示性判断,不代表 Periodic Labs 披露定价。
[CM032, CM033, CM034, CM035, CM036]自主 AI 材料发现价值链,从研究假设出发,经过机器人合成和表征,走向 IP 生成和商业部署。
[CM019, CM040]2.6 图表与证据
03竞争格局
3.1 直接同业:Neolab 与自主发现初创公司
Periodic Labs 最接近的直接竞争对手,是那些把 AI 基础模型与物理自主实验室基础设施结合起来、规模化生成新实验数据的公司。CuspAI(英国 Cambridge,成立于 2024 年)是最清晰的可比对象:它的平台充当「材料世界的搜索引擎」,为目标属性生成具备合成可行性的化学组成。CuspAI 在种子轮和 Series A 中累计融资 $130M(2025 年 9 月,由 NEA 和 Temasek 共同领投),投后估值约 $520M;随后它与 Meta、Kemira 和 Hyundai Motor Group 洽谈商业合同,把非正式估值推至约 $800M,并在 2026 年初进入 $200M+ 融资洽谈,目标为超过 $1B 的独角兽级估值。Tracxn 上 CuspAI 列出 73 个活跃竞争对手;其创始人 Prof. Max Welling(前 Microsoft Research Distinguished Scientist)和 Dr. Chad Edwards(前 Quantinuum)也给它带来强学术与深科技可信度,可与 Periodic 的 OpenAI/DeepMind 履历相提并论。 第二个直接同业是 FutureHouse,一家旧金山非营利机构,由 Eric Schmidt 支持(他也是 Periodic 投资人),明确瞄准自主 AI 科学家目标。TechCrunch 将 FutureHouse 与 Periodic、Tetsuwan Scientific 并列为同一使命空间的公司。Tetsuwan Scientific 是一家小型初创公司,也瞄准自主实验室科学工作流。材料科学之外,Isomorphic Labs(Google DeepMind 拆分公司)于 2025 年初融资 $600M,估值未公开披露,目标是带物理实验室集成的 AI 驱动药物设计;虽然其重点是制药而非材料,但团队和技术路径与 Periodic Labs 有实质重叠。 Periodic Labs 相对 CuspAI 和其他 AI 信息学同业宣称的关键差异在于:(1)全闭环自主实验室,生成物理实验数据,而不只是计算预测;(2)每次实验形成的自研数据飞轮;(3)范围更广——追求通用 AI 科学家,而不是特定领域搜索引擎。不过,Periodic 尚未发布验证其发现产出优于同业的基准;CuspAI 具备合成感知的生成模型,以及 DeepMind 的 GNoME(联合创始人 Cubuk 曾参与),都已经展示了强预测到验证管线,而且不需要完整物理实验室集成。 [CP004, CP005, CP006, CP007, CP008, CP019]
| 竞争对手 | 类别 | 规模 / 融资 | 目标客群 | 差异化 | 关键限制 |
|---|---|---|---|---|---|
| CuspAI | 直接同业(AI 材料搜索引擎) | 已融资 $130M;据称估值谈判约 $800M–$1B+(2026) | 企业 R&D:半导体、电池、水处理 | 具备合成感知的生成式 AI;自有材料数据集;合作伙伴包括 Meta、Kemira、Hyundai | 没有自主物理实验室;计算预测材料仍需另行合成 |
| FutureHouse | 直接同业(自主 AI 科学家、非营利) | 非营利;融资未公开披露 | 学术和独立研究群体 | 非营利开放科学使命;创始人 / 投资人网络与 Periodic 重叠 | 非营利模式限制商业规模和自有数据积累 |
| Isomorphic Labs | 相邻领域(带实验室集成的 AI 药物设计) | 已融资 $600M(2025);估值未披露 | 药物发现 | Google DeepMind 分拆背景;深度物理模型;临床管线 | 聚焦药企,而非材料科学;不是直接材料竞争对手 |
| Schrödinger (SDGR,上市公司) | 既有厂商(计算化学 + AI 平台) | 上市公司(NASDAQ);2026 Q1 收入 $58.6M;现金 $406M | 药物发现、材料科学、先进制造 | 30+ 年仿真积累;Bunsen AI 共同科学家计划 2026 年夏季上线;历史融资 $162M | SaaS 转型造成近期收入压力;物理实验室自动化有限 |
| Citrine Informatics | 既有厂商(材料信息学 SaaS) | 约 $81.3M 融资(2025 Series C);收入估计 $10–$100M | 特种化学品、涂层、电池、聚合物(企业) | 能处理小规模 / 稀疏工业数据集;客户包括 LyondellBasell、Panasonic、Michelin | 没有自主实验室组件;依赖客户提供数据;仅限仿真 |
| Microsoft Azure Quantum Elements | 既有厂商(超大规模云厂商材料 AI 平台) | Microsoft(NASDAQ MSFT,市值约 $3T) | 跨行业企业云客户 | 超大规模云厂商分发;云集成;Quantinuum 逻辑量子比特里程碑 | 平台广度大于深度;不是材料专精厂商;可能替代较小供应商 |
| Google DeepMind (GNoME) | 既有厂商(研究级 AI 材料发现) | Alphabet 子公司;研究预算未单独披露 | 学术和工业研究群体 | GNoME:预测 2.2M 个晶体、380K 个稳定结构;736 个已独立合成;开放访问 | 开放访问研究工具,不是商业产品;没有企业定价或 SLA |
| Emerald Cloud Lab | 相邻领域(云实验室即服务) | 私营;融资未披露;定价 >$250K/年 | 学术、生物技术、药企 | 200+ 台仪器;24/7 远程访问;可复现实验室即服务 | 聚焦生命科学(非材料);高成本限制学术采用;没有 AI 科学家层 |
| Recursion Pharmaceuticals | 相邻领域(AI + 物理实验室,药物发现) | 上市公司(NASDAQ RXRX);2026 Q1 现金 $665M;已融资 $452M;Q1 收入 $6.47M | 药物发现(肿瘤、罕见病) | 50PB 自有数据;端到端 AI 发现平台;Sanofi / Roche 合作 | 仅聚焦药企;年化净亏损约 $560M;股价较峰值下跌 >90% |
| Atinary | 相邻领域(自动驾驶实验室平台) | 私营;融资未披露 | 化学、材料科学、药企 R&D | 2026 年在波士顿 SLAS 开设首个物理自动驾驶实验室;跨领域优化 | 早期阶段;与 Periodic 的自主实验室雄心相比规模有限 |
| Automata (LINQ) | 相邻领域(实验室编排 OS) | $45M Series C(2026 年 1 月);Danaher 为战略投资人 | 药物发现、细胞生物学、生命科学 | API 优先;模块化 LINQ 平台;Danaher 产品组合集成 | 没有 AI 科学家愿景;只是编排层;不是材料优先平台 |
融资数据来自 Tracxn、TechCrunch、Forbes 和公司公告,截至 2026 年 6 月。私营公司收入数字来自公开来源估计 (如有披露)。Schrödinger 收入来自 2026 Q1 SEC 文件。Recursion 数字来自 2026 Q1 投资者发布。估值区间包括 已披露数字和据报道的交易谈判估值。
[CP001, CP004, CP006, CP007, CP009, CP012]Periodic Labs 位于高自主、高材料专属性象限;多数既有厂商集中在较低自主水平。CuspAI 是最接近的直接同业, 但缺少物理实验室基础设施。
坐标位置是有证据支撑的序数评分,不是数值测量。X 轴:自主水平(0=手动 / 脚本化, 1=完全自主闭环)。Y 轴:材料科学专属性(0=领域无关,1=材料优先)。位置来自截至 2026 年 6 月的公开产品描述、 SEC 文件和新闻来源。公开披露有限的公司不确定性较高。
[CP006, CP008, CP018, CP025, CP038]3.2 现有平台:计算化学与材料信息学
材料科学买方在考虑 Periodic Labs 前,通常会遇到三类现状替代方案:Schrödinger、Citrine Informatics 和 Microsoft Azure Quantum Elements。 Schrödinger(NASDAQ: SDGR,成立于 1990 年)是最成熟的基于物理的仿真和 AI 材料平台。其 2026 年 Q1 总收入为 $58.6M(ACV $28.4M,同比 +12%),不过随着公司从永久授权转向托管(订阅)授权,软件收入同比下降 21%。截至 2026 年 Q1,Schrödinger 持有 $406M 现金,并计划在 2026 年夏季推出「Bunsen」——一个智能体 AI 共同科学家,直接竞争 Periodic 的 AI 科学家愿景。Lilly 以 $2.3B 收购 Ajax Therapeutics(使用了 Schrödinger 发现的分子)验证了 Schrödinger 的历史记录,也让它在监管和制药信任上有强差异化。对材料科学客户,Schrödinger 把同一套物理优先核心用于分子仿真。 Citrine Informatics(Redwood City, CA)截至 2025 年 Series C 已累计融资约 $81.3M,客户包括 LyondellBasell、Eastman、Panasonic、Michelin 和 LANXESS。Citrine 的 DataManager 和 VirtualLab 工具面向特种化学品和先进材料公司,帮助它们从小而稀疏的数据集中提取价值。它的 SaaS 模式和企业材料 R&D 聚焦(而不是从零发现)让它成为仿真既有厂商的互补工具,也部分替代 Periodic 的数据生成价值主张。截至 2026 年,Citrine 收入估计在 $10–$100M 区间。 Microsoft Azure Quantum Elements 代表来自超大规模云厂商的平台风险。Microsoft 已进入 AI 驱动材料发现,可能冲击小型专业玩家。它的云规模和企业分发优势远超任何初创公司;已经在 Azure 生态内的材料科学家,可以以很低切换摩擦采用新的材料 AI 工具。同样,Google DeepMind 的 GNoME——预测了 220 万个新晶体结构(其中 380,000 个高度稳定),其第一作者正是 Periodic Labs 联合创始人 Ekin Cubuk——仍是开放且被广泛引用的研究工具,整个社区无需向 Periodic 付费即可在其基础上建设。 [CP009, CP010, CP011, CP012, CP013, CP018]
| 能力 | Periodic Labs | CuspAI | Schrödinger | Citrine Informatics | Microsoft Azure QE | Google DeepMind |
|---|---|---|---|---|---|---|
| 自主物理实验室 | 是(核心产品) | 否(仅计算) | 否 | 否 | 否 | 部分(Berkeley A-Lab 合作) |
| 用于材料组成的生成式 AI | 是 | 是(具备合成感知) | 是(Bunsen,2026 年夏季) | 是(VirtualLab) | 是(Azure QE) | 是(GNoME) |
| 基于物理的仿真(DFT/MD) | Unknown | 部分 | 是(核心强项) | 否 | 部分(量子) | 是(DFT 集成) |
| 自有实验数据集 | 是(增长中) | 部分 | 是(共同实验室项目) | 否(客户数据) | 否 | 是(内部;外部有限) |
| 企业 SaaS / API 产品 | 早期(半导体合作伙伴) | 是 | 是 | 是 | 是 | 否(研究工具) |
| 超导体 / 先进材料重点 | 是(已声明优先级) | 部分(广义材料) | 部分 | 部分(电池、聚合物) | 部分 | 是(GNoME 包含 52K 个层状化合物) |
| 临床 / 药物发现能力 | 否 | 否 | 是 | 否 | 部分 | 是(AlphaFold 谱系) |
| 开源 / 公共数据库访问 | 否 | 否 | 部分(部分工具) | 否 | 是(Materials Project 集成) | 是(已发布 380K 个 GNoME 结构) |
标为“未知”的单元格表示缺少公开证据;这不等于有证据证明不存在。能力评估基于官方产品页、SEC 文件和 已核实新闻来源,截至 2026 年 6 月。Schrödinger Bunsen AI 被描述为计划在 2026 年夏季上线;相关能力仅为上线前描述。
[CP006, CP008, CP011, CP013, CP018, CP020]Periodic Labs 是唯一把自主物理实验室与生成式 AI 材料设计规模化结合的竞争者;Schrödinger 和 Citrine 在企业 SaaS 深度上领先。
“未知”和“部分”单元格反映截至 2026 年 6 月公开披露有限。功能有无按已公开发布产品评估,不包括 beta 或路线图项目;Schrödinger Bunsen 例外,该产品已宣布 2026 年夏季发布,并有预发布描述。
[CP006, CP011, CP013, CP017, CP025, CP032]3.3 相邻自主实验室与云 CRO 平台
另一类竞争者提供实验室即服务基础设施层,而 Periodic Labs 目标是把这一层内化:云实验室、CRO 自动化平台和编排初创公司。 Emerald Cloud Lab(ECL,Austin, TX)运营一座完全由软件控制的设施,超过 200 种仪器型号可通过 ECL Command Center 全天候远程访问。ECL 面向生命科学和生物技术,而不是先进材料合成,但它的「云实验室」模式是 Periodic 自主实验室的直接概念先例。ECL 的定价——每年可能超过 $250,000——使其主要服务于资金充足的商业客户。Strateos 提供类似云实验室模式,用符号编程支持自主工作流,目标是制药和合成生物学应用。Arctoris 为药物发现 CRO 搭建自主机器人工作单元。 Recursion Pharmaceuticals(NASDAQ: RXRX)是端到端 AI 加物理实验室发现公司的公开市场先例。截至 2026 年 Q1,Recursion 股价约 $3.20–$3.50(从 $40 以上峰值下跌),Q1 收入为 $6.47M(低于预期),持有 $665M 现金。其 Recursion OS 平台摄入 50+ PB 自研生物数据。Recursion 的股价表现给 Periodic Labs 投资人敲响警钟:即便一家 AI 实验室公司资金充足、人员到位,也可能需要多年商业验证,市场才会给出溢价估值。Recursion 约 $560M 的年化净亏损和收入不及预期,凸显在临床 / 材料证据点出现之前,AI 实验室平台变现有多难。 在 SLAS 2026 上,15 家公司竞争成为 AI 赋能实验室的操作系统层——Biosero、Automata、Synthace、UniteLabs 等。Automata 于 2026 年 1 月完成 $45M Series C,战略投资方为 Danaher Ventures,直接把 Danaher 的仪器组合与 Automata 的 LINQ 编排平台连接起来。Ginkgo Bioworks 通过与 OpenAI 合作,于 2026 年 2 月发表结果,显示 GPT-5 自主执行 36,000 次蛋白质合成实验,将 sfGFP 生产成本降低约 40%。这个概念验证对 Periodic 的「AI 科学家」叙事形成压力:大规模 AI 实验室实验已经由既有玩家执行,而不只是资金雄厚的新进入者。Atinary 在 SLAS 2026 于 Boston 启动首个物理自驱实验室,直接进入材料与制药自主优化空间。 [CP014, CP015, CP016, CP017, CP025, CP026]
| 平台 | 模式 | 指示性价格 / 单位 | 包含能力 | 折扣 / 未知项 |
|---|---|---|---|---|
| Periodic Labs | 合同 / 合作(早期商业) | 未公开披露 | 自主实验室实验;AI 科学家智能体;定制数据生成;半导体 / R&D 合作 | 无公开标价;据 Forbes / TechFundingNews 称已产生收入,但条款未披露 |
| CuspAI | 企业 SaaS + 定制材料服务 | 未公开披露 | 材料属性搜索;生成式组成建议;具备合成感知的输出;合作伙伴集成(Meta、Kemira、Hyundai) | 合同条款未披露;双重收入模式(平台 + 内部 IP 开发) |
| Schrödinger | 托管订阅(按 ACV) | 2026 Q1 ACV $28.4M;过去 4 个季度 ACV 约 $201M | 基于物理的仿真;FEP+;药物 / 材料设计工作流;Bunsen AI(2026 Q3) | ACV 同比增长 12%;因永久许可转 SaaS,软件收入同比 -21% |
| Citrine Informatics | SaaS 平台订阅 | 总 ARR 估计 $10–$100M;按席位 / 按项目分层 | DataManager;VirtualLab;生成式 AI 实验设计;最多 1000x 虚拟实验;数字助手 | 多年期企业合同;入门服务;价格未公开列示 |
| Emerald Cloud Lab | 按量付费和年度订阅 | 全面访问 >$250,000/年 | 200+ 种仪器类型;24/7 访问;ECL Command Center;可复现性保证 | 学术套餐价格更低;可提供企业 / 大学设施选项 |
| Microsoft Azure Quantum Elements | Azure 云消耗 + 订阅 | Azure 定价层级(按用量) | 量子仿真;AI 化学工作流;Quantinuum 硬件访问;云集成 | 并入更广泛的 Azure 企业协议;定价高度可变 |
| Recursion OS | 内部平台(不对外授权) | 不可购买 | 50PB 自有数据集;AI 靶点识别;药物设计;临床管线管理 | N/A – 内部平台;药企合作里程碑包括 Sanofi 和 Roche 的 $500M+ |
该市场多数价格不公开列示;表中数值为公开披露合同价值、SEC 文件中的 ACV,或分析师估计。“未公开披露”表示截至 2026 年 6 月未找到经验证的公开标价。Periodic Labs 定价完全未披露;公司已确认来自半导体客户的收入,但未披露金额或条款。
[CP002, CP009, CP015, CP016, CP022, CP025]3.4 护城河耐久性、商品化风险与反向证据
Periodic Labs 的核心护城河主张建立在自研实验数据飞轮上:每次自主实验室运行都生成高质量、源自实验室的数据,用来训练更好的 AI 模型,并相对依赖互联网抓取或公开材料数据库的竞争对手拉开差距。这个概念很有吸引力,但在商业规模上尚未验证。 2025 年材料科学自驱实验室子分支约为 $0.12B,并以约 40% CAGR 增长至 2030 年 $0.65B;但跨所有垂直领域的「自主闭环发现」被评估为 TRL 6(试点规模),量产成熟预计在 2028–2030 年。按独立市场分析,物理实验室 AI 的「ChatGPT 时刻」是 2028–2030 年事件,并非近在眼前。在此期间,实验室自动化的价值创造几乎完全迁移到软件 / AI 编排层,Automata、Benchling 和 Strateos 等公司已经在激烈竞争。 商品化风险真实存在:液体处理和机器人工作单元自动化现在已处于 TRL 8–9,实际上已经商品化;可防御位置在 AI 编排和数据生成平台。正如「Lab OS wars」所示,这一层竞争异常激烈。Schrödinger 的 Bunsen AI 发布(2026 年夏季)意味着最成熟的计算平台正在获得智能体 AI 能力。Microsoft Azure Quantum Elements 带来超大规模云厂商分发。Google 持续以零边际成本向用户发布 GNoME 级研究工具,使基准预测能力免费可得。 反向证据:VIA News 的金融分析师指出 Periodic Labs 的 $300M 烧钱风险令人担忧,并提到超导体突破没有清晰商业时间线。Sapio Sciences 在 SLAS 2026 的一项调查发现,45% 的科学家在使用未经授权的影子 AI 工具,因为官方平台没能满足他们——这说明需求端确实有拉力,但也说明采购漏斗不成熟、组织信任度低。行业分析师指出,多数「自主实验室」实际处于 Level 2–3 自主(针对特定任务的脚本化闭环),而不是 Periodic Labs 宣传的通用科学自主。Periodic Labs 正从半导体行业客户获得早期商业收入,但没有披露收入数字、达到产品市场契合的时间线,或实验吞吐量基准。 [CP027, CP028, CP029, CP030, CP031, CP033]
| 护城河主张 | 威胁 | 严重性 | 缓解措施 / 尽调问题 |
|---|---|---|---|
| 自有自主实验室实验数据飞轮 | CuspAI 等公司也在积累自有材料数据集;开放访问的 GNoME 结构会部分商品化基线预测 | 高 | 核验 Periodic 的实验数据生成速度、数据集独特性,以及自有数据是否在发现产出上显著跑赢公共基线 |
| 精英创始团队人才护城河(Fedus/Cubuk 来自 OpenAI/DeepMind) | 大型 AI 实验室(Google、OpenAI、Microsoft)能招到同等人才,并转向材料 AI;FutureHouse 和 Isomorphic 也从同一人才池招人 | 中 | 跟踪创始团队构成能否转化为可防御 IP,还是只是短期品牌优势;评估 20+ 人团队的股权留存和人员留任风险 |
| 半导体行业早期商业牵引 | Schrödinger、Citrine 和 Microsoft 已经拥有成熟的半导体行业关系和分发 | 高 | 识别合同范围、排他性和续约管线;确认半导体收入是否覆盖运营成本,还是仍处于商业化前阶段 |
| AI 模型 + 机器人 + 数据的垂直整合 | 实验室 OS 编排初创公司(Automata、Synthace、UniteLabs)提供模块化替代方案,把 AI 与硬件解耦; Schrödinger Bunsen AI 可能无需投资物理实验室也能跑出智能体科学 | 高 | 观察企业客户更偏好 Periodic 的封闭栈,还是混搭式编排;判断纵向整合到底是护城河,还是资本开支负担 |
| 资本优势($300M 种子轮;Series A 洽谈超过 $500M) | Schrödinger 持有 $406M 现金,Recursion 持有 $665M;CuspAI 已有 $130M+,且在洽谈独角兽轮融资;在这个量级,资本本身不是护城河 | 中 | 对照 Recursion 年亏损 $560M 的先例评估烧钱速度;要求披露现金跑道和收入爬坡时间表 |
| 材料规模上缺少同等自主实验室同行 | 正在出现:Atinary 于 2026 年推出首个 SDL;Chemify/Chemifarm 提供 chemistry-as-code 网络;A-Lab(Berkeley)验证了全自主材料合成 | 中 | 确认 Berkeley A-Lab、Chemify 或资金充足的学术联盟,能否在没有创业公司成本结构的情况下复制 Periodic 的实验室能力 |
| 学术资助计划带来的网络效应 | FutureHouse 和学术云实验室(Emerald、CMU)以更低学术端成本提供类似的开放科学参与机制 | 低 | 评估学术资助计划能否转化为商业客户,还是主要服务于公关 |
严重程度评级基于公开证据作定性判断。「高」表示已有证据显示威胁在近期竞争中活跃。「中」表示威胁可信但尚未实质落地。尽调问题代表仍未解决、需要私营公司数据才能回答的问题。
[CP001, CP003, CP004, CP033, CP036, CP037]Periodic Labs 在实验室自主性和数据生成差异化上得分最高,但已验证商业规模最低;既有厂商在信任、分销和收入上占优。
市场规模和 CAGR 数字来自分析师报告估计(RobotToday、BusinessWire/ResearchAndMarkets)。 Periodic Labs 估值来自据报道的交易谈判,而非已交割轮次。CuspAI 估值是融资前非正式数字。影子 AI 采纳来自 Sapio Sciences 在 SLAS 2026 对 150 名科学家的调查——样本量限制了可推广性。
[CP005, CP012, CP021, CP027, CP028, CP030]3.5 图表与证据
04财务情况
4.1 收入模式与商业牵引
Periodic Labs 的主要收入模式是合同制 AI 科学服务,常被称为「AI-lab-as-a-service」;公司与半导体、航天和国防行业的企业客户合作,用自主机器人实验室和 AI 科学家平台加速材料 R&D。收入确认更偏定制化和里程碑驱动,而非订阅制,因此天然存在收入波动和客户集中风险。2025 年 9 月公司发布时,Liam Fedus 公开披露了一个具体用例:一家半导体制造商面临芯片散热问题,Periodic 正为其训练定制 AI 智能体,解读实验数据并更快迭代。公司官网也确认与航天和国防客户有合作。 定价未公开披露,可能围绕合同价值、研究范围、时间线,以及结果材料 IP 的潜在共同所有权来设计。来源称,高影响力项目的合同可能价值数千万美元。三条次级收入流有可能成立,但尚未确认:(1)如果公司发现具备商业可行性的化合物,可授权材料 IP;(2)向外部 AI 系统授权自研实验数据;(3)直接商业化突破性发现材料。截至 2026 年 6 月,收入数字、客户数和 ARR 均未公开。尽管公司声称已有商业牵引,没有任何分析师或投资人评论引用了具体收入指标。TechFundingNews 提到「不同于许多同业,公司正在产生收入」,这让 Periodic Labs 区别于收入前可比公司,但仍不是财务披露。 GTM 效率代理指标(CAC、回本周期、NRR、客户数)完全未披露。Go-to-market 动作看起来是直销企业模式,目标是大型工业公司的 R&D 负责人和首席科学家,价值主张集中在压缩多年材料发现周期。国防和半导体行业客户关系通常需要深度技术参与,这意味着销售周期长,相比软件 GTM 可扩展性有限。[CI011, CI012, CI013, CI014, CI015, CI016]
| 来源 | 机制 | 单位 / 合同结构 | 当前状态 | 收入质量 | 尽调问题 |
|---|---|---|---|---|---|
| 合约式 AI 科学服务 | 客户为定义明确的 R&D 问题支付自主实验室实验和 AI 科学家产出 | 定制化;可能为里程碑 + 预付款;估计单个项目数千万美元 | 已开展;已确认半导体、航天、国防客户 | 低(早期、波动大、规模未披露) | 收入金额、合同数量、客户集中度、续约率 |
| 面向半导体的 AI 智能体 | 定制 AI 智能体,训练目标是为芯片 R&D 团队解读实验数据 | 嵌入合同条款或单独授权;公开未定价 | 已开展;至少披露一名半导体客户 | 低(仅披露单个项目;规模未知) | 定价、使用指标、续约条款、共同 IP 条款 |
| 材料 IP 授权 | 将自主实验室发现的突破性材料授权给制造商 | 按单位收取版税或一次性收费;尚未部署 | 尚未开展;仅为未来路径 | 不适用(推测性) | 是否有已签 LOI、授权框架、IP 所有权结构 |
| 自研实验数据授权 | 出售 AI 生成材料数据集的独家或非独家访问权 | 订阅或按数据集收费;尚未部署 | 尚未开展;仅为未来路径 | 不适用(推测性) | 数据资产清单、授权模式、排他性条款 |
Periodic Labs 未披露收入金额。状态、质量和尽调问题由 a16z、Forbes、Observer 和 TechFundingNews 的报道推断得出。合同价值仅为分析师估计。推测性收入来源只是前瞻性可能性,并非已确认业务线。
[CI011, CI012, CI015, CI016, CI038, CI039]| 定价维度 | 已知 / 估计 | 来源依据 | 置信度 | 尽调问题 |
|---|---|---|---|---|
| 标价 | 未公开披露 | 无公开价目表;公司未发布定价 | 低 | 要求提供标准项目定价或参考合同 |
| 合同结构 | 可能为定制化里程碑制;高额预付款 + 成功费 | a16z 称其「在前沿落地并扩张」;TechFundingNews 提到保密合同 | 中 | 确认结构(固定费用、T&M、里程碑、按结果付费) |
| 典型合同价值 | 估计重大项目为数千万美元 | 可比深科技 R&D 合同的行业来源 | 低(估计) | 要求提供已签合同样本或匿名化交易规模 |
| IP 共同所有权 | 据报道,部分合同包含 IP 共同所有权条款 | 行业来源;描述为保密 | 低 | 审查客户合同中的 IP 所有权和授权条款 |
所有定价均来自估计或第三方报道推断;Periodic Labs 未披露定价模型。「估计数千万美元」是分析师推断,不是公司披露。在尽调确认前,所有数字只能视为方向性参考。
[CI014, CI015, CI040]客户研发问题如何经过 Periodic Labs 的自主实验室平台,转化为合同收入。
工作流根据 a16z 投资逻辑、Periodic Labs 网站和 Observer 报道推断;具体合同机制和里程碑结构未公开披露。
[CI012, CI015, CI039, CI040]4.2 成本结构与资本强度
Periodic Labs 的成本结构由三类主导:(1)自主实验室资本开支和持续维护;(2)AI 计算基础设施(用于训练和推理的 GPU 集群与云计算);(3)从 OpenAI、DeepMind 和 Meta 吸引研究者所需的顶尖人才薪酬。公司已经聘用了一批离开这些公司并放弃大量股权包的研究者,意味着需要远高于市场的薪酬来吸引这类人才。第四类较小成本是办公设施和 G&A。这种组合更像资本密集型物理科学基础设施业务,而不是软件公司。 自主实验室建设是最具结构新意的成本项。可比云实验室平台提供了有用的地板基准:Emerald Cloud Lab 自有数据显示,一个标准自动化化学设施的初始仪器成本为 $1.4M–$3.6M,年度维护为 $288K–$720K;NCBI/PMC 学术研究引用 $250K/year 作为仅访问 Emerald Cloud Lab 完整仪器套件的入门成本。Periodic Labs 正在建设完全自研的设施,包括定制机器人、粉末合成系统和集成 AI 反馈闭环——成本层级显著高于访问费模式。行业分析师估计,每个完全自主的材料发现站点建设成本为 $10–50 million,这与 $300 million 融资规模相符。若规划多个站点,实验室总资本可能达到 $100–200 million,在首轮回本之前就吃掉种子轮很大一部分。 毛利率在结构上不确定:AI 科学合同收费率高(单个项目可能 $5–20M),但要抵消实验室折旧、计算和人才成本,而这些成本很难按合同分摊。纯软件类比会高估毛利;纯 CRO(合同研究组织)类比会低估毛利。公司没有披露毛利率。ViaNews 分析师提示,如果早期结果令人失望,「资本结构给转向或时间线延长留下的空间有限」,这凸显了高固定成本相对软件同业会降低运营灵活性的风险。[CI019, CI020, CI021, CI022, CI023, CI032]
| 指标 | 数值 / 状态 | 置信度 | 重要性 | 尽调问题 |
|---|---|---|---|---|
| ARR / 收入运行率 | 未披露 | Unknown | 衡量商业牵引力和增长轨迹的基准 | 要求提供截至 2026 年 Q2 的审计后或管理层报告 ARR |
| 单合同毛利率 | 未披露;结构上好坏参半(高价值收费 vs. 高实验室 / 人才成本) | Unknown | 判断商业模式放大后能否在经济上持续 | 要求按项目提供 P&L,展示收入、直接实验室成本、算力、人才 |
| CAC / 销售回本周期 | 未披露;直销企业模式意味着周期长、CAC 高 | Unknown | 反映 GTM 的资本效率;是 Series B 承销的关键 | 根据销售团队规模、配额和合同价值估算 CAC |
| NRR(净留存率) | 未披露;合约式模式的 NRR 可能低于 SaaS | Unknown | 经常性收入与项目收入的占比,是估值倍数的核心 | 确认合同续约率和现有客户扩张收入 |
| 烧钱倍数 | 未披露;估计烧钱 $5–15M/mo,收入未知 | 低(分析师估计) | 衡量资本效率;对这一阶段的深科技至关重要 | 提供过去 4 个季度的净烧钱和新增 ARR |
| 客户数量 | 未披露;半导体 / 航天 / 国防合计可能为低个位数 | 低(由公开表述推断) | 集中度风险;流失一个客户就可能产生实质影响 | 披露客户数量、按客户拆分收入、流失历史 |
所有指标要么未披露,要么为估计;本表只总结公开信息状态。置信度反映可得证据,而非模型可靠性。每个空缺指标都需要在承销前提出具体尽调要求。
[CI013, CI017, CI019, CI027, CI038]| 公司 / 基准 | 类别 | 融资额(USD) | 估计实验室 / 基础设施资本开支 | 隐含月度烧钱 | 关键对比点 |
|---|---|---|---|---|---|
| Periodic Labs | AI 科学 / 自主实验室 | $300M 种子轮(2025 年 9 月);$500M Series A 洽谈中(2026 年 5 月) | 每个站点 $10–50M;计划建设多个站点 | $5–15M/mo(分析师估计) | 参照公司;所有数据如本章所述 |
| Emerald Cloud Lab | 商业云实验室(生物 / 化学) | 累计融资约 $47M(公开数据) | 每个设施 $1.4–3.6M 仪器 + $288–720K/yr 维护 | 未披露 | 访问费模式;资本开支远低于自研自主实验室 |
| Strateos(前称 Transcriptic) | 实验室自动化平台 | 融资约 $100M+(收购前) | 模块化;客户站点每种方法 $100K+ | 未披露 | 服务驱动模式;硬件密集度低于 Periodic 的自研建设 |
| Kebotix(AI 材料) | AI 驱动材料发现 | 融资约 $12M | 定制化;项目制;披露资本开支更低 | 未披露 | 规模小得多;收入前模式;可作为材料 AI 对标 |
| Atomwise(AI 药物发现类比) | AI 药物发现 | 融资约 $174M | 仅计算;无实体实验室资本开支 | 未披露 | ViaNews 将 Atomwise 作为警示案例:融资 $123M,截至 2025 年无获批分子 |
Periodic Labs 资本数据来自 Forbes、TechCrunch 和公司来源。Emerald Cloud Lab 成本数据来自公司网站和 NCBI/PMC 学术研究。Strateos 与 Kebotix 数据来自网络搜索,仍需核验。Atomwise 对比来自 ViaNews 的反向分析。所有可比数据均为近似值,来源是公开报道;直接财务对比需要私营数据披露。
[CI020, CI021, CI022, CI031, CI043]示意性展示 $300M 种子轮在主要成本类别中的分配;所有数字均为分析师估计。
所有分配数字均为分析师估计,来自行业基准(Emerald Cloud Lab、可比 AI 实验室)以及 ViaNews 和 Nextomoro 发布的烧钱速度估计。Periodic Labs 未披露实际分配。“剩余资本”数字假设公司按中点烧钱速度运营约 18 个月; 实际剩余现金取决于真实支出节奏和收入抵消。
[CI020, CI023, CI030, CI034]4.3 资本充足性与现金跑道
Periodic Labs 进入 2026 年时,已披露资本全部来自 $300 million 种子轮。月度烧钱速度未披露;分析师估计区间为每月 $5 million 到 $15 million,反映实验室建设、计算支出和竞争性人才成本的组合。按每月 $10 million 中位数计算,$300 million 种子轮可从 2025 年 9 月交割起提供约 30 个月现金跑道,也就是大约到 2028 年 3 月——明显早于 Series A 交割。下限($5M/month)把跑道延至 60 个月;上限($15M/month)压缩至 20 个月,可能在 Series A 资金到账前的 2027 年中触及。 如果 Series A 以 $500 million、$7.5 billion 估值交割,在相同烧钱速度下会再提供 33–100 个月现金跑道,在中等烧钱速度下基本可把公司资助到 2030 年代中期。这一点在结构上重要,因为材料科学发现周期——即便有 AI 加速——也以年为单位。累计 $800 million 融资将给公司一条可信的多年跑道,用来在再次接触资本市场前证明突破。 不过,资本充足性风险仍来自清算优先权结构:$300 million 种子轮优先股在所有普通股股东之前拥有清算优先权。任何等于或低于隐含 $1.2–$1.3 billion 种子轮估值的退出中,普通股持有人(包括员工)都拿不到回报。Series A 还会增加进一步的优先股堆叠。UpsideList 估计,如果商业化停滞,悲观情景下当前水平的股权价值可能下跌 -70%,这会完全抹去普通股价值。公司的债务和项目融资义务未披露;没有公开迹象显示有授信安排或设备租赁补充股权资本,不过 2026 年 5 月提交的、由 Sydecar 管理 SPV 的 SEC Form D 表明,围绕 Periodic Labs 的二级投资人融资活动仍在进行。[CI005, CI006, CI007, CI017, CI018, CI024]
| 项目 | 数值 / 估计 | 置信度 | 来源 | 备注 |
|---|---|---|---|---|
| 种子轮融资额 | $300 million(2025 年 9 月) | 高 | TechCrunch、Forbes、a16z、Periodic Labs 网站 | 公布时已披露的最大 VC 种子轮 |
| 种子轮投后估值 | $1.3 billion | 高 | TechCrunch、Forbes、TechFundingNews | Series A 前参考点 |
| 估计月度烧钱 | $5M–$15M(分析师估计) | 低 | ViaNews、分析师模型 | 公司未披露;区间反映实验室 + 人才 + 算力假设 |
| 估计种子轮现金跑道 | 自 2025 年 9 月起 20–36 个月(约 2027 年 3 月–2028 年 3 月) | 低 | 由 $300M 和估计烧钱区间推导 | 若新实验室建设推高烧钱,现金跑道会缩短 |
| Series A(计划 / 推进中) | $500 million,估值 $7.5 billion(深入洽谈;截至 2026 年 6 月尚未确认完成交割) | 中 | Forbes、Bloomberg、LetsDataScience | 由 AMP(Anjney Midha)领投;多方来源称超额认购 |
Periodic Labs 未披露在手现金、债务融资额度和项目融资义务。烧钱速度和现金跑道是分析师估计,并非公司提供的数据。Forbes(2026 年 5 月)称 Series A 处于「深入洽谈」,截至本报告日期 2026 年 6 月 10 日尚未确认完成交割。
[CI001, CI005, CI006, CI017, CI018, CI024]在硬数字缺失处,用来源支撑的区间框定 Periodic Labs 的关键财务参数。
烧钱速度和现金跑道区间来自 ViaNews 反向分析和分析师评论;Periodic Labs 未确认。Series A 规模和估值来自 Forbes 与 Bloomberg 报道(尚未确认交割)。实验室建设区间来自行业分析师估计和 Emerald Cloud Lab 基准。所有数值都应视为框定区间的估计。
[CI005, CI017, CI018, CI020, CI034]4.4 公开财务信息缺口
作为一家仍处在 Series A 阶段的私营公司,Periodic Labs 的公开财务披露义务很少,也没有主动公布财务指标;公开信息只确认公司在半导体、航天和国防垂直领域已有付费客户。标准 SaaS 和深科技指标——年经常性收入(ARR)、毛利率、净留存率(NRR)、获客成本(CAC)、回本周期、客户数、流失率、收入同比增长——全部未披露。财务预测、单位经济模型和产品定价也没有发布。 Periodic Labs 本身没有直接提交 SEC Form D,这一点值得注意:公司要么通过其他实体名称使用 Regulation D 豁免且未公开备案,要么依赖结构性豁免。已识别的 SEC Form D(2026 年 5 月提交)属于投资 Periodic Labs 的 AGC-Sydecar SPV,能确认有老股交易活动,但不能提供公司资产负债表或损益表层面的直接资本化数据。 尽调中最关键的私有指标请求包括:月度烧钱速度、按季度计算的现金跑道、按合同拆分的毛利率、客户数与收入集中度、ARR 或收入运行率、IP 共同所有权条款、下一轮融资触发条件,以及任何债务或授信文件。截至 2026 年 6 月,公开来源拿不到这些信息;任何潜在股权投资人或商业伙伴都会面临显著信息不对称。[CI013, CI027, CI035, CI036, CI037]
| 缺失指标 | 对分析的影响 | 缺失原因 | 尽调路径 |
|---|---|---|---|
| ARR 和收入运行率 | 无法评估商业牵引力、增长率或估值倍数 | 私营公司;无公开披露要求 | 要求提供审计财务报表或管理层编制的收入明细表 |
| 按合同拆分的毛利率和 EBITDA | 无法评估利润率路径或资本密集度取舍 | 未披露;当前规模下可能为负 | 要求提供含收入和直接成本分摊的利润表 |
| 月度烧钱速度和现金余额 | 无法充分验证现金跑道说法或资本充足性 | 未披露;估计差异很大 | 要求提供月度现金流量表和资金余额 |
| 客户数量和收入集中度 | 无法评估集中度风险或客户流失 | 被描述为保密;国防 / 半导体客户通常如此 | 要求提供匿名化客户名单,含收入层级和合同状态 |
| Periodic Labs 直接 SEC 备案 | 无法核验股权结构表或已披露发行条款 | 未在「Periodic Labs」实体名称下找到 Form D;可能豁免或使用替代名称 | 用所有可能实体名称检索 EDGAR;尽调中要求提供股权结构表 |
缺口来自截至 2026 年 6 月的公开来源审查。影响评级反映这些缺口对正式投资决策的重要性。Periodic Labs 本体(区别于投资者 SPV)缺少 SEC 备案数据,是一个值得注意的缺口。
[CI013, CI027, CI035, CI036, CI037]4.5 财务结论与尽调阻断项
对一个深科技登月项目而言,Periodic Labs 在种子轮到 Series A 阶段讲出了一个财务上可信的故事:融资能力突出、使命有差异化、早期商业信号已经出现,资本基础也足以支撑数年运营。不到八个月估值从 $1.3 billion 跳到 $7.5 billion,反映投资人确实热情高涨;但这个价格也嵌入了一个乐观情景——自主实验室效率得到证明,半导体客户扩大投入,材料突破能在投资人的时间窗口内出现。 目前收入质量偏低:收入仍处早期,定制化、项目制明显,没有披露复购、规模或多元化。毛利率路径存在结构性不确定——实验室模式资本开支重,前期实验室成本摊销后,早期收入毛利率可能为负。对一家软件相邻公司来说,资本强度极高;它更像专业 CRO、纳米科技公司或早期制药公司,而不是纯 AI 公司。ViaNews 的反向分析和关于 AI 投资动态的 arXiv 论文都提示,VC 支持的自主科学初创公司面临二元化结果和漫长资本周期,可能不符合典型基金回报时间表。 关键尽调阻断项包括:(a)确认 Series A 是否已按报道的 $7.5 billion 估值完成交割;(b)私下披露烧钱速度、现金跑道和单项目毛利率;(c)客户合同条款(收入金额、期限、IP 所有权);(d)通往下一次融资事件的里程碑路线图;(e)任何关于自主实验室相对传统实验方法表现的独立验证。在这些问题解决之前,财务图景高度依赖管理层叙事和投资人热情,而不是可验证的牵引指标。[CI009, CI019, CI031, CI033, CI034, CI038]
从合同收入走向毛利时会抵消收入的关键成本类别;所有量级均未披露。
毛利位置存在结构性不确定;没有财务披露。所有成本类别均根据 a16z 论述、ViaNews 分析和行业基准推断。 图表仅为定性;不要把节点大小或边权重解读为比例关系。
[CI017, CI019, CI023, CI034]4.6 附录
05产品与技术
5.1 产品愿景与 AI 科学家核心架构
Periodic Labs 的核心产品是一个「AI 科学家」系统,目标是在物理科学中自动化完整科学发现周期。公司在公开发布材料和 a16z 投资公告中阐明的创始论点是:大语言模型已经耗尽估计 10 trillion token 的互联网语料,需要一种质变的新数据源——靠直接与物理世界互动生成的自有实验结果。产品架构把 AI 推理系统放进物理实验室设备的闭环反馈中,让 AI 提出假设、用机器人实验检验假设、观察预测是否经得住物理现实,再据此更新模型。自然本身就是强化学习环境——每一次合成实验都会确认或反驳 AI 的材料预测,提供互联网语料中没有的、明确且高保真的训练信号。 公司用「从比特到原子」描述自己的路径——把数字 AI 推理(bits)与物理世界实验(atoms)接起来。初始目标应用选择高温超导体发现,是因为相关物理实验相对较快,结果高度可验证(导电性和临界温度可客观测量),物理仿真生态(密度泛函理论)也足够成熟,能在投入机器人合成实验前缩小候选搜索空间。每次实验都会产生数 GB 高质量数据,包括传统科学文献很少发表的失败结果——由此形成一套会随实验次数增长、仅靠公共数据库无法复制的自有训练语料。 第二条商业产品线已经启动:为一家未具名半导体制造商训练定制 AI 智能体,帮助其解决芯片散热问题。这条产品线把 Periodic Labs 的实验推理能力转化为近期开票的企业产品,让工程师能更快解释并迭代实验数据,而不必等完整自主机器人集成落地。这种双模式策略——研究阶段发现超导体,商业阶段交付工程智能体——让公司在更长周期的材料发现平台成熟前拥有近期收入载体。公司还服务航天和国防客户,但没有披露项目细节。 [CE002, CE004, CE007, CE008, CE013, CE014]
| 用户任务 | 当前工作流(无 Periodic) | Periodic Labs 方案 | 声称可衡量收益 | 已知限制 |
|---|---|---|---|---|
| 发现新超导体 | 人类研究人员多年开展定向实验;只发表正结果 | AI 从文献和模拟生成假设;机器人执行;捕获所有结果数据 | 实验迭代量级提升,并包含失败数据 | 截至 2025 年 10 月,全自主机器人闭环尚未运行 |
| 识别稳定晶体结构 | DFT 计算和人工筛选;受算力和人力带宽限制 | GNoME 风格 GNN 预测稳定性;AI 在合成前预筛数百万候选 | AI 预测候选在机器人合成规模上验证 | 新型化合物类别的预测准确率未经独立基准测试 |
| 解读芯片散热实验数据 | 工程师手动分析实验结果;迭代周期慢 | 定制 AI 智能体解读实验数据,并提出设计干预建议 | 半导体热管理 R&D 迭代更快 | 半导体客户未具名;效率提升未披露 |
| 生成 AI 模型训练数据 | 依赖静态互联网抓取文本语料(约 10T tokens 已耗尽) | 自主实验室每次运行生成 GB 级实验数据,并持续更新 | 竞争对手拿不到的新鲜、高信号、自研训练语料 | 当前数据生成规模未披露;模型改进归因未经验证 |
| 航天和国防材料发现 | 长周期政府和内部 R&D 项目 | AI 科学家瞄准新材料,并配有明确物理性能评估 | 压缩任务关键材料的发现周期 | 具体应用和结果未披露;安全合规未说明 |
| 跨学科科学 R&D | 物理学家和 ML 研究人员在孤立团队中工作,共同语言有限 | 每周跨学科教学;强制建立 ML 与物理之间的共同流利度 | Bits/Atoms 整合更紧;科学迭代更快 | 员工增长后,这种知识转移模式能否扩展仍不确定 |
用例来自公司公开表述、a16z 投资公告、TechCrunch 报道和招聘信息分析。声称收益由公司提出;暂无独立验证。
[CE004, CE007, CE008, CE013, CE014, CE033]七步自主发现周期,从 AI 假设生成开始,经过物理合成和表征,走向自研数据捕获和模型更新。
工作流综合自 a16z 投资文章、periodic.com、TechCrunch 2025 年 10 月报道和 AI Insider 报道。第 3 步 (机器人合成)截至 2025 年 10 月尚未完全自动化。
[CE004, CE005, CE006, CE008, CE026, CE027]5.2 技术栈:Bits 与 Atoms 双轨平台
Periodic Labs 围绕两条明确命名的轨道组织技术,公开招聘和内部沟通中都能看到:一条是「Bits」,覆盖 LLM 研究、机器学习基础设施和分布式训练工程;另一条是「Atoms」,覆盖物理实验室自动化、机器人、粉末合成工艺工程、薄膜沉积和材料表征。双轨架构反映了公司的根本判断:要交付 AI 科学家,必须同时在两个很少共存于同一组织的领域做到极强。 「Atoms」平台以粉末合成实验室为核心,机器人手臂混合前驱体化学品,在炉中加热到指定温度,再用表征仪器测量所得材料的电导率、磁响应、晶体结构等性质。公司在 2025 年 10 月确认,加州 Menlo Park 的实验室正在建设,但当时完整机器人栈尚未运行。截至 2026 年 6 月,招聘信息显示公司正在招自动化工程师、工艺工程师(粉末)、材料合成与薄膜方向研究科学家,以及聚焦半导体的多物理场仿真科学家——说明 Atoms 轨道仍在积极建设,范围已经超出最初的粉末合成重点。 「Bits」平台以面向科学推理的 LLM 训练和微调为核心。公司通过三类训练来构建领域专用 AI 模型:用科学文献预训练,用自有实验数据中期训练,并根据实验结果做强化学习。联合创始人 Cubuk 将这一路径描述为把物理当作类似数学和代码的「可验证环境」——在这些领域,RL 驱动的 AI 进展最快。Bits 团队招聘分布式训练工程师、ML 系统工程师和超级计算工程师,显示训练大规模领域专用模型所需的基础设施规模。 连接两条轨道的是一层量子力学仿真,用来在进入物理合成前缩小候选搜索空间。这一层直接继承 GNoME 方法——用图神经网络预测晶体稳定性——以及密度泛函理论软件工具,让 AI 能快速筛选数百万种候选化合物组成,再选择一个子集做物理验证。a16z 的投资逻辑明确写道:「模型会阅读文献、运行量子力学仿真、在实验室中行动,并从自然本身获得反馈。」负向实验结果会被有意捕获并纳入训练数据;相比只偏向正向结果的已发表科学文献,这构成结构性优势。 [CE003, CE005, CE006, CE019, CE026, CE027]
| 层级 / 组件 | 系统角色 | 关键依赖 | 技术风险 |
|---|---|---|---|
| LLM 假设生成(Bits) | 从文献和既往实验数据提出候选材料组成和合成条件 | 前沿 LLM 能力;特定领域科学训练数据 | a16z 称 AI 推理在凝聚态物理上低于人类专家;输出质量未经外部验证 |
| 量子力学模拟(Bridge) | 缩小搜索空间;在物理合成前过滤低概率候选 | DFT 软件(VASP、Quantum ESPRESSO 或同等工具);模拟运行所需算力 | 面对新型化合物类别,模拟准确率会下降;假阳性会浪费实验室产能 |
| 机器人粉末合成(Atoms) | 混合前驱体粉末,在炉中加热,并规模化产出物理样品 | 可靠的粉末合成机械臂;材料供应链;化学安全基础设施 | 截至 2025 年 10 月尚未完全运行;建设时间表和吞吐量未披露 |
| 材料表征(Atoms) | 测量合成样品的电导率、临界温度、结构和磁性 | 分析仪器(XRD、SQUID 磁强计等);校准和计量规程 | 仪器吞吐量限制每日实验量;测量误差会传导到训练数据 |
| 专有数据管线 | 摄取、标注并存储所有实验结果,包括失败样本;供模型训练使用 | 扎实的数据工程;科学实验记录的元数据标准 | 数据集质量、标注准确性和覆盖范围尚未经过独立评估 |
| AI 模型训练(Bits) | 在专有实验数据上对领域模型做中期训练和 RL 微调 | GPU / 超算集群;通过 NVentures 与 Nvidia 形成战略协同;分布式训练基础设施 | 算力依赖带来基础设施成本,也可能形成供应商集中风险 |
| 商业代理层(Bits) | 将 AI 科学家能力打包给外部客户(半导体、航天、国防) | 客户数据访问协议;工程导入;IP 授权条款 | GTM 模式尚未公开;客户集成复杂度未量化 |
| 实体实验室基础设施(Atoms) | 承载机器人和表征设备;提供安全与合规系统 | Menlo Park 设施;化学品处理安全基础设施;监管合规 | 实验室监管状态、安全认证和化学品处理规程未见公开文件 |
架构判断来自公司公开表述、a16z 投资文章、TechCrunch 报道和招聘信息分析。Periodic Labs 尚未发布独立的架构文档。
[CE003, CE004, CE005, CE006, CE018, CE019]五层架构从物理世界仪器出发,穿过实验室自动化、量子仿真、AI/ML 核心,最终到商业应用交付。
架构根据公司公开表述、a16z 文章、TechCrunch 报道和 2026 年 6 月招聘信息分析推断。具体软件组件和层间接口未公开记录。
[CE003, CE004, CE005, CE026, CE027]5.3 技术血统:GNoME、ChatGPT 与 MatterGen 谱系
创始团队的技术履历,是 Periodic Labs 产品主张和执行 AI 科学家论点能力的核心。联合创始人兼 CEO Liam Fedus 曾任 OpenAI 后训练研究副总裁,领导团队创造了 ChatGPT 和首个万亿参数神经网络。他在后训练上的专长——决定模型行为的关键 RL 微调和指令对齐阶段——可以直接迁移到 Periodic Labs 正在构建的「从自然中 RL」训练范式。联合创始人 Ekin Dogus Cubuk 曾领导 Google Brain 和 DeepMind 的材料与化学研究团队。他最知名的成果是 GNoME(Graph Networks for Materials Exploration),即 2023 年 Nature 论文:团队用基于 DFT 计算晶体能量训练的图神经网络,识别出超过 2.2 million 个潜在稳定的无机晶体结构,这是已知材料史上最大规模的一次扩展。GNoME 证明,AI 预测材料可以在机器人合成实验室中验证;Berkeley A-Lab 在同一个 2023 研究周期中,按 AI 配方合成了 41 种新化合物。这个先例直接支撑 Periodic Labs 的架构,也验证了闭环路径的可行性。 除两位联合创始人外,团队还包括 Alexandre Passos(OpenAI o1 和 o3 推理模型共同创造者)、Eric Toberer(曾任 Colorado School of Mines 的材料科学家,做出过关键超导体发现)以及 Matt Horton(Microsoft MatterGen 的创造者之一,MatterGen 是用于生成式材料发现的 LLM)。公司称已从 OpenAI、DeepMind、Meta、Databricks 和 Samsung 招募超过 20 名研究人员,其中许多人放弃数千万到数亿美元未归属股权加入公司。《纽约时报》将这场人才迁移报道为近年 AI 史上最值得关注的事件之一。 一个独特的团队实践强化了技术融合:每周跨学科教学,物理学家教 LLM 围绕量子力学推理,ML 研究人员学习物理直觉。联合创始人 Cubuk 表示:「我们确实认为紧密耦合极其重要。」创始人还记录称,粉末合成工作流中的机器人手臂可靠性直到最近才达到足以支撑可靠自主实验的门槛;创始人认为,这种时点上的汇合正是现在适合创建 Periodic Labs 的关键原因。 [CE001, CE009, CE010, CE011, CE012, CE016]
| 模块 / 资产 | 主要用户 | 成熟度状态 | 差异化 | 关键尽调缺口 |
|---|---|---|---|---|
| AI Scientist Core(LLM 假设引擎) | 内部研究团队 | 商业化前;正在训练 | 继承 GNoME + ChatGPT 后训练经验;采用 nature-as-RL 设计 | 无外部基准;相对人类基线的能力未经验证 |
| Powder Synthesis Atoms Lab(机器人) | 内部研究团队 | 建设中(Menlo Park);截至 2025 年 10 月尚未完全运行 | 与 AI 闭环;每次运行生成 GB 级自研数据 | 机器人落地时间表未披露 |
| 量子力学模拟(桥接层) | 内部研究 / AI 训练 | 已运行(继承 GNoME) | 基于 DFT 的晶体稳定性筛选,在物理实验前缩小搜索空间 | 与 LLM 假设引擎的整合深度未披露 |
| 自研实验数据集 | AI 模型训练 | 已开展;随每次实验运行增长 | 包含负结果;公共数据库无法获得 | 规模、覆盖范围和治理政策未披露 |
| 半导体热管理智能体(商业) | 半导体制造商工程师 | 试点 / 活跃客户项目 | 定制智能体解读自有实验数据,服务芯片 R&D 迭代 | 客户未具名;结果、定价和合同范围未披露 |
| 超导体发现计划 | 研究 + 未来商业化 | 研究阶段;无已发表结果 | AI 指导合成,目标是高温超导体 | 无同行评议产出;实验室到市场通常需 10–20 年 |
| 国防 / 航天材料计划 | 国防和航天行业客户 | 商业化前 | 公司称航天和国防是当前客户行业 | 客户名称、应用和结果未披露 |
| 薄膜 / 半导体工艺实验室(计划) | 半导体 R&D 团队 | 招聘阶段;尚未运行 | 将 Atoms 实验室延伸到半导体薄膜工艺工程 | 建设时间表和范围未公开披露 |
成熟度状态基于 2025 年 10 月 TechCrunch 报道、2026 年 6 月招聘信息和公司公开表述。暂无独立验证可确认运行状态或商业结果。
[CE002, CE003, CE005, CE013, CE014, CE015]5.4 运营状态、路线图与关键依赖
截至 2025 年 10 月,Periodic Labs 确认已在 San Francisco 搭建研究实验室,并在积极处理实验数据、仿真和测试部分预测。联合创始人 Cubuk 明确告诉 TechCrunch,机器人组件「还需要一点时间训练」——说明发布时完整 Atoms 轨道自主闭环尚未运行。截至 2026 年 6 月,Menlo Park 实验室仍在招聘自动化工程师、工艺工程师(粉末)以及材料合成与薄膜方向研究科学家。面向半导体的多物理场仿真科学家招聘,显示技术路线图正在从最初的粉末合成超导体重点,扩展到薄膜沉积和半导体工艺工作流;这也确认产品范围比公开叙事暗示的更宽。 公司的产品路线图有三个可见阶段:(1)为超导体发现建设并扩展自主粉末合成 Atoms 实验室;(2)为半导体客户部署商业化工程智能体;(3)扩展到更多材料目标,包括电池、催化剂、磁体、隔热盾,最终到药物化合物。a16z 投资逻辑将先进制造、材料科学、半导体、能源和航空航天列为优先市场,合计约占全球 GDP 的 $15 trillion。Forbes 2026 年 5 月报道称,公司后续一轮融资在 $7.5 billion 估值下大幅超额认购,且已经在讨论以更高估值快速跟进的下一轮融资。 关键依赖包括:(a)实验室规模下可靠的机器人硬件;据联合创始人证词,这项能力最近才足够成熟;(b)训练大型领域专用模型所需规模的 GPU / 计算资源,Nvidia 的 NVentures 作为战略种子投资人,暗示硬件层面对齐;(c)前沿 ML 与物理科学交叉处的专业人才,公司正在从顶级 AI 实验室积极招募;(d)与 R&D 预算大、实验评价标准清晰的行业客户建立关系。闭源权重策略降低了开源复制风险,但也限制了外部验证能力主张。公司没有披露与云服务商的具体计算基础设施安排,也没有说明与 NVentures 的战略关系性质。 [CE015, CE018, CE020, CE023, CE024, CE037]
| 日期 / 阶段 | 里程碑 | 状态 | 含义 | 来源 |
|---|---|---|---|---|
| 2025 年 3 月 | 公司由 Liam Fedus 和 Ekin Dogus Cubuk 创立;Felicis 在公司注册前开出第一张支票 | 已完成 | 创始团队成形;Fedus 公开离开 OpenAI 后,VC 追捧随之而来 | TechCrunch 2025 年 10 月 |
| 2025 年 1 月(发布前) | Periodic First Release 闭源权重模型发布 | 已完成 | 首个模型工件;架构和基准表现未公开 | Nextomoro 档案 |
| 2025 年 9 月 30 日 | 宣布 $300M 种子轮;公司以 $1.3B 投前估值走出隐身 | 已完成 | 创纪录种子轮;a16z 领投;Bezos、Schmidt、Dean、Gil 作为天使投资人参与 | TechCrunch 2025 年 9 月;a16z 公告 |
| 2025 年 10 月 | 旧金山研究实验室已运行;正在使用实验数据和仿真;机器人尚未跑起来 | 已完成 | 实验室基础设施仍处早期;Menlo Park 建设推进中 | TechCrunch 2025 年 10 月 |
| 2026 年上半年(进行中) | 为 Menlo Park Atoms 实验室招聘自动化工程师、工艺工程师(粉末)、多物理场仿真科学家 | 进行中 | Atoms 线扩张;薄膜和半导体工艺能力被纳入范围 | Ashby 招聘信息 2026 年 6 月 |
| 2026 年 5 月 | 正深入谈判,由 AMP(Anjney Midha)领投、按 $7.5B 估值融资 $500M+;据称该轮大幅超额认购 | 据称进行中 | 不到 8 个月估值跳升六倍;更高估值的快速跟进轮已被讨论 | Forbes 2026 年 5 月 |
时间线基于公开报道日期。未来里程碑——机器人投入运行、超导体发现结果和模型发布——公司尚未公开承诺。
[CE001, CE011, CE016, CE018, CE020, CE023]AI 科学家平台的关键上游依赖——硬件、计算、人才、仿真软件和数据——以及下游客户关系。
依赖边来自公开公司表述和投资人材料推断。具体合同关系(例如与 Nvidia 或云厂商的算力合同)尚未公开披露。
[CE003, CE015, CE016, CE019, CE028, CE034]基于截至 2026 年 6 月的公开证据,评估八项产品与技术能力的成熟度。
成熟度评级基于公开证据评估;Periodic Labs 的技术就绪度尚无独立审计。所有“可运行”判断均基于公司说法。
[CE013, CE014, CE015, CE018, CE019, CE020]5.5 信任、安全、技术风险与关键证据缺口
Periodic Labs 所处领域中,AI 预测与物理验证之间的差距既是护城河来源,也是最大的执行风险。负责尽调的 a16z 牵头合伙人指出,前沿 AI 模型在凝聚态物理中的科学分析能力「客观上很糟糕」,并且「相对人类研究者更差」——公司自己的牵头投资人承认,起点 AI 能力低于人类专家基线。Periodic Labs 声称会靠闭环训练弥合这一差距,但目前这仍是一个开放且未经验证的问题。 超导体发现目标带有结构性风险。尽管国际研究投入数十年,室温超导体仍停留在理论层面。2023 年 LK-99 事件——一个曾短暂吸引巨大关注、随后无法被独立复现的主张——说明超导体声明在审查下可以多快坍塌。Via.news 指出,传统材料开发从实验室到商业部署平均需要 10 到 20 年,这在投资人回报周期和物理科学时间线之间制造了结构性张力。AI 在材料科学中的应用迄今商业成功有限;物理验证瓶颈不能仅靠提高预测速度消除。 截至 2026 年 4 月,Periodic Labs 尚未发表关于其 AI 科学家能力的同行评审论文或外部基准。闭源权重策略排除了外部评估。Sakana AI 的开源 AI Scientist-v2 系统可作为理解行业成熟度的有用参照——它自主生成了一篇被 ICLR workshop 接收的论文——但该系统运行在数字 ML 研究中,而不是物理世界材料发现,也与 Periodic Labs 无关联。 信任、安全和数据安全协议尚未公开记录。公司用工业规模机器人进行化学合成,带来材料处理、安全认证和监管合规问题,而公开材料没有回应这些问题。自有实验数据集的知识产权保护——公司声称的主要护城河——仍是治理缺口。公开材料中没有出现已披露的数据治理政策、商业客户许可条款或 AI 安全监督协议。 [CE021, CE022, CE028, CE029, CE030, CE031]
| 控制 / 认证 | 状态 | 范围 | 已知缺口 |
|---|---|---|---|
| 化学品处理与实验室安全 | 未公开记录 | 实体 Atoms 实验室(Menlo Park) | 公开材料未披露安全认证或实验室合规标准 |
| 专有数据集的数据安全与 IP 保护 | 未公开记录 | 专有实验数据集(核心护城河资产) | 未披露数据治理政策;商业客户的 IP 授权条款未披露 |
| AI 模型安全与监督规程 | 未公开记录 | AI 科学家 LLM 和 RL 系统 | 未见 AI 安全文档;自主实验设计没有披露人类在环监督规程 |
| 实验可复现标准 | 公司称采用迭代确认方法;未发布规程 | 超导体和材料发现主张 | 未见独立复现实验规程;闭源权重政策使社区无法复现 |
| 国防和航天客户监管合规 | 未公开记录 | 国防和航天领域产品线 | ITAR、EAR 或政府安全合规状态未披露;许可要求未说明 |
所有条目反映公开披露缺失,并不等于确认缺少控制。Periodic Labs 仍是早期私营公司,可能已有尚未公开的内部规程。
[CE020, CE021, CE022, CE029, CE030]5.6 附录
06客户情况
6.1 买方与用户分层
Periodic Labs 服务的是一个窄但高价值的客户宇宙:半导体制造、航空航天和太空探索、国防系统等大型工业 R&D 组织中的工程师和研究人员。买方通常是 R&D 负责人、材料科学负责人或工程副总裁,手里有可观实验数据预算,也有明确材料难题——先进芯片散热、再入飞行器隔热、国防系统中的磁体性能——传统方法未能在既定时间内解决。用户则是实验室工程师或计算科学家,他们接收 AI 智能体输出,并决定是否启动下一轮实验周期。公开信息中没有消费者端界面、没有渠道合作伙伴层,也没有披露地理分层。 a16z 将目标行业描述为「代表数万亿美元 R&D 支出」,并明确点名半导体、先进制造、能源和航空航天为主要影响垂直。Fedus 将理想客户描述为「并没有特别好用工具」且拥有「庞大 R&D 预算」的组织,把 ICP 锚定在技术需求和预算能力,而不是公司规模。官方网站公布 Academic Grant Program 后,一个更早期的次级用户群也开始出现,公司开始触达研究机构。按地理、收入区间、员工数或渠道划分的客户分层尚未披露。[CU001, CU005, CU006, CU007, CU009, CU011]
| 细分 | 买方 / 用户 / 付款方 | 用例 | 规模 / 范围 | 收入 / 战略价值 | 尽调缺口 |
|---|---|---|---|---|---|
| 半导体制造商 | 买方:研发负责人 / 材料负责人;用户:实验室工程师 / 计算科学家;付款方:研发资本开支预算 | 芯片材料的散热分析、实验数据解读、仿真自动化 | 年研发预算达数十亿美元的大型晶圆厂和 IDM;合作深度未知 | 已确认产生收入;未具名;未披露 ACV 或合同规模 | 具名客户、合同结构、试点与生产状态、续约历史 |
| 航空航天和航天公司 | 买方:首席科学家 / 工程副总裁;用户:材料科学家;付款方:IRAD 或政府合同预算 | 发射载具的隔热材料发现、先进磁体、结构复合材料 | 有暗示但未确认;无案例研究;a16z 和 Inc. 在半导体之外提到该领域 | 收入贡献未知;考虑到发射载具对材料成本敏感,战略价值高 | 具名客户、用例具体性、生产还是试点状态、采购工具类型 |
| 国防承包商和机构 | 买方:项目经理 / 总工程师;用户:材料科学家 / 仿真工程师;付款方:IRAD、DARPA 或项目预算 | 用于武器系统、雷达、推进和屏蔽的先进材料;超导体探索 | 有暗示但未确认;投资人和媒体报道把该领域与半导体并列提及 | 收入贡献未知;若通过 ITAR,可能获得大型 IRAD 资助合同 | ITAR 许可状态、具名客户、合同工具、政府主承包商 vs. 直接销售模式 |
| 学术和研究机构 | 用户 / 资助接受方;付款方:基金会或大学预算 | 通过学术资助计划支持材料发现;可能贡献训练数据 | 早期、非收入;通过官网 Academic Grant Program 公布 | 目前非收入;是未来转化为企业客户的战略管线 | 资助计划范围、机构数量、转化意图 |
细分描述基于公司和投资人披露。规模和收入值为推断;官方未披露客户数量、ACV 数据或分细分收入。
[CU001, CU005, CU006, CU012, CU035]| 指标 | 数值 | 日期 | 来源 | 置信度 | 含义 | 缺失分母 |
|---|---|---|---|---|---|---|
| 收入运行率 | <$5M(估算) | Q1 2026 | ZoomInfo(第三方情报) | 低 | 商业化仍处早期;远低于已投入资本($300M 种子轮) | 无分母;ZoomInfo 估算常常滞后于现实 |
| 已确认产生收入 | 是(未披露金额) | 2026 年 3 月 | TechFundingNews 引述 Bloomberg | 中 | 证明至少有一个客户在付费;金额未披露 | 未披露合同价值或客户数量 |
| 已确认客户领域 | 3(半导体、航天、国防) | 2025 年 9 月 | a16z 投资公告 | 高 | 发布时已具备多领域牵引;披露领域不等于可换算客户数 | 各领域内部没有数量;三类可能只对应 1 个客户 |
| 投资人信心信号(超额认购轮) | 按 $7.5B 估值融资 $500M,该轮大幅超额认购 | 2026 年 5 月 | Forbes | 高 | 投资人需求可作为客户管线可信度代理;不是直接客户指标 | 未披露客户数量或管线 ARR |
| Forbes AI 50 Brink 认可 | 入选(2026 年首届榜单) | 2026 年 4 月 | Forbes.com.au 报道 | 高 | 编辑层面对早期牵引的认可;评选标准包含早期牵引 | 不是客户指标;榜单方法论为编辑判断 |
| 员工数(交付能力代理) | 40-48 名员工 | Q1 2026 | AI Market Watch / ZoomInfo 数据 | 中 | 相对企业研发支持需求,团队规模小;限制并行服务客户数量 | 未披露每员工客户比或支持能力 |
所有数值都是估算或代理信号。官方未披露客户数量、ARR 或留存数据。ZoomInfo 对未产生收入或早期收入创业公司的收入估算不确定性很高。
[CU008, CU009, CU010, CU019, CU039]从目标市场认知到账户扩张的示意漏斗,体现半导体和国防 R&D 企业深科技采购摩擦高、周期长的特征。
所有漏斗值均为示意估计,来自员工数(约 40-48 人)、已确认部署信号(1 个公开半导体案例)和典型企业深科技转化率推断。没有官方管线或转化数据。
[CU003, CU009, CU012]6.2 客户证明与具名部署
公开记录中最具体的客户证明,是一家半导体制造商部署 Periodic Labs 的定制 AI 智能体,用来分析散热实验数据并加快芯片工程迭代周期。官方公司网站描述了这一部署,a16z 和 Inc. 也独立确认,但客户名称没有披露。a16z 指出「Periodic 已经在与航天、国防和半导体客户合作」,确认牵引力不止半导体案例,已经跨多个行业。TechFundingNews 在 2026 年 3 月独立报道,Periodic Labs「已经拿下半导体行业客户」,且「与许多同行不同,公司正在产生收入」。 客户证明质量受不透明限制:本次研究没有找到具名客户、已发布案例、G2 或 Gartner Peer Insights 评价,也没有独立客户证言。Forbes 将 Periodic Labs 纳入其首届 2026 AI 50 Brink List,而入选标准要求「早期牵引力」,提供了有限独立验证。截至 2025 年 10 月,TechCrunch 报道称公司的机器人手臂「尚未启动运行」,暗示部分早期客户项目可能处在 AI 智能体和仿真阶段,而不是完整闭环自主实验室部署。任何公开来源都没有确认半导体项目究竟属于试点还是生产部署。[CU002, CU003, CU004, CU008, CU010, CU013]
| 客户 / 细分 | 领域 | 部署 / 用例 | 生产 vs. 试点 | 已披露结果 | 限制 |
|---|---|---|---|---|---|
| 未具名半导体制造商 | 半导体 | 面向散热分析的定制 AI 代理;实验数据解读;为芯片工程师加快迭代 | 不清楚——描述为活跃部署,但未确认试点还是生产 | 实验数据迭代更快;未披露量化的速度或成本改善 | 客户未具名;结果为定性;生产状态含糊;无独立确认 |
| 航天领域客户——未具体说明 | 航空航天 / 航天 | 暗示用于发射载具部件的材料发现;可能包括隔热罩和结构材料 | 未知——未披露部署细节 | 未披露 | 客户未具名且未确认;用例从 a16z 和公司表述推断;无案例研究 |
| 国防领域客户——未具体说明 | 国防 | 暗示用于国防应用的先进材料研发;可能包括电磁、推进或屏蔽材料 | 未知——未披露部署细节 | 未披露 | 客户未具名且未确认;用例为推断;公开记录未处理 ITAR 影响 |
本表列出公开客户合作记录的残片。所有条目均基于公司和投资人表述;没有任何一行获得独立客户确认或第三方案例研究。没有具体具名客户行,反映真实不透明,而不是研究缺口。
[CU001, CU002, CU003, CU007, CU020, CU021]示意客户从首次认知到潜在生产部署与扩张的旅程,并标出各阶段哪些信息已确认、哪些来自推断。
航天和国防阶段来自行业提及推断;只有半导体路径有明确公开证据。扩张阶段没有公开确认案例。
[CU002, CU003, CU004, CU030]对每个已确认或隐含客户行业,在客户证明维度上给出证据质量评级。评级为分析师评估;绿色 = 强,黄色 = 部分,红色 = 缺失。
证据质量评级使用 0–1 标尺:1 = 已确认,0.5 = 隐含 / 部分,0 = 缺失 / 未知。所有数值由分析师基于截至 2026 年 6 月的公开来源推导;没有任何一行代表具名客户。
[CU001, CU007, CU010, CU020, CU021]6.3 伙伴生态与进入市场
Periodic Labs 的投资人联合体同时也是早期进入市场网络。NVIDIA 的 NVentures 参与 $300M 种子轮,为公司连接半导体硬件基础设施,并可能向芯片 R&D 组织提供热引荐。领投方 a16z 围绕先进制造和材料科学对 $15 trillion GDP 的影响有明确论点,也主动撰写了关于半导体、航天和国防牵引力的客户叙事。Felicis(第一张支票)、DST Global 和 Accel 补齐机构网络。个人投资人包括 Jeff Bezos(Amazon)、Eric Schmidt(前 Google CEO)、Jeff Dean(Google Senior Fellow)和 Elad Gil,各自带来个人行业网络。 硅谷顶级科技律师事务所之一 Wilson Sonsini Goodrich & Rosati 为创始轮提供法律顾问,确认交易合法性。Bromley Capital Partners(UK)披露其在 2026 年 1 月为一笔投向 Periodic Labs 的数百万美元私募配售提供顾问服务,显示投资人基础正在扩展到全球机构参与者。2026 年 5 月针对「AGC Wealt Periodic Labs I」提交的 SEC Form D——AGC AI Nexus Fund 旗下基金载体——为持续结构化投资活动提供了监管确认。Forbes 2026 年 5 月报道的 $500M 后续轮被描述为「大幅超额认购」,且公司正在讨论快速跟进的下一轮融资——这是投资人和市场对客户管线有强信心的又一个信号。[CU016, CU017, CU018, CU031, CU032, CU033]
估算各目标客户细分的战略价值指数(0-100),反映 R&D 预算规模、材料发现紧迫性和已确认合作深度。分数越高,近期商业机会越大。
数值为分析师估计,依据公开 R&D 预算数据、已确认合作信号和行业评论。Periodic Labs 未披露按细分划分的官方收入权重。能源和学术细分基于公司使命表述推测。
[CU006, CU015, CU038]6.4 留存、持久性与扩张
截至 2026 年 6 月,Periodic Labs 没有披露留存、NRR、GRR、流失率或合同期限数据,任何第三方来源也没有识别到这些信息。即使最早期采用者,客户关系也不到十二个月,仅靠公开数据做队列层面的留存分析在结构上不可能。公司声称的「在前沿领域落地并扩张」策略——先解决有清晰评估标准的关键点问题,再在账户内扩展——暗示其扩张模型围绕深度领域编码和工作流扩展设计,而不是订阅层级增购。a16z 公告所述的「通过中期训练和强化学习编码深度领域知识」,一旦客户自有实验数据嵌入定制模型,就可能形成高切换成本。 客户扩张的商业模式尚未公开定义。Periodic Labs 没有披露是按模型训练、按实验周期、按 API 调用收费,还是通过年度企业许可证收费。这种模糊性让外部无法预测 NRR、留存经济性或头部客户收入集中度。Academic Grant Program 表明公司愿意与机构建立有粘性的非付费关系,这些关系未来可能转化为付费企业关系。没有发现正式续约、长期承诺或客户合同结构的证据。[CU004, CU025, CU026, CU036, CU037, CU040]
| 指标 | 数值 / 状态 | 细分 | 置信度 | 尽调要求 |
|---|---|---|---|---|
| 净收入留存(NRR) | 未披露 | 全部 | 低 | 在管理层数据室索取按 cohort 划分的 NRR;与企业级 deeptech 同行做基准比较 |
| 总收入留存(GRR)/ 流失 | 未披露 | 全部 | 低 | 获取 GRR 和 logo 流失;确认是否已有试点阶段客户停止合作 |
| 合同期限 / 续约结构 | 未披露;商业模式未公开定义(平台 vs. 服务 vs. IP 授权) | 全部 | 低 | 确认合同结构:订阅、工时材料、IP 授权或其他;核验续约节奏 |
| 客户满意度 / NPS | 未披露;独立评论平台(G2、Gartner Peer Insights、Capterra)没有 Periodic Labs 条目 | 全部 | 低 | 若有跟踪,索取 NPS 或 CSAT;识别可供访谈的客户联系人,用于直接尽调 |
| 账户内扩张速度 | 未披露;“在前沿落地并扩张”是公司表述的战略,但没有可用扩张指标 | 全部 | 低 | 索取 cohort 级扩张 ARR 数据;确认首个半导体合作是否已扩展到更多用例 |
所有指标目前均未披露。公司发布后不到 12 个月;cohort 级数据可能尚无意义。单元格显示 null 是因为未披露,而不是没有客户。
[CU004, CU025, CU026, CU036, CU037]6.5 采购障碍与采用风险
半导体和国防行业采购自主 AI 实验室平台,在结构上速度很慢。这些垂直行业典型验证、安全审查和批准周期为 12-24 个月;国防采购还需要 ITAR 合规审查、数据驻留保证,以及法律、IT、采购和高管层多方对齐。IP 所有权不确定是实质障碍:在国防和半导体 R&D 中,AI 发现的材料洞见归客户还是归 Periodic Labs,具有显著商业和战略价值含义,而公开信息尚未澄清。 更大的行业层面风险是,AI 材料发现还没有出现商业化「尤里卡时刻」。MIT Technology Review 2025 年 12 月报道,没有一家自主实验室初创公司证明自己在稳定性预测之外有实用价值,计算仿真与物理合成之间的差距仍是主要瓶颈。Vianews.market 引述称,AI 材料项目走到商业化生产的失败率为 60-70%。这些行业层面信号会影响所有买方信心,不只是 Periodic Labs。公司的「自然作为 RL 环境」论点——真实物理实验会生成可操作、自有的数据——架构上成立,但尚未在规模商业化上得到证明;早期企业买方很可能要求大量试点,才会承诺生产级合同。竞争平台(Lila Sciences、CuspAI、Radical AI)在重叠垂直中面临同样的采用摩擦。[CU022, CU023, CU024, CU027, CU028, CU029]
| 扩张驱动 / 集中风险 | 类型 | 影响 | 尽调路径 |
|---|---|---|---|
| 单一垂直集中:公开确认的部署只有半导体 | 集中风险 | 高——如果半导体是唯一付费垂直,收入高度集中;航天 / 国防未确认 | 确认各领域客户数量和收入权重;索取分细分 ARR 拆分 |
| 客户名称集中:所有客户均匿名 | 集中风险 | 高——无法判断收入是分散在 5+ 个客户,还是集中于 1-2 个 logo | 索取付费客户数量和最大客户收入占比 |
| 落地后扩张:编码专有领域知识会制造切换成本 | 扩张驱动 | 中等偏正面——基于客户数据做定制模型训练会提高黏性;规模化未经验证 | 确认当前账户内是否已有第二用例扩张;记录扩张节奏 |
| 伙伴生态杠杆:NVIDIA / a16z 网络带来暖启动引荐 | 扩张驱动 | 中等偏正面——NVIDIA 的战略兴趣可能加速半导体管线;价值未确认 | 确认 NVIDIA 或 a16z 被投公司是否为客户;梳理暖启动引荐管线 |
| 商业模式含糊:不清楚是平台、IP 还是服务模式 | 集中风险 | 高——不同模式意味着 NRR、CAC、毛利率和扩张经济性会截然不同 | 获取商业模式文档;确认发现材料的 IP 归 Periodic Labs 还是客户所有 |
影响和类型评估为分析师根据公开证据推断。没有披露合同数据、分领域收入或 IP 所有权结构,无法量化集中度或扩张经济性。
[CU001, CU025, CU026, CU038]6.6 附录
07风险
7.1 技术与 AI 模型风险
Periodic Labs 的核心论点建立在一串技术假设上,每一环都有重大失败风险。最根本的是合成缺口:包括 DeepMind GNoME 等前代系统在内,AI 模型经常预测出在物理世界中很难或无法实现的结构。GNoME 预测了 2.2 million 个晶体结构;其中只有大约 380,000 个被认为潜在稳定,截至 2025 年底,只有约 736 个得到独立合成和验证。Periodic Labs 的自主系统会面对同样约束——数字预测并不保证物理可实现,瓶颈已经从模型推理转移到实验室验证。第二个技术风险是 AI 幻觉。2025 年 4 月发表于 arXiv 的学术研究区分了良性模型错误和「腐蚀性幻觉」——这类输出科学上看似合理,但事实错误且难以识别。在 AI 输出直接驱动机器人实验排队的闭环自主实验室中,腐蚀性幻觉可能在人工审查发现前穿过多个实验周期,浪费昂贵试剂,并为下游模型再训练生成误导性数据。底层数据质量是第三个担忧。主动学习循环依赖密度泛函理论计算提供真值标签;系统性 DFT 偏差会把错误传入迭代训练周期。物理实验也无法像数字 AI 训练那样以接近零边际成本扩展,限制了投资人可能期待的迭代速度。随着系统自动化程度提高而逐步减少人在环监督,会进一步压缩发现失败的窗口。 [CR001, CR002, CR003, CR004, CR005, CR006]
| 风险类别 | 机制 | 证据 | 概率 | Periodic Labs 暴露 |
|---|---|---|---|---|
| 合成缺口 | AI 预测热力学稳定的晶体结构,但在现实实验室条件下无法合成 | GNoME:2.2M 个预测,截至 2025 年末约 736 个被合成并验证 | 高 | 吞吐量的核心约束;自主机器人无法跨越根本性的合成障碍 |
| 腐蚀性幻觉 | 模型生成科学上看似合理、事实却错误的输出,并逃过标准错误检测 | arXiv 2025 年 4 月:幻觉分类研究发现,AI 生成科学文本中的幻觉比例会影响下游实验设计 | 中高 | 闭环自主实验会在人工复核之前放大传播 |
| DFT 训练数据偏差 | 主动学习所用密度泛函理论标签带有系统性近似误差 | 计算材料科学文献已充分记录 DFT 近似误差 | 中 | 模型重训练循环会继承系统性误差;重新校准很耗资源 |
| 可复现性失败 | AI 预测的合成路线无法在不同实验室设备或条件下复现 | 多篇 GNoME 后续批评指出缺少独立实验验证 | 中高 | 客户验证和 IP 申报都依赖可复现性;单一实验室合成不够 |
| 自主范围漂移 | 人类监督减少后,AI 会选择预定发现领域之外的实验方向 | 自主智能体安全文献记录了这一概念风险;未报告 Periodic Labs 直接事故 | 中低 | 如果实验设计流程缺少双用途筛查,可能生成双用途化学品 |
概率估计是基于既有技术的定性评估;目前没有 Periodic Labs 专属事故数据。
[CR001, CR003, CR004, CR005, CR006, CR007]7.2 监管、法律与安全风险
Periodic Labs 运营自主机器人实验室,处理反应性、新型且可能有害的化学化合物。这会立即触发 OSHA Laboratory Standard(29 CFR 1910.1450)以及其 2024 Hazard Communication Standard 更新下的合规义务;该更新与 GHS 第 7 版对齐,要求所有被处理化学品满足新的标签和 SDS 要求,包括毒性特征未知的 AI 生成新候选物。OSHA 机器人标准要求所有自动化设备满足 lockout/tagout 要求,2025 年 9 月发布的 ANSI/A3 R15.06-2025 也要求每一个新增或修改后的机器人流程都要有记录化风险评估。OSHA 事故数据库记录了 2024 年 2 月一起员工被机器人手臂压死的事故,说明即便成熟工业机器人部署也可能造成致命后果。双重用途暴露带来性质不同的监管风险。2022 年一项实验记录显示,AI 化学工具在关闭安全过滤器后,数小时内可生成数千种潜在可武器化分子。CSIS、RAND 和 US National Security Commission on Emerging Biotechnology 都已指出,自主 AI 实验室是新兴生物安全风险来源,而现有治理框架不足以应对。EU AI Act(2024 amendment)将 AI 驱动的实验室平台归为高风险。Arms Control Association 2025 年 11 月报道称,台式合成监管缺口会制造结构性生物安全漏洞,与自主材料实验高度类似。若监管机构对自主实验室 AI 采取执法行动——CSIS 和 RAND 都主张这么做——公司可能被要求停运或承担当前未预算的高成本许可制度。IP 和法律风险是第三个维度。AI 生成材料发现的发明人身份在 USPTO 和 EPO 都未解决。训练数据诉讼风险正在升级;如果新材料与战略国防应用交叉,还可能触发出口管制义务。 [CR009, CR010, CR011, CR012, CR013, CR014]
| 监管领域 | 具体义务 | Periodic Labs 暴露 | 合规状态 | 风险级别 |
|---|---|---|---|---|
| OSHA Lab Standard(29 CFR 1910.1450) | 所有处理的新型化合物都需要化学卫生计划、PPE、通风橱和培训 | AI 生成的新候选材料毒性未知;缺少毒理数据时必须按危险品处理 | 未公开披露;推定进行中 | 高 |
| OSHA HCS 2024 / GHS Rev 7 | 2024 年起更新所有危险化学品的 SDS 和标签;2026 年前全面合规 | 自主合成新材料会生成没有既有 SDS 数据的化合物 | 未见公开合规声明 | 高 |
| OSHA Lockout/Tagout(29 CFR 1910.147) | 所有机械臂和自动化设备维护都需要能源控制程序 | 拥有多套机器人系统的自主实验室需要成文 LOTO 计划 | 未披露 | 中 |
| ANSI/A3 R15.06-2025 | 每个新增或修改的机器人流程前都必须完成成文风险评估;2025 年 9 月生效 | AI 持续改变实验变量,意味着机器人配置会频繁更新 | 未披露;新标准截至 2025 年 9 月生效 | 高 |
| 双重用途 / 生物安全 | NSCEB、CSIS 和 EU AI Act 对自主 AI 实验室平台的高风险分类 | 未披露双重用途筛查规程;CSIS 和 RAND 已识别结构性缺口 | 未披露政策 | 严重 |
合规状态仅反映公开披露;私人文档可能存在。NSCEB 和 EU AI Act 高风险 AI 条款截至 runDate 已生效;更多实施细则仍待出台。
[CR009, CR010, CR013, CR014, CR015]7.3 资本、商业与竞争风险
Periodic Labs 于 2025 年 9 月完成的 $300 million 种子轮,是科学 AI 初创公司史上最大的种子轮之一。资本带来异常充足的现金跑道,但也同时把公司锁进 VC 规模回报预期,而这种预期与材料科学商业化时间线在结构上错位。历史上,突破性材料——包括 Periodic Labs 瞄准的超导体——从初始发现到商业可用产品需要 10 到 15 年。第三方财务分析估计,考虑自主实验室基础设施、来自 OpenAI 和 DeepMind 的顶尖研究人才以及 GPU 计算需求,年运营成本为 $50-75 million,意味着如果没有商业收入流,$300M 种子轮可能在四到六年内耗尽。MIT Technology Review 2025 年 12 月指出,AI 材料发现初创公司尚未跨过实验室到市场的转化门槛,而这是证明 VC 规模回报假设所必需的;PitchBook 2026 年分析也称该品类通往 VC 回报是一条「曲折道路」。竞争压力进一步压缩商业可选项。Lila Sciences 完成 $550M 融资,CuspAI 融资 $154M,Microsoft 于 2025 年 1 月以 MIT 开源许可证发布 MatterGen,使 AI 预测层商品化。Google DeepMind 的 GNoME 正在公开发表结果。Orbital Industries 于 2026 年获得 $50M Series B,正在推进 AI 加制造的垂直整合模式。投资人集中还制造额外金融风险向量:Andreessen Horowitz 和 Nvidia 是锚定投资人,而 Nvidia 同时持有 Orbital Industries 的组合敞口,可能在未来融资或合作讨论中形成潜在利益冲突。如果公司寻求 Series A 资本,而早期里程碑让任何锚定投资人失望,逆向选择动态可能加速团队流失并拖延商业时间线。WEF 分析强调,把 AI 预测转化为可上市材料,需要监管批准、制造验证和客户采用步骤;Periodic Labs 尚未公开披露如何处理这些步骤。 [CR020, CR021, CR022, CR023, CR024, CR025]
| 实体 | 类型 | 融资 / 资源 | 竞争重叠 | 依赖 / 冲突风险 |
|---|---|---|---|---|
| Lila Sciences | 直接竞争对手 — AI 自主生物 / 化学实验室 | 已融资 $550M(2025) | 高:自主实验室平台,投资人层级相同 | 未披露;有人才竞争风险 |
| CuspAI | 直接竞争对手 — AI 驱动材料发现 | 已融资 $154M | 高:材料预测和合成自动化 | 未披露 |
| Microsoft / MatterGen | 拥有开源模型的大型科技竞争对手 | 2025 年 1 月按 MIT License 发布;Microsoft 内部研发预算 | 高:MatterGen 让 Periodic Labs 依赖的 AI 预测层商品化 | 无;但开源可得性会抹掉单靠预测建立的护城河 |
| Google DeepMind / GNoME | 前身平台 / 持续竞争对手 | Alphabet 研发;GNoME 仍在发表论文 | 高:创始人 Cubuk 参与打造 GNoME;DeepMind 继续发表成果 | 如果复用 GNoME 训练数据,可能引发 IP 纠葛 |
| Nvidia(投资人 + 生态) | 锚定投资人,同时持有冲突组合敞口 | Periodic Labs 锚定投资人;也持有 Orbital Industries 敞口 | 中:Periodic Labs 依赖 Nvidia GPU 供应;该投资人也资助竞争对手 | 后续融资轮存在利益冲突;优惠 GPU 定价可能撤回 |
融资数字来自公开披露和 PitchBook;可能未反映最新轮次。
[CR025, CR026, CR027, CR041, CR044]7.4 治理、执行与缓释框架
Periodic Labs 的创始团队由 Ekin Dogus Cubuk(Google DeepMind 的 GNoME 架构师)和 Liam Fedus(前 OpenAI 研究副总裁)组成,具备极强科学与商业信誉。但这种集中依赖也是重大关键人风险:任一创始人离开,都会削弱公司吸引顶级科学人才、维持投资人信心和执行高难度研究路线图的能力。公开信息中没有披露继任计划或领导层厚度。董事会 VC 色彩重——a16z、Nvidia 以及 Bezos 关联方——可能会施压公司更快拿出商业证明,牺牲有条理的科学验证。治理风险还延伸到自主系统监督。截至报告运行日,Periodic Labs 未披露 AI 治理政策、双重用途筛查协议,也没有披露面向自主实验室运营的 AI 安全框架。在 CSIS、RAND 和 Nature Biotechnology 都呼吁对实验室场景中的生成式 AI 工具强制内置生物安全保障的环境下,这会带来声誉和监管暴露。即使公共事故并非 Periodic Labs 直接造成,只要涉及 AI 化学工具,也可能触发全行业监管审查并扰乱运营。Periodic Labs 及其投资人可采取的缓释措施包括:采用 ANSI/A3 R15.06-2025 机器人安全协议并记录风险评估;在 AI 实验设计管线中嵌入双重用途筛查;对所有机器人控制系统做网络安全监测;在规模化合成前设置强制人工监督检查点。投资人应监控的终止标准包括:(1)18 个月内未能合成并独立验证任何 AI 预测材料;(2)任何针对自主实验室运营的联邦监管执法行动;(3)首个商业里程碑前任一联合创始人离职;(4)在没有披露商业合作或许可协议的情况下消耗超过 75% 已融资资本;(5)任何涉及双重用途或不安全自主合成的公开事件。 [CR028, CR029, CR030, CR031, CR032, CR033]
| 风险因素 | 描述 | 当前缓释 | 剩余风险 | 建议行动 |
|---|---|---|---|---|
| 联合创始人离职(Cubuk) | Ekin Dogus Cubuk 是 GNoME 架构师;他的网络和信誉是科研招聘与投资人信心的核心 | 未披露;没有公开接班计划 | 关键 — 商业化前阶段若创始人离职,公司可能撑不住 | 谈判留任股权并设置 4 年悬崖期;搭建自主研究管理层 |
| 联合创始人离职(Fedus) | Liam Fedus 曾任 OpenAI 研究副总裁;带来商业和合作伙伴信誉 | 未披露 | 高 — 失去他会削弱 Series A 融资和企业合作管线 | 聘请科学副总裁和商务副总裁作为冗余;联合创始人股权归属绑定竞业限制 |
| VC 董事会压力 | a16z、Nvidia 和 Bezos 关联投资人对回报时间表的期待,与材料科学 10-15 年周期不匹配 | 未公开披露;董事会构成尚未完全公开 | 中 — 过早转向商业里程碑可能牺牲科学严谨性 | 谈判把里程碑定义绑定科学进展,而非市场收入 |
| 缺失 AI 治理政策 | 未披露公开 AI 治理框架、双用途筛查协议或安全政策 | 未披露 | 高 — 监管或声誉事故可能触发运营停摆或投资人撤资 | 发布 AI 治理政策;采用 CSET 或 RAND 建议的生物安全筛查 |
| 机器人系统网络安全 | 自主机器人实验室是联网目标;对抗性提示注入或数据投毒可能破坏实验完整性 | 未披露 | 中 — 行业级风险;未报告 Periodic Labs 事故 | 为合成系统部署物理隔离控制;开展第三方渗透测试 |
风险评估仅反映公开披露审阅;私有文件可能缓释剩余风险评分。
[CR028, CR029, CR032, CR034, CR035]| 标准 | 阈值 | 理由 | 监控来源 | 优先级 |
|---|---|---|---|---|
| 合成验证失败 | 运营启动后 18 个月内没有独立验证的 AI 预测材料 | 核心价值主张是物理发现,不是数字预测;失败会推翻投资逻辑 | 同行评议论文、专利申请 | 关键 |
| 监管执法行动 | 任何针对自主实验室运营的联邦(OSHA、EPA、DHS)执法通知或停工令 | 运营停摆或同意令会把商业化时间表至少重置 2-3 年 | OSHA 检查记录、监管机构数据库 | 关键 |
| 联合创始人离职 | 首份商业许可协议或 Series A 完成前,任一联合创始人离职 | 关键人物风险;见 TR004 — 公司信誉和团队吸引力依赖联合创始人 | LinkedIn、新闻稿、SEC 文件(如公开) | 高 |
| 资本消耗率 | 在未披露商业合作或许可协议的情况下,已消耗融资资本超过 75% | $300M 按 $50-75M 年烧钱估计可支撑 4-6 年;未披露收入路径 | 投资人报告;新闻稿 | 高 |
| 公开双用途事故 | 任何危险化合物合成或 AI 生成涉武器输出的确认报告 | 行业级监管反应会施加超出 Periodic Labs 控制的运营限制 | CSB 报告、学术文献、监管公告 | 高 |
阈值是投资人复核的示例触发器,并非有约束力的合同条款;列出的监控来源仅作指示。
[CR020, CR021, CR022, CR036, CR046]08估值
8.1 投资逻辑与反向逻辑
Periodic Labs 建立在一个可防守且时点正确的论点上:AI 进步的前沿已经从文本和代码转向物理实验,能够打造自驱动实验室——自主猜想、实验和学习的系统——的初创公司,将拥有下一代真实世界训练数据。创始人 Liam Fedus(前 OpenAI 研究副总裁、ChatGPT 共同创造者)和 Ekin Dogus Cubuk(前 Google Brain 与 DeepMind 材料科学负责人、里程碑式 GNoME 晶体发现论文共同作者)履历清晰有力。团队有 20 多名来自 Meta、OpenAI 和 DeepMind 的研究人员,许多人放弃了可观股权包加入;这是任何 AI 垂直中最精英的创始阵容之一。公司已经与一家受芯片散热困扰的半导体制造商展开早期商业合作,这是一个可触摸的证明点,也让 Periodic 区别于没有收入的纯研究同行。 反向逻辑有三根支柱。第一,公司没有公开披露收入基础,任何收入倍数锚点都无法建立;拟议 $7.5B 估值完全建立在团队质量、战略叙事和早期信号上。第二,不到八个月估值跳涨六倍,意味着投资人情绪已经跑在可识别运营里程碑之前——从 2025 年 9 月种子轮到 2026 年中轮次讨论之间,公开信息中没有产品、客户规模或实验室基础设施的阶跃变化。第三,工业规模自主实验室系统仍未经技术验证;截至 2025 年末,Periodic 的机器人自动化层尚未运行,从概念验证走到可靠生产实验室的时间线存在重大不确定性。投资含义是观察建议:团队值得持续覆盖,但若要在 $7.5B 估值投入新资本,应以商业规模里程碑和机器人实验室上线证据为前提。[CV001, CV010, CV011, CV031, CV032, CV033]
| 类型 | 论点 | 证据强度 | 什么会改变判断 |
|---|---|---|---|
| 投资逻辑 | 世界级创始人(OpenAI 副总裁 + DeepMind 材料负责人)履历高度相关,正在搭建结构性独特的数据资产 | 高 | 无需改变;基础团队就是差异化因素 |
| 投资逻辑 | 自主实验室形成自有实验数据闭环,能滚出纯文本 AI 竞争者拿不到的复利护城河 | 中 | 竞争对手复制实验室基础设施会削弱护城河优势 |
| 投资逻辑 | 与半导体制造商的早期商业牵引验证近期产品市场匹配 | 中 | 收入爬坡且客户从 1 家扩到 3 家以上,将上调至买入 |
| 反向逻辑 | 没有披露收入却给到 $7.5B 估值,在任何近期 ARR 情景下都意味着 150x+ EV/Revenue,提前计入数十年的发现价值 | 高 | 收入证据和更低进入价格(相对估值纪律)会化解高估担忧 |
| 反向逻辑 | 截至 2025 年底,自主机器人实验室尚未完全投入运营;商业规模下的技术交付时间表仍不确定 | 高 | 确认实验室上线,并拿出可复现的实验发现结果,将大幅缓释该风险 |
| 反向逻辑 | 不到 8 个月估值跃升六倍,超过任何可识别的运营或产品里程碑,说明市场热度跑在基本面前面 | 中 | 商业规模、平台扩张或发现突破的证据,将为估值台阶式上升提供支撑 |
证据强度反映基于公开报道的分析评估。「高」信心代表有一手来源支持的主张;「中」代表第三方报道但一手佐证有限的主张。
[CV031, CV032, CV033, CV036, CV037, CV038]从市场规模、团队深度、产品差异化和早期牵引力,经由估值风险,推导出观察建议的逻辑链。
[CV005, CV031, CV036]8.2 估值背景、可比公司与情景
Periodic Labs 2025 年 9 月种子轮投后估值 $1.3B,对一家尚未有收入的深科技初创公司已经偏激进。若 2026 年拟议轮次按 $7.5B 推进,公司估值将高于 Sakana AI 据报道 $2.65B 的 Series B 标记(2025 年 11 月),高于 Pathos AI $1.6B 的 Series D 估值,也高于所有已披露的私有 AI 科学可比公司。只有 Isomorphic Labs 的融资规模接近:它带着 DeepMind 的 AlphaFold 这项已被验证的先例进入市场,与 Eli Lilly 和 Novartis 有活跃药企合作,潜在里程碑付款最高 $3B,并在 2025 年 3 月完成首笔外部融资 $600M (Series A),随后在 2026 年 5 月完成 $2.1B Series B;但公司未披露投后估值。 Finro 2026 年一季度覆盖 575 家 AI 公司的数据给出市场参照。LLM 供应商的 EV/Revenue 中位数为 39.5x(平均 73.5x),种子阶段 AI 公司的中位数为 20.2x,Data Intelligence 公司的中位数为 14.3x。 Finerva 数据显示,上市 Robotics and AI 公司的 EV/Revenue 中位数只有 3.4x,低于 2021 年高点的 6x。若估值 $7.5B、假设 ARR 为 $50M,隐含倍数将超过 150x——高于 LLM 供应商类别中位数,说明估值完全锚定长期发现经济性,而不是近期收入。私有 AI 市场相对上市可比公司的溢价真实且持续存在,但按 Periodic 的隐含倍数看,即便以 2026 年标准衡量,这一差距也异常大。 这里按 5 年退出窗口(2031 年)建模三种情景。乐观情景假设公司凭借在半导体、电池和特种材料垂直领域的平台主导地位实现 $500M+ ARR,并以 20-30x 倍数退出,估值 $12-20B。基准情景预计 ARR 为 $150-250M,有 2-3 个合作伙伴,自动化实验室到 2027 年投入运行,对应退出估值 $5.5-10B——相对 $7.5B 入场价大致持平到小幅回报。悲观情景假设自动化实验室延迟到 2028 年以后,收入规模未能兑现,倍数压缩到 10-15x,估值低于当前标记。Gartner 2026 年 4 月预测,2026 年全球半导体收入将超过 $1.3T(同比增长 64%,AI 芯片占总量 30%),这验证了可解决问题的规模;但 Periodic 近期能吃下多少市场,仍取决于尚未去风险的实验室落地和商业化执行。[CV005, CV006, CV008, CV009, CV012, CV013]
| 情景 | 关键假设 | 隐含 2031 年退出估值(USD M) | 从 $7.5B 进入的隐含回报 | 关键执行风险 | 概率信号 |
|---|---|---|---|---|---|
| 乐观 | 自主实验室在 2026 年 H2 投入运营;到 2031 年在半导体、电池和特种材料领域实现 $500M+ ARR;2 家以上超大规模云厂商伙伴认可平台领导力;按 25-35x EV/Revenue 退出 | 12,000–20,000 | 1.6x–2.7x | 超大规模云厂商内部竞争和倍数压缩 | 超额认购和快速跟进轮谈判;管线中有两项超大规模云厂商投资或合作 |
| 基准 | 到 2031 年在 2-3 个垂直领域实现 $150–250M ARR;自主实验室在 2027 年上线;1-2 个战略合作;倍数较 2026 年高点压缩;按 25-40x EV/Revenue 退出 | 5,500–9,000 | 0.7x–1.2x | 收入爬坡慢于模型;下行市场重置估值标记 | 早期半导体客户牵引;Finro/Finerva 2026 年 AI 板块中位倍数 |
| 悲观 | 自主实验室运营推迟到 2028 年之后;半导体试点未能转化为更广泛收入;AI 倍数压缩至 10-15x;到 2031 年 ARR 低于 $50M | 500–2,000 | <0.3x | 实验室自动化技术失败,以及多个学期跨度的 AI 热潮周期压缩 | Finerva 数据显示公开机器人 / AI 中位 EV/Revenue 为 3.4x,相比私募溢价;历史 AI 投资泡沫模式 |
情景是基于分析师模型和公开市场背景的前瞻性估计,并非预测或保证。估值单位为百万美元。回报数字假设以 $7.5B 进入,不含业绩报酬、费用、后续轮次稀释和优先股堆叠影响。
[CV029, CV030, CV036, CV037, CV044]| 公司 | 轮次或交易 | 披露估值(USD M) | 资本或交易金额(USD M) | 领域与重点 | 相关性与局限 |
|---|---|---|---|---|---|
| Periodic Labs(种子轮) | 种子轮 — 2025 年 9 月 | 1,300 | 300 | AI 材料科学 / 自主实验室 | 标的公司基准;最新公开锚点 |
| Lila Sciences | Series A 轮(2025,多批次) | 未披露 | 500+ | AI 科学超级智能 / 实验室自动化(Flagship Pioneering) | 自主 AI 科学中最接近的可比公司;未披露估值锚点;由 Nvidia NVentures 出资 |
| Isomorphic Labs | Series A 轮 — 2025 年 3 月 | 未披露 | 600 | AI 药物设计 / AlphaFold 谱系 / 药企合作 | 相邻 AI 科学;$3B 药企里程碑承诺验证平台价值;Alphabet 支持 |
| Isomorphic Labs | Series B 轮 — 2026 年 5 月 | 未披露 | 2,100 | AI 药物设计 / 临床管线推进 | 迄今最大 AI 科学融资;确认机构资金在规模化阶段仍有胃口;未披露估值 |
| Sakana AI | Series B 轮 — 2025 年 11 月 | 2,650 | 135 | 自然启发 AI 模型 / 日本企业 AI | 已披露估值;资本强度更低;来自 MUFG 等企业客户的收入;区域重点限制直接可比性 |
| Pathos AI | Series D 轮 — 2025 年 5 月 | 1,600 | 365 | AI 肿瘤学发现 / 临床阶段资产 | 融资时已有收入;临床验证部分降低估值风险;成熟度高于 Periodic |
| Dotmatics(Siemens 收购) | 并购 — 2025 年 4 月 | 5,100(交易价格) | N/A | AI 科学研发软件(GraphPad Prism、SnapGene) | AI 科学软件的并购退出可比;已有收入;显示 AI 赋能科学工具可实现的战略退出价格 |
标为「未披露」的估值,指公司确认已融资但未披露投后估值。Isomorphic Labs 已融资外部资本 $2.7B,但未披露估值。Dotmatics 行是收购价格,不是股权估值。所有数字按报道口径以百万美元计;货币按当时汇率折算。访问日期 2026-06-10。
[CV001, CV011, CV012, CV013, CV014, CV015]在五个 ARR 情景下,展示以 $7.5B 进入时隐含的 EV/Revenue 倍数,并说明达到各收入里程碑需要多大倍数压缩。
ARR 情景为分析师估计,基于可比 AI 科学公司的爬坡速度;并非公司指引。EV/Revenue 倍数假设进入估值恒定为 $7.5B。Periodic Labs 尚未公开披露收入。
[CV025, CV028, CV036]Periodic Labs 在 5 年期(2031 年)的乐观、基准和悲观退出估值区间;每个情景对应自主实验室运营化、商业收入爬坡和 AI 倍数环境的不同假设。
所有数字均为百万美元。区间反映分析师基于同业基准和公开报道建模的情景;并非公司发布的指引。进入估值假设为 $7.5B。未反映后续轮次稀释。
[CV029, CV030, CV044]8.3 建议、尽调与退出准备
证据权重支持给出观察建议,但估值立场偏高。Periodic Labs 在团队、投资逻辑和市场规模上都明确达标。现阶段的否决因素是入场价格:在没有公开收入锚点的情况下按 $7.5B 入场,回报曲线要求公司近乎完美执行,晚进入的资本才可能拿到符合创投要求的收益。如果投资逻辑兑现,早期按 $1.3B 进入的种子轮投资人位置仍然很好。新资本按 $7.5B 进入,门槛显著更高:要拿到 3x 回报,就需要以 $22.5B 退出,这要么靠公司在 AI 科学赛道取得品类主导地位,要么靠超大规模云厂商以显著高于今日隐含价值的价格战略收购。 2026 年轮次超额认购,以及市场已在讨论更高估值的快速跟投轮,都是正面的流动性信号,也同时是风险信号。在高价格下投资者需求超过额度,可能反映坚定信念,也可能是羊群效应——2026 年 AI 科学环境里两者都存在。SEC Form D 证据(CIK 0002122824,2026-05-29 提交)显示,一个由 Sydecar 管理的微型 SPV(AGC Wealt Periodic Labs I)从 7 名投资者募集 $4.7M,说明部分资本并非通过直接机构持仓进入,而是经由面向散户可触达的共同投资工具进入;即便在私有市场,这种模式有时也会出现在风险转移动态之前。 2025-2026 年 AI 科学 M&A 环境提供了可信的退出路径:Siemens 在 2025 年以 $5.1B 收购 Dotmatics,超大规模云厂商正在扩建 AI 科学部门(OpenAI 的「OpenAI for Science」部门),Isomorphic Labs $2.1B 的 Series B 也确认机构愿意为规模化 AI 发现平台下注。如果 Periodic 在一两个材料垂直领域实现经过商业验证的发现,5 年内以 $10-15B 被战略收购是有可能的。IPO 路径需要 7-10 年,前提是公司达到公开市场能接受的收入规模。若要按 $7.5B 估值进入,关键尽调问题应集中在商业收入基线、自动化实验室上线状态、已发现材料的 IP 归属、股权结构表和优先股堆叠细节,以及相对超大规模云厂商实验室项目的竞争差异化能持续多久。[CV002, CV003, CV007, CV021, CV022, CV035]
| 维度 | 评估 | 投资含义 |
|---|---|---|
| 总体建议 | 观察 | 在 $7.5B 估值进入前,先监控商业规模里程碑;种子轮投资人应继续持有 |
| 信心 | 中 | 基于二手报道;未直接接触财务、股权结构表或产品路线图 |
| 风险评级 | 高 | 尚无收入,自主实验室未验证,8 个月估值跳涨六倍,未披露收入锚点 |
| 估值立场 | 偏高 | 在任何合理的近期 ARR 假设下,$7.5B 都意味着 150x+ EV/Revenue;估值押注长期发现期权,而非当前证据点 |
这是截至 2026-06-10、仅基于公开二手报道的快照评估;分析师未直接接触 Periodic Labs 财务、数据室或股权结构表细节。建议对价格和证据都敏感。
[CV005, CV009, CV036, CV037]| 触发事件 | 阈值或条件 | 对投资逻辑的传导 | 行动含义 |
|---|---|---|---|
| 自主实验室未按计划上线 | 到 2027 年 Q3(假设 Series A 完成后 18 个月)仍无商业规模机器人实验室投入运营 | 核心差异化 —— 自有实验数据闭环 —— 没有落地;投资逻辑退化为团队和叙事押注 | 如果里程碑较既定目标延后超过 12 个月,下调至回避 |
| Series A 或后续下行轮 | 新融资估值低于 $5B(较拟议 $7.5B 折价 33%) | 投资人信心流失;人才留存和股权结构表复杂度风险上升;信号显示发现经济性弱于模型 | 重新评估股权结构表稀释;监控优先股堆叠对普通股回报的影响 |
| 客户集中 | 两个或更少客户贡献已披露收入的 >70% | 客户集中风险超过分散化逻辑;任何合同流失都会形成重大收入断崖 | 升级尽调;追加部署前要求多年合同可见性和管线证据 |
| 超大规模云厂商进入实验室自动化 | Google DeepMind、OpenAI 或 Microsoft 推出具备规模资源的可信自主实验室能力 | Periodic 的基础设施护城河被显著侵蚀;竞争优势收窄到既有数据资产和品牌 | 重新评估回报曲线与战略收购概率;超大规模云厂商 M&A 可能成为上行触发器 |
| 关键创始人离职 | Fedus 或 Cubuk 在 Series A 完成后 24 个月内退出 | 人才磁吸力和研究方向连续性承压;20 多名顶尖研究人员的招聘建立在创始人声誉之上 | 立即复核持仓;在评估新领导层前,可能从持有下调至观察 |
触发阈值是分析师构建的监控指标,不是公司披露的契约。所有触发器在行动前都需要实地尽调验证;阈值可能需根据尚未公开的 Series A 条款调整。
[CV037, CV038, CV048, CV049]| 主题 | 缺失证据 | 重要性 | 负责人或尽调路径 |
|---|---|---|---|
| 商业收入基线 | 未公开披露 ARR、合同数量或收入结构;半导体合作条款未披露 | 整个乐观 / 基准估值案例都依赖收入爬坡;没有锚点,150x+ 隐含倍数就无法证伪 | 公司管理层;直接数据室访问;经审计财务或商业伙伴 LOI |
| 自主实验室上线状态和时间表 | 没有确认的里程碑日期;TechCrunch 报道称截至 2025 年底机器人尚未运行 | 自有数据生成护城河是核心逻辑;它尚未在商业规模验证,时间表也不确定 | 技术尽调;实地走访 Menlo Park 设施;与 CTO 审阅里程碑计划 |
| 股权结构表和优先股堆叠 | Series A 条款、优先股堆叠和反稀释条款未公开披露;参与权未知 | 在平轮或下行轮退出情景下,优先权悬置可能显著压低普通股回报 | 法律尽调;由公司法律顾问提供股权结构表模型;在投资条款清单中谈判投资人权利 |
| 已发现材料的 IP 所有权 | 未公开披露 AI 发现材料的 IP 转让、专利策略或许可模式 | 收入模型和长期护城河取决于 Periodic 是否拥有或许可已发现化合物和材料的 IP | IP 审计;审阅实验室记录本、NDA 范围以及与发明人的专利转让协议 |
| 治理和创始人控制权 | 投票权、董事会构成、保护性条款和信息权未公开披露 | 高估值会放大治理风险;创始人控制权和投资人保护影响战略可选性 | 审阅投资条款清单;确认董事会结构、拖售权和强制报告要求 |
| 自主科学竞争格局 | 没有独立基准测试对比 Periodic 实验室能力与 DeepMind A-Lab、GNoME 或学术自主化学实验室 | 如果机构发表的自主实验室工作复制 Periodic 路线且不需要自有商业化,护城河可能被侵蚀 | 外部技术专家审阅;咨询学术顾问;对发现产率做可比基准测试 |
尽调问题仅反映可从公开报道识别的缺口。获得数据室访问后,还会浮现更多重大尽调事项。优先级顺序反映其对回报路径清晰度的相对影响。
[CV032, CV033, CV036, CV038, CV039, CV049]面向投委会的投资 KPI 记分卡,覆盖七个维度:市场机会、团队质量、产品差异化、商业证明、财务可见性、估值吸引力和证据质量。
分数为分析师基于截至 2026-06-10 的公开证据,在 1-10 标尺上的评估。财务可见性分数反映收入未披露;估值吸引力分数因 $7.5B 进入且无收入锚点而被扣分。
[CV031, CV032, CV036, CV037]8.4 附录
免责声明
本报告是基于截至 2026-06-10 公开信息生成的 AI 辅助尽调摘要,不构成投资建议。Periodic Labs 是一家私营公司,披露有限;收入、客户数、定价以及据报道 2026 年融资状态等多项关键指标不可得,或基于二手报道而非经审计的公司文件。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | Periodic Labs was co-founded in March 2025 in San Francisco California by Liam Fedus and Ekin Dogus Cubuk. | 高 | SO006, SO025, SO002 |
| CO002 | Periodic Labs emerged from stealth on September 30 2025 simultaneously announcing a $300 million seed round reported as the largest disclosed seed round in venture-capital history at that time. | 高 | SO001, SO004, SO008 |
| CO003 | The $300 million seed round was led by Andreessen Horowitz (a16z) with Felicis Ventures cutting the first institutional check. | 高 | SO004, SO002, SO017 |
| CO004 | OpenAI did not invest in Periodic Labs; despite initial signals in Fedus departure tweet the founders confirmed to TechCrunch that OpenAI is not a backer. | 高 | SO002, SO021 |
| CO005 | Additional seed round investors included DST Global NVentures (NVIDIA venture arm) Accel and individual investors Jeff Bezos Eric Schmidt Jeff Dean and Elad Gil. | 高 | SO004, SO001, SO005 |
| CO006 | The seed round valued Periodic Labs at approximately $1.3 billion post-money with a pre-money valuation of approximately $1.0 billion. | 高 | SO001, SO004, SO025 |
| CO007 | Periodic Labs mission is to build AI scientists and autonomous robotic laboratories that can form hypotheses run physical experiments and iteratively discover new materials and scientific knowledge in the physical sciences. | 高 | SO005, SO001 |
| CO008 | Periodic Labs first commercial focus is discovering high-temperature superconductors that operate more efficiently than existing materials. | 中 | SO005, SO002 |
| CO009 | Periodic Labs is working with at least one unnamed semiconductor manufacturer to solve chip heat dissipation problems using custom AI agents trained on the manufacturer experimental data. | 中 | SO005, SO008, SO009 |
| CO010 | Periodic Labs customer base also includes companies in the space and defense sectors in addition to semiconductor customers. | 中 | SO008, SO014 |
| CO011 | Liam Fedus served as Vice President of Research for Post-Training at OpenAI from October 2024 through March 17 2025 when he announced his departure to found a materials science AI startup. | 高 | SO006, SO010, SO025 |
| CO012 | Fedus was a co-creator of ChatGPT and served as data-flywheel lead at OpenAI; he also led post-training research and development for GPT-4o o1-mini and o1-preview. | 高 | SO001, SO025, SO017 |
| CO013 | Liam Fedus holds a BS in physics from MIT (2010) an MS in physics from UC San Diego (2016) and a PhD in computer science from Universite de Montreal and MILA (2020) co-advised by Yoshua Bengio and Hugo Larochelle. | 中 | SO025, SO017 |
| CO014 | Liam Fedus serves as Chief Executive Officer of Periodic Labs. | 中 | SO025, SO003 |
| CO015 | Ekin Dogus Cubuk led the materials and chemistry research team at both Google Brain and Google DeepMind where he also founded the materials science research group. | 高 | SO001, SO017, SO013 |
| CO016 | Cubuk co-authored the 2023 GNoME paper that identified approximately 2.2 million novel stable crystal structures using AI one of the largest materials discovery results published to date. | 高 | SO001, SO017, SO024 |
| CO017 | Cubuk earned his PhD from Harvard University and completed a postdoc at Stanford University before his career at Google Brain and DeepMind. | 中 | SO017, SO025 |
| CO018 | Alexandre Passos a creator of OpenAI o1 and o3 reasoning models joined Periodic Labs as a senior researcher. | 中 | SO002, SO024 |
| CO019 | Eric Toberer a materials scientist who has made prior superconductor discoveries is a researcher at Periodic Labs. | 中 | SO002 |
| CO020 | Matt Horton a creator of Microsoft MatterGen and MatterSim generative materials science AI tools joined the Periodic Labs team. | 中 | SO002, SO015 |
| CO021 | More than twenty researchers from Meta OpenAI DeepMind Databricks and Samsung were recruited to Periodic Labs many foregoing substantial unvested equity to join. | 中 | SO003, SO014, SO002 |
| CO022 | Wilson Sonsini Goodrich and Rosati led by Yokum Taku Avi Emanuel MJ Han and Jinny Park advised Periodic Labs on the $300 million seed round transaction. | 中 | SO004 |
| CO023 | As of May 7 2026 Forbes reported Periodic Labs was in advanced talks to raise at least $500 million in a new funding round. | 高 | SO003, SO018, SO022 |
| CO024 | The reported 2026 follow-on round targets a $7.5 billion valuation representing approximately a 5.8-fold increase from the $1.3 billion seed valuation in under nine months. | 中 | SO003, SO013 |
| CO025 | Bloomberg first reported in March 2026 that Periodic Labs was in deal talks targeting approximately $7 billion in valuation for a new round. | 中 | SO022, SO013 |
| CO026 | The 2026 follow-on round is reportedly being led by AMP an investment vehicle founded by Anjney Midha a former general partner at Andreessen Horowitz. | 中 | SO003 |
| CO027 | The 2026 follow-on round was described as significantly oversubscribed with sources reporting discussions for a fast-follow additional round at an even higher valuation. | 中 | SO003 |
| CO028 | Periodic Labs product strategy centers on autonomous robotic labs generating proprietary experimental data including negative results forming a training corpus unavailable to competitors; each experimental run can produce gigabytes of data. | 中 | SO005, SO007, SO024 |
| CO029 | The company argues that frontier AI models have effectively exhausted the approximately ten trillion tokens of text data available on the internet as training material. | 中 | SO005, SO001, SO015 |
| CO030 | Cubuk co-published a 2023 paper demonstrating that a fully automated robotic lab (A-Lab) synthesized 41 novel compounds in 17 days using AI-generated recipes proving the feasibility of AI-driven autonomous laboratory paradigm. | 高 | SO002, SO024 |
| CO031 | Periodic Labs scientific advisory board is chaired by Nobel Chemistry laureate Carolyn Bertozzi of Stanford and includes authorities in superconducting physics and materials science from Stanford and MIT. | 中 | SO024, SO016 |
| CO032 | Periodic Labs is headquartered in San Francisco California. | 高 | SO003, SO005, SO001 |
| CO033 | Periodic Labs debuted on the Forbes AI 50 Brink list in 2026. | 中 | SO003 |
| CO034 | Periodic Labs headcount as of early 2026 is estimated at approximately 32 to 48 employees based on third-party business directories; Forbes reported more than 20 researchers specifically recruited from Meta OpenAI and DeepMind. | 低 | SO003, SO008 |
| CO035 | A 2025 Wiley/Futurism survey of working scientists found that the share believing AI surpasses human abilities dropped from over 50 percent in 2024 to under 33 percent in 2025 with 64 percent expressing concern about AI hallucinations up from 51 percent in 2024. | 中 | SO019 |
| CO036 | A 2025 peer-reviewed Nature study found that AI tools expand individual scientists output and impact but may simultaneously narrow the diversity and breadth of scientific inquiry at the field level. | 中 | SO020, SO019 |
| CO037 | Yann LeCun Meta Chief AI Scientist and Turing Award winner argues that current AI models engage in pattern-matching rather than genuine mental-model formation and therefore cannot perform the kind of original reasoning required for autonomous scientific discovery. | 中 | SO023 |
| CO038 | Felicis Ventures partner Peter Deng committed to invest in Periodic Labs before the company was even incorporated or had a name making Felicis the first institutional backer. | 高 | SO002, SO017 |
| CO039 | By October 2025 Periodic Labs had established its San Francisco laboratory with experimental data and simulations running though robotic systems were still being trained and not yet fully operational. | 中 | SO002 |
| CO040 | The founding concept of Periodic Labs arose from a conversation between Fedus and Cubuk approximately seven months before the September 2025 stealth launch around February 2025 when both recognized that robotic automation materials simulation and LLM reasoning had matured enough to build a genuine AI-science platform. | 中 | SO002, SO024 |
| CM001 | AI-driven materials discovery encompasses software platforms and integrated hardware-software systems that apply generative AI, graph neural networks, and large language models to propose, simulate, and experimentally validate novel material compositions, compressing discovery cycles from decades to months or years. | 中 | SM011, SM016 |
| CM002 | The AI materials discovery software segment—purpose-built platforms for discovery hypothesis generation and experimental design, distinct from general lab automation—is valued at approximately $970 million in 2026. | 中 | SM005, SM006 |
| CM003 | Adjacent spend pools representing Periodic Labs' upstream opportunity include the $8.83B total lab automation market (2026), the approximately $300B global pharmaceutical R&D budget, and comparable advanced-materials and chemicals R&D spend, creating a much larger budget pool than the narrow AI discovery software TAM alone. | 中 | SM009, SM019, SM022 |
| CM004 | Status-quo substitutes for AI-driven materials discovery include manual combinatorial synthesis, traditional computational chemistry, and contracted CRO or CDO services, all of which are measurably slower and more capital-intensive than AI-directed closed-loop systems. | 中 | SM011, SM014 |
| CM005 | The autonomous chemical laboratory market, which combines hardware robotics with AI control layers and represents a broader superset of the AI discovery software segment, is projected at approximately $5.75 billion in 2026 and growing at a 14.5% CAGR through 2035. | 中 | SM007, SM008 |
| CM006 | The AI in materials discovery market was approximately $740 million in 2025 and is forecast at $970 million in 2026 at a 30.3% CAGR, growing to approximately $2.77 billion by 2030, making it one of the fastest-growing software verticals in laboratory science. | 中 | SM005, SM006 |
| CM007 | The AI in lab automation market, a broader segment encompassing materials discovery, drug discovery, and general scientific automation, was $3.54 billion in 2025 and is projected at $4.19 billion in 2026, representing approximately 18.4% year-over-year growth. | 中 | SM008, SM009 |
| CM008 | The total lab automation market—hardware plus software plus AI—is valued at $8.03 billion in 2025 and projected at $8.83 billion in 2026, growing at approximately 9.9% annually; this represents the broadest sizing anchor for the laboratory technology stack. | 中 | SM009 |
| CM009 | By 2030, the AI in materials discovery segment alone is projected to reach approximately $2.77 billion assuming the 30.3% CAGR sustained from 2025–2026 is maintained, implying a roughly 3.7× increase in five years. | 中 | SM005, SM006 |
| CM010 | Autonomous lab robotics—the physical hardware infrastructure layer—was approximately $1.8 billion in 2025 and is growing at a projected 19.5% CAGR through 2033, confirming sustained investment in the physical substrate that AI discovery platforms depend on. | 低 | SM010 |
| CM011 | Global pharmaceutical R&D spending reached approximately $300 billion in 2025, representing one of the highest R&D-to-revenue intensity levels of any industry at roughly 18% of revenue, constituting the largest single accessible budget pool for AI-driven discovery platforms. | 高 | SM019, SM022 |
| CM012 | Total global R&D spending across all sectors reached approximately $2.53 trillion in 2025 according to WIPO estimates incorporating data from Eurostat, OECD, RICYT, and UNESCO UIS, establishing the macro context in which AI materials discovery tools compete for R&D budget allocation. | 高 | SM020, SM026 |
| CM013 | The top pharmaceutical companies—Roche, Merck, Pfizer, and Johnson & Johnson—are each projected to spend $10–15 billion on R&D in 2026, representing a high-value target customer pool for AI-accelerated discovery platforms. | 中 | SM019, SM022 |
| CM014 | Primary buyers of AI materials discovery platforms are materials scientists and R&D heads at pharmaceutical, battery and energy, semiconductor, and specialty chemicals companies, while payers are typically R&D division or business-unit budget owners operating with multi-year capital allocation cycles. | 中 | SM011, SM014 |
| CM015 | Pharmaceutical and biotechnology companies currently represent the largest deployment base for AI lab automation broadly, accounting for the highest share of autonomous lab deployments globally, driven by the urgency of drug discovery timelines and the maturity of existing laboratory infrastructure. | 中 | SM008, SM016 |
| CM016 | Asia-Pacific is the fastest-growing region for autonomous laboratory adoption, driven by China's 16%+ annual pharma R&D growth between 2020 and 2024 and state-backed advanced-materials investment programs targeting AI-driven research capabilities. | 中 | SM008, SM019 |
| CM017 | North America remains the largest single geographic market for AI materials discovery tools by absolute dollar value, hosting the majority of private VC investment in the autonomous science sector, including Periodic Labs' $300M seed and the in-progress $500M round. | 中 | SM015, SM026 |
| CM018 | Government agencies and academic research consortia represent a distinct second buyer tier, particularly for superconductor and advanced semiconductor research supported by national competitiveness programs in the US, EU, and China, with funding that is independent of enterprise R&D procurement cycles. | 中 | SM011, SM024 |
| CM019 | AI can reduce materials discovery-to-commercialization cycles from decades to approximately one to two years for many material innovations, representing the primary time-compression value proposition over traditional manual or computational methods. | 中 | SM011, SM024 |
| CM020 | DeepMind's GNoME model identified 2.2 million stable materials including 380,000 stable crystals using deep learning, demonstrating at scale that AI can generate validated candidate materials orders of magnitude faster than any human-led combinatorial approach. | 中 | SM029 |
| CM021 | Generative AI models including graph neural networks and transformer-based crystal structure generators are now capable of proposing novel material compositions with targeted electronic, thermal, or electrochemical properties faster than traditional combinatorial methods, reducing the in-silico screening phase from years to weeks or days. | 中 | SM011, SM013 |
| CM022 | SandboxAQ's AQVolt26 initiative demonstrates commercial deployment of AI models for solid-state battery materials discovery, with pre-trained model checkpoints publicly available on Hugging Face, signaling rising commercial maturity and accessible tooling in the battery vertical. | 中 | SM012 |
| CM023 | National competitiveness programs in the US (CHIPS Act, DOE materials mandates), EU (MPIE-led 33-partner AI battery consortium across 12 European countries), and China's advanced-materials plans create a sustained policy tailwind supplementing private-sector demand for AI materials discovery. | 中 | SM011, SM023 |
| CM024 | Major corporations in battery, semiconductor, and pharmaceutical sectors are partnering with or acquiring AI materials startups to secure next-generation capabilities, creating demand for platform licensing and collaboration agreements that extend beyond pure software subscriptions. | 中 | SM011, SM015 |
| CM025 | Only 29% of enterprises report achieving significant ROI from AI deployments as of 2026, indicating that AI materials discovery platforms must demonstrate concrete, measurable outcomes before enterprise R&D budget owners commit at scale; this structural hurdle extends early sales cycles. | 中 | SM018, SM025 |
| CM026 | Data quality, proprietary data ownership, and inconsistent experimental data formats are cited as the leading technical bottleneck for AI adoption in enterprise R&D, with materials science datasets particularly siloed, often non-standardized, and subject to complex IP agreements between partners and service providers. | 中 | SM017, SM018 |
| CM027 | AI regulation had expanded to 68 countries as of 2026, introducing mandatory audit requirements, explainability standards, and cross-border data flow restrictions that add compliance overhead to enterprise AI science deployments and extend regulatory review of novel AI-discovered materials. | 中 | SM017 |
| CM028 | Only approximately 25% of organizations had mature AI governance frameworks as of 2026 per Deloitte, creating institutional readiness risk for large-scale deployment of autonomous laboratory systems and adding enterprise risk committee scrutiny to procurement decisions. | 中 | SM018 |
| CM029 | The AI talent gap reached approximately 3.5 million unfilled roles worldwide by 2026, creating a workforce bottleneck that limits the internal capacity of potential customers to deploy, integrate, and maintain AI-driven discovery platforms without substantial vendor support. | 中 | SM017, SM025 |
| CM030 | Enterprise AI adoption is structurally slower than forecast because the implementation infrastructure needed to convert AI capability into production value consistently takes longer to build than the underlying technology itself, a recurring dynamic ComputeForecast identifies as independent of the specific AI capability wave. | 中 | SM021 |
| CM031 | Capital intensity of integrated autonomous laboratories—including robotics hardware, sensors, AI compute, and wet-lab facilities—creates a high initial investment barrier restricting the total addressable customer count and favoring large pharmaceutical companies and well-funded battery startups over smaller R&D organizations. | 中 | SM007, SM016 |
| CM032 | AI is compressing superconductor discovery cycles from decades to months using tools such as AtomGPT and MatterGPT for crystal structure generation, with commercial applications in quantum computing and power grid infrastructure potentially emerging from 2026 onward, though no commercially deployed room-temperature superconductor has yet been verified. | 中 | SM011, SM024 |
| CM033 | AI-accelerated solid-state battery research directly addresses critical near-term commercial needs in EV and grid storage markets; SandboxAQ's AQVolt26 and the MPIE-led EU consortium demonstrate that AI battery discovery is already in transition from research to early commercial pilots. | 中 | SM012, SM023 |
| CM034 | In the semiconductor vertical, AI platforms can now propose novel gallium alloys and two-dimensional materials with targeted electronic properties for advanced chip manufacturing and optoelectronics, with TechXplore reporting active pilot deployments as of May 2026. | 中 | SM013, SM011 |
| CM035 | Pharma-adjacent discovery tools for excipients, drug delivery scaffolds, and biocompatible materials represent a growing sub-segment of the $300 billion pharmaceutical R&D spend, with IQVIA's Global Trends in R&D 2026 confirming AI's growing role in pre-IND material screening workflows. | 中 | SM022, SM019 |
| CM036 | SaaS-model AI discovery platforms can achieve gross margins of 70–90%, substantially above the 30–50% margins of traditional CRO services, making the business model highly defensible at scale and justifying premium valuation multiples for platforms achieving commercial traction. | 中 | SM011, SM014 |
| CM037 | Periodic Labs raised a $300 million seed round in September 2025 at a $1.3 billion valuation, with investors including Andreessen Horowitz, Nvidia, Jeff Bezos, Accel, DST, Eric Schmidt, and Jeff Dean, establishing the company as the best-funded new entrant in AI-driven science. | 高 | SM001, SM028 |
| CM038 | As of May 2026, Periodic Labs was in advanced talks to raise an additional $500 million at a $7.5 billion valuation led by AMP (Anjney Midha), representing a nearly 6× valuation step-up in approximately eight months and reflecting exceptional investor conviction in the AI-driven materials discovery thesis. | 中 | SM002, SM003 |
| CM039 | The AI-driven science sector—including Periodic Labs and peer Lila Sciences—had collectively raised over $1.3 billion in venture funding by early 2026, reflecting exceptional VC conviction in the autonomous science factory thesis substantially ahead of verified commercial revenue. | 中 | SM015, SM001 |
| CM040 | The Royal Society's 2025 review of autonomous self-driving laboratories identifies the sector as transitioning from academic pilot projects to broader commercial adoption, validating the near-term market-entry timing of platforms like Periodic Labs. | 中 | SM016 |
| CM041 | Switching costs for enterprise R&D customers adopting AI discovery platforms are elevated due to proprietary training data lock-in, deep integration with existing ELN and LIMS systems, and scientific staff retraining requirements, creating defensible retention once initial deployments are established. | 中 | SM018, SM021 |
| CM042 | Process reimagination—not just tool substitution—is identified by Forbes as the primary success factor for enterprise AI adoption, requiring change management investment that extends customer onboarding timelines for AI science platforms beyond what pure technology readiness would suggest. | 中 | SM017, SM018 |
| CP001 | Periodic Labs emerged from stealth in September 2025 with a $300 million seed round at a $1.3 billion valuation, backed by Andreessen Horowitz, NVIDIA, Felicis, Accel, Jeff Bezos, Eric Schmidt, and Jeff Dean. | 高 | SP002, SP003, SP011 |
| CP002 | As of May 2026, Periodic Labs was in advanced talks to raise at least $500 million in a round led by AMP (Anjney Midha), targeting a $7.5 billion valuation — nearly a sixfold increase from its September 2025 valuation. | 高 | SP003, SP019, SP024 |
| CP003 | Periodic Labs was co-founded by Liam Fedus (former VP of Research at OpenAI, co-creator of ChatGPT) and Ekin Dogus Cubuk (former research scientist at Google DeepMind, who led GNoME and the materials science team). | 高 | SP002, SP012, SP021 |
| CP004 | CuspAI raised $130 million across seed and Series A rounds (Series A in September 2025 co-led by NEA and Temasek), valuing the company at approximately $520 million post-money at the time of the Series A. | 中 | SP004, SP020, SP027 |
| CP005 | In early 2026, CuspAI's informal valuation reached approximately $800 million based on new commercial contracts, and the company entered talks to raise $200 million or more at a valuation above $1 billion. | 中 | SP004, SP027 |
| CP006 | CuspAI describes its platform as 'a search engine for the material world' that accepts desired material properties (strength, conductivity, thermal tolerance) and returns AI-generated chemical compositions up to ten times faster than traditional methods. | 中 | SP004, SP020 |
| CP007 | CuspAI's commercial customers include Meta, Kemira, and Hyundai Motor Group, representing partnerships across semiconductors, specialty chemicals, and automotive sectors. | 中 | SP004, SP027 |
| CP008 | CuspAI's generative AI models are described as synthesis-aware — they propose materials that can actually be manufactured rather than only theoretically simulated, which is cited as a key differentiator from earlier computational methods. | 中 | SP004 |
| CP009 | Schrödinger reported Q1 2026 total revenue of $58.6 million and annualized contract value (ACV) of $28.4 million, representing 12% year-over-year ACV growth. | 高 | SP007, SP017 |
| CP010 | Schrödinger's Q1 2026 software revenue declined 21% year-over-year primarily due to a deliberate transition from upfront perpetual licenses to a hosted (cloud subscription) licensing model. | 高 | SP007, SP026 |
| CP011 | Schrödinger announced plans to launch 'Bunsen,' an agentic AI co-scientist designed for autonomous execution of complex molecular discovery workflows, in summer 2026. | 高 | SP007, SP017 |
| CP012 | Schrödinger held approximately $406 million in cash and marketable securities as of the end of Q1 2026. | 高 | SP007, SP017 |
| CP013 | Schrödinger's platform applies the same physics-based, AI-augmented simulation core to both pharmaceutical drug discovery and materials science applications. | 中 | SP007, SP017 |
| CP014 | Recursion Pharmaceuticals (NASDAQ: RXRX) was trading at approximately $3.20–$3.50 per share in mid-2026, down more than 90% from peak prices above $40 per share. | 中 | SP025 |
| CP015 | Recursion reported Q1 2026 revenue of $6.47 million, missing analyst expectations, and an estimated annualized net loss of approximately $560 million for 2026. | 高 | SP008, SP025 |
| CP016 | Recursion Pharmaceuticals held $665 million in cash as of Q1 2026, with management guidance indicating runway through at least early 2028 without additional financing. | 高 | SP008, SP025 |
| CP017 | Recursion's platform, Recursion OS, ingests over 50 petabytes of proprietary biological and chemical data, which the company cites as a core competitive moat. | 中 | SP008 |
| CP018 | Google DeepMind's GNoME AI tool predicted 2.2 million new crystal structures, of which 380,000 were assessed as highly stable and suitable for experimental synthesis. | 高 | SP012, SP016 |
| CP019 | GNoME was co-authored by Ekin Dogus Cubuk while at Google DeepMind; Cubuk is now co-founder and co-CEO of Periodic Labs, directly connecting Periodic's founding to the deepest public AI-materials research. | 高 | SP002, SP012, SP016 |
| CP020 | Over 736 GNoME-predicted materials were independently synthesized by external research labs worldwide, and the A-Lab at Lawrence Berkeley National Laboratory autonomously synthesized 41 of 58 proposed compounds in 17 days using GNoME data. | 高 | SP012, SP016 |
| CP021 | GNoME expanded the number of known stable inorganic crystals from approximately 48,000 to 421,000 — a roughly tenfold increase — and was released as open-access to the research community. | 高 | SP012, SP016 |
| CP022 | Citrine Informatics has raised approximately $81.3 million through 12 funding rounds including a Series C that closed in early 2025, backed by investors including Innovation Endeavors and Prelude Ventures. | 中 | SP015 |
| CP023 | Citrine Informatics' enterprise customers include LyondellBasell, Eastman, Panasonic, Michelin, and LANXESS, spanning specialty chemicals, coatings, and battery materials. | 中 | SP009, SP015 |
| CP024 | Citrine Informatics' platform is specifically optimized for handling small, sparse datasets typical of specialty chemicals and materials R&D — a niche capability that distinguishes it from general-purpose AI platforms. | 中 | SP009, SP015 |
| CP025 | Emerald Cloud Lab operates over 200 different instrument models accessible remotely via a single unified software interface (ECL Command Center), available 24 hours a day, 365 days a year. | 中 | SP010 |
| CP026 | Emerald Cloud Lab access can exceed $250,000 per year for comprehensive institutional accounts, pricing that is incompatible with standard academic grant budgets according to independent analysis. | 中 | SP014 |
| CP027 | The global laboratory robotics market was approximately $8.5 billion in 2025, with projections toward $18 billion by 2030 at a consensus CAGR of 7–9.4%. | 中 | SP014 |
| CP028 | The materials science/chemistry self-driving lab sub-segment was approximately $0.12 billion in 2025, with projected growth at approximately 40% CAGR to reach $0.65 billion by 2030. | 中 | SP014 |
| CP029 | As of April 2026, liquid handling and robotic workcell automation are assessed at TRL 8–9 and considered effectively commoditized; the value creation frontier has migrated to AI orchestration and scheduling software. | 中 | SP014 |
| CP030 | The global materials informatics market is projected to grow at over 20% CAGR, reaching more than $820 million by 2033, driven by demand to compress 10–20 year development timelines to 2–5 years. | 中 | SP015 |
| CP031 | Battery materials represent approximately 30% of the materials informatics market by value, followed by advanced polymers (20%) and catalysts (15%), with pharmaceutical materials and renewable energy as fastest-growing segments. | 中 | SP015 |
| CP032 | Microsoft's entry into AI-driven materials discovery through Azure Quantum Elements has been cited by market analysts as potentially disrupting smaller specialist players' market positions. | 中 | SP015 |
| CP033 | Schrödinger's platform benefits from deep physics-based simulation expertise and over three decades of established pharmaceutical-industry relationships, which create customer-switching friction for incumbent users. | 中 | SP007, SP017, SP026 |
| CP034 | Automata raised a $45 million Series C in January 2026 with Danaher Ventures as a strategic investor, linking the Danaher instrument portfolio (Beckman Coulter, Molecular Devices) to Automata's LINQ orchestration platform. | 中 | SP006 |
| CP035 | A Sapio Sciences survey of 150 scientists at SLAS 2026 found that 45% are using unauthorized shadow AI tools because their official platforms are not keeping pace with researcher needs. | 中 | SP006 |
| CP036 | Independent analysts characterize most commercially deployed autonomous labs in 2026 as operating at Level 2–3 autonomy (closed-loop optimization for specific, scripted experimental tasks), not the general-purpose scientific autonomy promoted in marketing materials. | 中 | SP005, SP014 |
| CP037 | Financial analysts flagged Periodic Labs' $300 million capital deployment risk as material, citing the absence of a clear commercial timeline for superconductor breakthroughs and the long monetization cycles typical of materials discovery. | 低 | SP023 |
| CP038 | In February 2026, OpenAI's GPT-5 model autonomously executed over 36,000 protein synthesis experiments in Ginkgo Bioworks' cloud lab, reducing sfGFP production costs by approximately 40%, demonstrating that AI-lab closed-loop execution is achievable outside dedicated materials startups. | 中 | SP006 |
| CP039 | At SLAS 2026 (February 2026 in Boston), 15 companies were identified as competing to become the standard orchestration layer for AI-enabled laboratories, including Biosero, Automata, Synthace, and UniteLabs. | 中 | SP006 |
| CP040 | Chemify's Chemifarm chemistry-as-code synthesis network has raised over $50 million and provides chemistry-as-a-service across multiple physical lab facilities, representing a modular alternative to Periodic's vertically-integrated autonomous lab model. | 中 | SP014 |
| CP041 | Periodic Labs co-founder Liam Fedus confirmed in the company's launch blog post that autonomous labs are central to Periodic's strategy because they provide 'huge amounts of high-quality data that exist nowhere else' and supply valuable negative results that are seldom published. | 中 | SP001 |
| CP042 | Periodic Labs has stated it is working with a semiconductor manufacturer facing heat-dissipation issues and is training custom agents for researchers to iterate faster on experimental data. | 中 | SP001, SP011 |
| CI001 | Periodic Labs raised a $300 million seed round in September 2025 at a $1.3 billion post-money valuation. | 高 | SI002, SI020, SI004 |
| CI002 | Andreessen Horowitz (a16z) led Periodic Labs' $300 million seed round. | 高 | SI002, SI004, SI007 |
| CI003 | Seed round institutional investors include Nvidia (via NVentures), DST Global, Accel, and Felicis. | 高 | SI002, SI010, SI025 |
| CI004 | Individual investors in the seed round include Jeff Bezos, Eric Schmidt, Jeff Dean, and Elad Gil. | 高 | SI002, SI014, SI007 |
| CI005 | As of May 2026, Periodic Labs is in advanced talks to raise a $500 million Series A at a $7.5 billion post-money valuation. | 高 | SI001, SI013, SI015 |
| CI006 | The Series A round is described by multiple sources as 'significantly oversubscribed.' | 中 | SI001, SI008 |
| CI007 | Sources report that discussions for a fast-follow round at an even higher valuation were underway concurrent with the Series A close. | 低 | SI001 |
| CI008 | The Series A is led by AMP, an investment vehicle founded by Anjney Midha, a former general partner at Andreessen Horowitz. | 中 | SI001, SI005 |
| CI009 | If the Series A closes at $7.5 billion, Periodic Labs' valuation will have increased nearly sixfold from its $1.3 billion seed in under eight months. | 中 | SI001, SI005 |
| CI010 | Periodic Labs had approximately 40 employees as of March 2026. | 中 | SI008, SI010 |
| CI011 | Periodic Labs generates revenue from commercial engagements with customers in the semiconductor, space, and defense industries. | 中 | SI004, SI005, SI014 |
| CI012 | One confirmed customer engagement involves training custom AI agents for a semiconductor manufacturer facing chip heat dissipation problems. | 高 | SI020, SI004, SI014 |
| CI013 | Periodic Labs has not publicly disclosed revenue figures, ARR, revenue run rate, or financial statements as of June 2026. | 高 | SI010, SI011, SI023 |
| CI014 | Periodic Labs' service pricing is bespoke, case-by-case, and not publicly disclosed. | 中 | SI011, SI010 |
| CI015 | The primary revenue model is contract-based AI-science services ('AI-lab-as-a-service'), engaging enterprise clients to accelerate specific materials R&D problems. | 中 | SI004, SI005, SI011 |
| CI016 | Secondary revenue paths under consideration include materials IP licensing, proprietary experimental data licensing, and direct commercialization of discovered materials. | 低 | SI011, SI012 |
| CI017 | Analyst estimates for Periodic Labs' monthly cash burn range from $5 million to $15 million, reflecting lab buildout, AI compute, and talent costs; no figure has been confirmed by the company. | 低 | SI003, SI010 |
| CI018 | Estimated operational runway from the $300M seed is 20–60 months from the September 2025 close, depending on actual burn rate realization. | 低 | SI003, SI010 |
| CI019 | Gross margin for Periodic Labs is not disclosed; the model structurally combines high-margin AI services revenue with high capital-intensity lab costs, making margin direction uncertain. | 低 | SI003, SI011 |
| CI020 | Industry analyst estimates place each fully autonomous materials discovery lab buildout at $10–50 million depending on scale and specialization. | 低 | SI003, SI010 |
| CI021 | Emerald Cloud Lab's own data shows initial instrumentation costs of $1.4M–$3.6M and annual maintenance of $288K–$720K for a standard automated chemistry facility. | 中 | SI016, SI017 |
| CI022 | Access-fee entry costs for comparable commercial cloud labs are $250K/year (Emerald Cloud Lab) and $100K+ per method (Strateos); Periodic Labs builds proprietary infrastructure at a substantially higher cost tier. | 高 | SI017, SI016 |
| CI023 | Periodic Labs' cost structure comprises four primary categories: autonomous lab capital equipment and maintenance, AI compute (GPU clusters), elite researcher talent compensation, and facilities and G&A. | 中 | SI004, SI003, SI010 |
| CI024 | The $300 million seed round was characterized as one of the largest VC seed rounds in history at the time of the September 2025 announcement. | 高 | SI002, SI012 |
| CI025 | Bloomberg reported in March 2026 that Periodic Labs was in deal talks at approximately $7 billion valuation, corroborating later Forbes reporting of $7.5 billion. | 中 | SI022, SI001 |
| CI026 | Sources indicate a fast-follow additional financing round at an even higher valuation was in discussion concurrent with the primary Series A. | 低 | SI001 |
| CI027 | As of June 2026, Periodic Labs has not publicly disclosed ARR, gross margin, NRR, CAC, payback period, customer count, or any standard SaaS or deep-tech financial metrics. | 高 | SI010, SI023, SI024 |
| CI028 | UpsideList's equity analysis models a base case of +75% valuation upside over two years, a bull case of +350% on breakthrough materials, and a bear case of -70% on commercialization failure. | 低 | SI009 |
| CI029 | Investors hold $300 million in liquidation preferences ahead of common stockholders; in any exit at or below the $1.2–$1.3 billion seed valuation, common shares would receive no proceeds. | 中 | SI009, SI010 |
| CI030 | Periodic Labs stated the $300M seed is earmarked for hiring, laboratory scale-out, and bringing first products to industry partners. | 中 | SI007, SI020 |
| CI031 | Materials science timelines from discovery to commercial production typically span 5–10 years; ViaNews cites a 60–70% failure rate for AI-materials ventures reaching commercial production. | 中 | SI003, SI018 |
| CI032 | Periodic Labs recruited researchers who left substantial equity packages at Meta, OpenAI, and DeepMind, implying above-market compensation to attract this cohort. | 中 | SI001, SI010 |
| CI033 | The global materials science market is estimated at over $2 trillion; the AI-driven autonomous discovery sub-segment is described as greenfield with no established revenue benchmarks. | 低 | SI011, SI004 |
| CI034 | ViaNews analysts warn that monthly burn could reach $10–15 million before revenue generation, and that the binary-outcome nature of materials discovery limits the company's recovery optionality if initial results disappoint. | 中 | SI003 |
| CI035 | No benchmarks evaluating Periodic Labs' AI scientist capabilities, autonomous lab throughput, or materials discovery efficiency have been published as of April 2026. | 中 | SI010 |
| CI036 | An SEC Form D filed on 2026-05-29 identifies an entity named 'AGC Wealt Periodic Labs I a Series of AGC AI Nexus Fund LLC,' administered by Sydecar, as having raised approximately $4.74 million from 7 investors — an SPV investing into Periodic Labs. | 中 | SI021 |
| CI037 | No Form D or similar regulatory disclosure filed directly by Periodic Labs (as the issuer) is publicly identified in SEC EDGAR as of June 2026; the company's direct financing terms are not in the public record. | 中 | SI021 |
| CI038 | Periodic Labs' contract-based revenue model creates revenue lumpiness and customer concentration risk unlike recurring SaaS revenue structures. | 中 | SI011, SI003 |
| CI039 | Periodic Labs disclosed customers in space and defense sectors at its launch in September 2025, confirming multi-sector commercial engagement from inception. | 中 | SI014, SI004 |
| CI040 | Industry sources describe early semiconductor contracts as confidential and potentially involving IP co-ownership clauses for jointly developed materials data. | 低 | SI005, SI008 |
| CI041 | a16z's investment thesis estimates the industries Periodic Labs targets — advanced manufacturing, materials science, semiconductors, energy, aerospace — represent roughly $15 trillion of global GDP. | 低 | SI004 |
| CI042 | UpsideList estimates an approximately 6-year time to liquidity event (IPO or acquisition) for Periodic Labs equity from the Series A stage. | 低 | SI009 |
| CI043 | NCBI/PMC-cited benchmarks for cloud lab access show entry costs of over $250K/year for Emerald Cloud Lab and over $100K to automate a single method at Strateos, with minimum one-year contracts. | 高 | SI017, SI016 |
| CE001 | Periodic Labs was founded in 2025 by Liam Fedus, former VP of Research at OpenAI and co-creator of ChatGPT, and Ekin Dogus Cubuk, former head of materials and chemistry research at Google Brain and DeepMind. | 高 | SE001, SE002, SE003 |
| CE002 | Periodic Labs' core product is an 'AI scientist' system designed to build artificial intelligence that can autonomously form scientific hypotheses, run physical experiments, and learn iteratively from results. | 高 | SE001, SE002 |
| CE003 | Periodic Labs structures its technology into two explicit tracks: 'Bits' covering LLM research, machine learning, and distributed training infrastructure; and 'Atoms' covering physical lab robotics, powder synthesis, and materials characterization. | 中 | SE011, SE018 |
| CE004 | The Periodic Labs AI scientist operates in a closed loop: AI generates hypotheses from literature and prior data, quantum mechanical simulations filter candidates, robotic synthesis produces physical samples, characterization instruments measure results, and all outcomes feed back into model training. | 高 | SE001, SE003, SE015 |
| CE005 | Periodic Labs' Atoms platform uses powder synthesis laboratories where robotic arms mix precursor chemicals and heat them in furnaces to produce candidate materials for superconductors and other compounds. | 高 | SE001, SE003, SE009 |
| CE006 | Each autonomous lab run generates gigabytes of proprietary, high-quality experimental data that does not exist in any public database or internet corpus. | 中 | SE001, SE003 |
| CE007 | Periodic Labs' founding thesis is that large language models have exhausted the estimated 10 trillion token internet corpus and require experimental data generated through direct physical interaction with the world to advance further. | 高 | SE001, SE003, SE005 |
| CE008 | Periodic Labs treats nature itself as the reinforcement learning environment: when the AI predicts a material's properties and robots synthesize it, the physical outcome provides an unambiguous training signal unavailable from text data. | 中 | SE001, SE003 |
| CE009 | Ekin Dogus Cubuk was a co-author on the 2023 GNoME paper at Google DeepMind that used graph neural networks to identify over 2.2 million potentially stable inorganic crystal structures, the largest such expansion in materials science history. | 高 | SE002, SE003, SE015 |
| CE010 | GNoME's technical approach uses graph neural networks trained on DFT-computed crystal energies to predict thermodynamic stability of candidate materials, enabling high-throughput screening of millions of candidate crystal structures. | 中 | SE003, SE015 |
| CE011 | Liam Fedus co-created ChatGPT and led the post-training team at OpenAI including development of the first trillion-parameter neural network; he departed OpenAI in March 2025 to found Periodic Labs. | 高 | SE002, SE015 |
| CE012 | The founding team includes contributors to OpenAI's Operator/Agent system, Microsoft's MatterGen LLM for materials science, and the neural attention mechanism underlying modern transformers. | 中 | SE001, SE002, SE010 |
| CE013 | Periodic Labs' primary research target is the discovery of high-temperature superconductors that operate above current cryogenic thresholds, with potential applications in next-generation power grids and chip efficiency. | 高 | SE001, SE003, SE009 |
| CE014 | Periodic Labs has an active commercial product: custom AI agents trained for an unnamed semiconductor manufacturer to help engineers interpret experimental data and address chip heat dissipation problems faster. | 中 | SE001, SE003, SE005 |
| CE015 | The company states it has current customers in space, defense, and semiconductor sectors as of its stealth emergence in September 2025. | 中 | SE003, SE012, SE016 |
| CE016 | Periodic Labs hired more than 20 researchers from OpenAI, DeepMind, Meta, Databricks, and Samsung, many forgoing tens to hundreds of millions of dollars in unvested equity to join the startup. | 高 | SE004, SE022 |
| CE017 | Key hires beyond the co-founders include Alexandre Passos (co-creator of o1 and o3), Eric Toberer (materials scientist with superconductor discoveries), and Matt Horton (creator of Microsoft's MatterGen). | 中 | SE015, SE023 |
| CE018 | As of October 2025, Periodic Labs confirmed it had set up a research lab in San Francisco and was working with experimental data and simulations, but co-founder Cubuk stated the robotic components were not yet running. | 高 | SE015, SE022 |
| CE019 | Periodic Labs' June 2026 job listings show active hiring for Automation Engineer, Process Engineer (Powder), Research Scientist Materials Synthesis, Research Scientist Thin Films, and Multiphysics Simulation Scientist (Semiconductors) in the Atoms track. | 中 | SE011 |
| CE020 | Periodic Labs maintains a closed-weights model policy; the 'Periodic First Release' model (January 2025) is closed and the broader product strategy beyond the AI scientist framing has not been publicly disclosed. | 中 | SE018 |
| CE021 | The a16z lead partner conducting due diligence noted that frontier AI models are objectively terrible at scientific analysis in condensed matter physics and relatively worse than human investigators—the company's own lead investor acknowledges the starting capability is below human expert baseline. | 中 | SE003 |
| CE022 | No peer-reviewed publications or external benchmarks for Periodic Labs' AI scientist capability have been published as of April 2026; the closed-weights model policy precludes external evaluation. | 中 | SE018 |
| CE023 | Bloomberg reported in March 2026 that Periodic Labs was in deal talks for a follow-on round at approximately $7 billion valuation, a more than fivefold step-up from the $1.3 billion seed valuation. | 中 | SE020, SE021 |
| CE024 | Forbes reported in May 2026 that Periodic Labs was in advanced talks to raise at least $500 million led by AMP (Anjney Midha's investment vehicle) at a $7.5 billion valuation, described as significantly oversubscribed. | 高 | SE004, SE017 |
| CE025 | Periodic Labs debuted on the Forbes AI 50 Brink List in 2026, recognized for training models to accelerate scientific discovery in semiconductors, magnetism, and superconductivity. | 高 | SE023, SE004 |
| CE026 | Quantum mechanical simulations bridge the Bits and Atoms tracks by narrowing the compound search space before committing to physical synthesis; this directly inherits from the GNoME methodology applied at Periodic Labs. | 中 | SE003, SE015 |
| CE027 | Periodic Labs' powder synthesis process involves mixing precursor powders, heating them in furnaces, and characterizing material properties including conductivity and critical temperature. | 中 | SE001, SE003, SE009 |
| CE028 | The proprietary experimental dataset—including failure data rarely published in conventional scientific literature—is Periodic Labs' primary stated competitive moat, creating a training advantage competitors cannot replicate from public literature. | 中 | SE001, SE003, SE005 |
| CE029 | Via.news noted that traditional materials development timelines average 10 to 20 years from laboratory to commercial deployment, creating structural tension between investor return expectations and physical science timelines. | 中 | SE007, SE025 |
| CE030 | Room-temperature superconductors remain theoretical despite decades of international research; recent high-profile claims such as LK-99 in 2023 failed independent replication, illustrating the risk that AI-predicted materials will similarly fail physical validation. | 中 | SE007, SE025 |
| CE031 | AI applications in materials science have shown limited commercial success to date; physical validation bottlenecks cannot be eliminated by AI prediction speed, and predictions always require experimental confirmation. | 中 | SE007, SE025 |
| CE032 | Sakana AI's open-source AI Scientist-v2 system generated the first ICLR workshop-accepted paper produced entirely by AI, demonstrating that AI-driven autonomous scientific discovery is feasible in the ML research domain but has not yet been demonstrated for physical-world materials discovery. | 高 | SE013, SE014 |
| CE033 | Periodic Labs explicitly captures negative experimental results in its proprietary training data, which conventional scientific publication norms typically exclude—creating a structurally more complete training corpus. | 中 | SE001, SE005 |
| CE034 | Co-founder Cubuk cited that reliable robotic arms for powder synthesis workflows only recently became mature enough for autonomous materials science experiments, making 2025 the right moment to found the company. | 中 | SE015 |
| CE035 | The company holds weekly cross-discipline teaching sessions in which physicists teach LLMs to reason about quantum mechanics and ML researchers learn physics intuitions, reinforcing tight Bits/Atoms integration. | 中 | SE015 |
| CE036 | The a16z investment thesis describes Periodic Labs as building systems that encode deep domain knowledge through mid-training and reinforcement learning against physical experimental outcomes, not through text-only pre-training. | 中 | SE003 |
| CE037 | Forbes reported that the Periodic Labs follow-on round was significantly oversubscribed and that talks for a fast-follow additional round at an even higher valuation were already underway as of May 2026. | 中 | SE004 |
| CE038 | The a16z investment announcement identifies advanced manufacturing, materials science, semiconductors, energy, and aerospace as Periodic Labs' priority market sectors, collectively representing approximately $15 trillion of global GDP. | 中 | SE003 |
| CE039 | As of October 2025, Periodic Labs had set up a lab and was working with experimental data, simulations, and testing some predictions, but co-founder Cubuk told TechCrunch the robots 'will take a bit to train.' | 中 | SE015 |
| CE040 | Periodic Labs' June 2026 job listings include a Multiphysics Simulation Scientist (Semiconductors) and Research Scientist Thin Films roles, signaling expansion of the Atoms lab scope beyond powder synthesis into semiconductor thin-film process engineering. | 中 | SE011 |
| CU001 | Periodic Labs has confirmed active customer relationships in semiconductor, space, and defense sectors as of September 2025, with ongoing engagements at the time of the a16z investment announcement. | 高 | SU001, SU002, SU004 |
| CU002 | The Periodic Labs official website confirms the company is helping an unnamed semiconductor manufacturer address heat-dissipation issues by training custom AI agents for engineers and researchers to iterate faster on experimental data. | 高 | SU001, SU002 |
| CU003 | Custom AI agents trained on customer proprietary experimental data help semiconductor engineers interpret results and decide which experiments to run next, serving as an AI-assisted R&D co-pilot. | 高 | SU001, SU002, SU005 |
| CU004 | Periodic Labs' stated go-to-market strategy is 'land and expand at the frontier': solve a specific critical problem with clear evaluations first, demonstrate physical-reality optimization superiority, then scale across the customer account. | 中 | SU002 |
| CU005 | The target buyer and user for Periodic Labs is engineers and researchers in advanced industrial R&D organizations, not enterprise software procurement teams or C-suite executives. | 中 | SU005, SU002 |
| CU006 | Target customer industries — semiconductors, advanced manufacturing, materials science, energy, and aerospace — represent 'roughly $15 trillion of global GDP' per a16z, establishing addressable market scale. | 中 | SU002 |
| CU007 | As of June 2026, no named customers, client companies, or formally attributed case studies have been publicly disclosed by Periodic Labs. | 高 | SU001, SU008, SU015 |
| CU008 | Revenue is estimated at under $5M by third-party intelligence as of Q1 2026, indicating early-stage monetization relative to the $300M capital deployed. | 低 | SU020 |
| CU009 | Customer count has not been publicly disclosed; multiple customers are implied by the enumeration of three sectors, but the minimum could be a single organization spanning all three. | 中 | SU002, SU004 |
| CU010 | TechFundingNews reported in March 2026 that Periodic Labs 'already secured customers in the semiconductor industry' and 'unlike many peers, the company is generating revenue.' | 中 | SU024, SU009 |
| CU011 | Ideal customer contacts are engineers and researchers — not software procurement — who 'don't really have particularly good tools' for analyzing complex experimental data in advanced industrial R&D. | 中 | SU005 |
| CU012 | ICP is concentrated in organizations with 'massive R&D budgets' in semiconductor, space, and defense; no minimum company-size, revenue, or headcount threshold has been publicly stated. | 中 | SU005, SU002 |
| CU013 | The customer workflow involves AI scientists forming hypotheses, running quantum-mechanical simulations, planning syntheses, and feeding experimental results back into the model — creating a closed loop between AI and physical reality. | 中 | SU001, SU006 |
| CU014 | Periodic Labs trains custom agents per customer to help engineers process experimental data and iterate faster, suggesting a bespoke onboarding model rather than a standardized self-serve product. | 中 | SU001, SU002, SU005 |
| CU015 | Target customer industries account for 'roughly $15 trillion of global GDP' per a16z, which Periodic Labs cites as the commercial validation of its focus on physical sciences. | 中 | SU002 |
| CU016 | Periodic Labs' strategic investor ecosystem includes NVIDIA (NVentures), a16z, DST Global, Accel, and Felicis, as well as individuals including Jeff Bezos, Eric Schmidt, Jeff Dean, and Elad Gil. | 高 | SU001, SU006, SU023 |
| CU017 | NVIDIA's strategic investment through NVentures aligns with Periodic Labs' GPU-compute-intensive simulation and model-training workloads but does not constitute a confirmed customer deployment. | 中 | SU008 |
| CU018 | Bromley Capital Partners (UK) confirmed advising on a multi-million dollar private placement into Periodic Labs concluded in January 2026, expanding the investor base internationally. | 中 | SU010 |
| CU019 | Forbes included Periodic Labs on its inaugural 2026 AI 50 Brink List, which requires 'early traction' as a selection criterion, providing limited but independent editorial validation. | 高 | SU018, SU003 |
| CU020 | No entries for Periodic Labs were found on G2, Gartner Peer Insights, Capterra, or any other independent customer review platform during research conducted June 2026. | 中 | SU008, SU015 |
| CU021 | No independent customer testimonials, named customer conference talks, or formally published case studies have been identified for Periodic Labs in any source reviewed. | 中 | SU008, SU001 |
| CU022 | Competitors Lila Sciences, CuspAI, and Radical AI target overlapping enterprise verticals without publicly naming customers, suggesting the opacity pattern is sector-wide rather than unique to Periodic Labs. | 中 | SU016 |
| CU023 | Enterprise procurement cycles in semiconductor and defense R&D typically span 12-24 months due to technical validation requirements, security reviews, and multi-stakeholder approvals. | 中 | SU008, SU014 |
| CU024 | Defense sector procurement for AI platforms carrying experimental data from a defense R&D program requires ITAR compliance review and data-residency assurances, adding regulatory complexity to the sales cycle. | 中 | SU008, SU013 |
| CU025 | IP ownership over AI-discovered material insights is unresolved: it is unclear whether customers retain rights to discoveries made using Periodic Labs' platform from their own experimental data. | 中 | SU008, SU013 |
| CU026 | Periodic Labs' commercial model — platform license, discovered-IP ownership, or research services — has not been publicly defined, creating budget-cycle and risk-allocation ambiguity for enterprise buyers. | 中 | SU008 |
| CU027 | MIT Technology Review reported in December 2025 that no AI materials discovery startup had produced a 'eureka moment' and that the gap between simulation prediction and physical synthesis remains the central bottleneck. | 高 | SU014, SU013 |
| CU028 | Third-party risk analysis cites a 60-70% failure rate for AI-materials ventures reaching commercial production, even with promising laboratory results. | 中 | SU012 |
| CU029 | Traditional materials science development from discovery to commercial market typically takes 10-20 years, creating structural tension between Periodic Labs' investor return expectations and materials-market realities. | 中 | SU013, SU014 |
| CU030 | As of October 2025, robotic arms at Periodic Labs' laboratory were 'not yet up and running,' per TechCrunch, though the lab was already working with experimental data and simulations. | 中 | SU021 |
| CU031 | A May 2026 SEC Form D filing for 'AGC Wealt Periodic Labs I, a Series of AGC AI Nexus Fund LLC' provides regulatory confirmation of structured investment activity around Periodic Labs as recently as May 2026. | 中 | SU025 |
| CU032 | Wilson Sonsini Goodrich & Rosati (WSGR), a leading Silicon Valley technology law firm, confirmed advising Periodic Labs on the $300M seed round, validating the transaction's legal integrity. | 高 | SU023, SU006 |
| CU033 | Felicis Ventures committed the first check into Periodic Labs before the company had a name, incorporation, or bank account, reflecting exceptionally high conviction in the founders and customer thesis. | 中 | SU022 |
| CU034 | Forbes' AI 50 Brink selection methodology is based on 'business promise, early traction and the use of AI in solving a new type of problem,' making inclusion implicit evidence of commercial progress. | 中 | SU018 |
| CU035 | Periodic Labs launched an Academic Grant Program for research institutions, creating a secondary non-paying user base that could serve as a long-term pipeline for future enterprise relationships. | 中 | SU001 |
| CU036 | No retention, NRR, GRR, contract length, or cohort data has been disclosed by Periodic Labs or found in any third-party source as of June 2026. | 中 | SU008, SU015 |
| CU037 | No evidence of formal customer contract renewals, long-term commitments, or multiyear agreements has been identified in any public source. | 中 | SU008 |
| CU038 | Customer concentration risk is elevated: all publicly confirmed deployments are in the semiconductor vertical, while space and defense traction is implied but unconfirmed with no use-case detail. | 中 | SU002, SU004, SU005 |
| CU039 | Forbes reported in May 2026 that the $500M follow-on round was 'significantly oversubscribed,' with active talks for a fast-follow additional round, serving as a proxy signal for strong customer pipeline confidence. | 高 | SU003, SU018 |
| CU040 | Periodic Labs' autonomous laboratories run continuous experiments without human working-hour constraints, potentially compressing experimental iteration timelines from months to days for customer R&D teams. | 中 | SU001, SU006 |
| CR001 | Google DeepMind's GNoME AI system predicted approximately 2.2 million crystal structures, of which roughly 380,000 were assessed as potentially stable; only approximately 736 have been independently synthesized and verified as of late 2025 — a realization rate of roughly 0.03%. | 高 | SR002, SR009, SR020 |
| CR002 | Periodic Labs is building an AI-directed autonomous laboratory targeting materials discovery with a focus on superconductors, using a closed-loop system in which AI generates experimental hypotheses that robotic systems directly execute. | 高 | SR001, SR004 |
| CR003 | arXiv research published April 2025 identified 'corrosive hallucinations' — scientifically plausible but factually incorrect AI outputs that resist standard detection — as the highest-risk failure mode for AI-integrated experimental design pipelines. | 高 | SR006, SR003 |
| CR004 | The synthesis gap — discrepancy between AI-predicted crystal structures and physically realizable materials — is the primary unsolved constraint in autonomous materials discovery; the bottleneck has shifted from model inference to physical validation. | 高 | SR009, SR020, SR002 |
| CR005 | Active learning loops in AI materials discovery depend on density functional theory labels for ground-truth, and systematic DFT approximation errors propagate into iterative training cycles, requiring expensive recalibration with each new model generation. | 中 | SR018, SR006 |
| CR006 | Autonomous experimental systems operating with reduced human oversight narrow the window for detecting AI errors before they propagate through multiple costly experimental cycles. | 中 | SR006, SR015 |
| CR007 | Reproducibility failure — AI-predicted synthesis routes not replicating across different laboratory equipment — is a documented key challenge in post-GNoME materials validation literature. | 中 | SR009, SR020 |
| CR008 | Room-temperature superconductivity has not been confirmed in any peer-reviewed study as of June 2026; prior claims such as LK-99 in 2023 were not reproducible. | 中 | SR009, SR020 |
| CR009 | OSHA's Laboratory Standard (29 CFR 1910.1450) requires chemical hygiene plans, PPE protocols, and hazard assessments for all novel compounds handled, including AI-generated candidates with unknown toxicity profiles. | 高 | SR010, SR017 |
| CR010 | ANSI/A3 R15.06-2025, published September 2025, requires documented risk assessments before each new or modified robotic process, imposing significant administrative burden on AI-varied autonomous workflows. | 高 | SR011, SR012 |
| CR011 | OSHA's accident investigation database records a February 2024 fatality in which an employee was crushed by a robot arm, illustrating that even mature industrial robot deployments can produce fatal outcomes without adequate safety protocols. | 高 | SR024, SR011 |
| CR012 | The US National Security Commission on Emerging Biotechnology identified autonomous AI laboratory platforms as sources of emerging biosecurity risk that current governance frameworks do not adequately address. | 高 | SR019, SR014 |
| CR013 | A 2022 demonstration showed an AI chemistry tool generated thousands of potentially weaponizable molecules in hours when safety filters were disabled, triggering biosecurity community calls for mandatory safeguards. | 高 | SR013, SR014 |
| CR014 | The EU AI Act 2024 amendment classifies AI-driven laboratory platforms as high-risk applications, requiring conformity assessment, human oversight documentation, and logging of AI-generated experimental decisions. | 中 | SR014, SR030 |
| CR015 | OSHA's lockout/tagout standard (29 CFR 1910.147) requires documented energy control procedures for all robotic equipment maintenance; the Arms Control Association's November 2025 report identified benchtop synthesis regulatory gaps as structural biosecurity vulnerabilities. | 高 | SR011, SR013 |
| CR016 | OSHA's 2024 Hazard Communication Standard update (GHS 7th revision) requires updated safety data sheets and labeling for all hazardous chemicals, creating compliance challenges for novel AI-generated compounds without pre-existing toxicology data. | 高 | SR010, SR025 |
| CR017 | Both the USPTO and EPO require human inventors on patent applications; AI-generated materials discoveries must identify a human contributor to the inventive concept, creating legal uncertainty for autonomously discovered materials. | 高 | SR012, SR016, SR029 |
| CR018 | Training-data litigation risk for AI systems used in materials discovery is escalating, with multiple pending cases challenging the use of published scientific literature for AI model training without licensing agreements. | 中 | SR016, SR029 |
| CR019 | NSCEB's 2025 biosecurity report specifically flagged the structural governance gap for autonomous laboratory AI platforms, advocating mandatory built-in safeguards before commercial deployment. | 高 | SR019, SR014 |
| CR020 | Periodic Labs raised $300 million in seed funding in September 2025, with anchor investors including Andreessen Horowitz, Nvidia, Jeff Bezos, and Eric Schmidt, making it one of the largest seed rounds on record for a scientific AI startup. | 高 | SR001, SR004, SR005 |
| CR021 | Third-party analysis estimates Periodic Labs' annual operational costs at $50-75 million given autonomous laboratory infrastructure, top-tier research talent, and GPU compute requirements, suggesting the $300M seed could be consumed within four to six years without commercial revenue. | 中 | SR003, SR004 |
| CR022 | Via News reported in December 2025 that Periodic Labs faces mounting commercial pressure to demonstrate a viable revenue path before the $300M seed capital is exhausted, with investors expecting venture-scale returns misaligned with materials science timelines. | 中 | SR007, SR003 |
| CR023 | MIT Technology Review reported in December 2025 that AI materials discovery startups have yet to cross the lab-to-market translation threshold necessary to justify venture-scale return assumptions. | 高 | SR022, SR007 |
| CR024 | PitchBook's 2026 analysis identified the 'winding road to VC returns' for AI materials discovery companies, noting that historical materials science timelines of 10-15 years are incompatible with standard 7-10 year VC fund cycles. | 高 | SR023, SR022 |
| CR025 | Multiple well-capitalized competitors are active in AI materials discovery as of 2026: Lila Sciences ($550M raised), CuspAI ($154M), Orbital Industries ($50M Series B in 2026), and Microsoft MatterGen (MIT open-source license). | 高 | SR026, SR033, SR023 |
| CR026 | Microsoft released MatterGen under the MIT open-source license in January 2025, commoditizing the AI prediction layer and eliminating prediction-alone as a defensible competitive differentiator. | 高 | SR026, SR028 |
| CR027 | Google DeepMind continues to publish GNoME results openly, creating a dynamic where the founder's predecessor platform serves simultaneously as a public benchmark and a competitive threat. | 中 | SR002, SR028 |
| CR028 | Periodic Labs was co-founded by Ekin Dogus Cubuk (GNoME architect at Google DeepMind) and Liam Fedus (former VP of Research at OpenAI), creating exceptional scientific credibility but concentrated key-person dependency. | 高 | SR001, SR004, SR005 |
| CR029 | The Periodic Labs board includes representatives from Andreessen Horowitz, Nvidia, and Bezos-affiliated entities, which may generate pressure for commercial milestones at timescales inconsistent with materials science discovery cycles. | 中 | SR005, SR001 |
| CR030 | RAND Corporation's 2025 report on autonomous AI laboratory platforms assessed current regulatory frameworks as inadequate for the rate of deployment and advocated mandatory biosecurity and dual-use screening requirements. | 高 | SR030, SR031 |
| CR031 | CSIS published analyses in both 2025 and 2026 calling for mandatory governance of autonomous AI laboratory platforms and classifying the regulatory gap as a national security concern. | 高 | SR014, SR031 |
| CR032 | Periodic Labs has not publicly disclosed any AI governance policy, dual-use screening protocol, AI safety framework, or succession plan for its co-founders as of June 2026. | 高 | SR001, SR004 |
| CR033 | Networked autonomous laboratory robotic systems face documented cybersecurity threats including adversarial data poisoning, prompt injection attacks, and robotic control hijacking that could compromise experimental integrity. | 中 | SR015, SR006 |
| CR034 | Absence of a published AI safety framework exposes Periodic Labs to reputational risk from sector-wide incidents: any public AI chemistry event could trigger regulatory scrutiny affecting all autonomous laboratory operators. | 中 | SR030, SR031, SR013 |
| CR035 | CSIS and RAND have both called for human oversight checkpoints and mandatory dual-use screening to be embedded in AI-directed experimental design pipelines before autonomous synthesis operations proceed. | 高 | SR031, SR030 |
| CR036 | Departure of either Periodic Labs co-founder before the company's first commercial milestone would impair its ability to attract top-tier scientific talent, maintain investor confidence, and execute its research roadmap. | 中 | SR001, SR007 |
| CR037 | Via News analysis from December 2025 identified Periodic Labs as subject to mounting commercial pressure, noting that the deep-tech materials discovery sector has produced few commercial successes despite substantial AI investment. | 中 | SR007, SR027 |
| CR038 | MIT Technology Review's December 2025 investigation found that no AI materials discovery company had yet achieved a commercial licensing agreement for an AI-predicted material as of the publication date. | 高 | SR022, SR023 |
| CR039 | RAND and CSIS analyses recommend adoption of ANSI/A3 R15.06-2025 robotic safety documentation and independent biosecurity review as foundational mitigations for autonomous laboratory operators. | 高 | SR030, SR031 |
| CR040 | Cybersecurity monitoring, air-gapped controls for synthesis systems, and third-party penetration testing are identified in expert literature as necessary protective measures for networked autonomous laboratory environments. | 中 | SR015, SR006 |
| CR041 | The AI materials discovery sector attracted over $1 billion in investor capital in the 18 months prior to June 2026, validating the thesis while creating significant competitive pressure on any single entrant's ability to build a durable commercial moat. | 中 | SR023, SR033, SR005 |
| CR042 | WEF analysis from September 2025 identifies regulatory approval, manufacturing validation, and customer adoption as three distinct post-discovery barriers that AI prediction capability alone does not address and that Periodic Labs has not publicly disclosed plans to overcome. | 中 | SR032, SR022 |
| CR043 | WEF stresses that transforming AI-predicted materials into market-ready products requires multi-year engagement with supply chain partners, regulatory agencies, and industrial customers — a commercialization track distinct from laboratory discovery. | 中 | SR032 |
| CR044 | Nvidia holds anchor investment positions in both Periodic Labs and Orbital Industries, a direct competitor, creating a potential conflict of interest in future financing negotiations and GPU supply arrangements. | 中 | SR005, SR033 |
| CR045 | Orbital Industries raised a $50 million Series B in 2026 and is pursuing a vertically integrated model combining AI discovery with downstream manufacturing, directly competing with Periodic Labs on commercial translation capability. | 中 | SR033, SR023 |
| CR046 | Consumption of more than 75% of raised capital without a disclosed commercial partnership or licensing agreement represents a critical investor kill criterion, given third-party estimates of $50-75M annual burn and a $300M seed raise implying a 4-6 year runway. | 中 | SR003, SR021 |
| CV001 | Periodic Labs raised a $300 million seed round at a $1.3 billion post-money valuation on September 30, 2025, led by Andreessen Horowitz. | 高 | SV004, SV003 |
| CV002 | The Periodic Labs seed round included participation from Felicis, DST Global, Accel, and NVentures (Nvidia's venture arm). | 高 | SV004, SV020 |
| CV003 | Individual angel investors in the Periodic Labs seed round include Jeff Bezos, Eric Schmidt, Jeff Dean, and Elad Gil. | 高 | SV003, SV016 |
| CV004 | Wilson Sonsini Goodrich and Rosati served as legal counsel for Periodic Labs in the September 2025 seed round, led by partner Yokum Taku. | 中 | SV004 |
| CV005 | Periodic Labs is in advanced talks as of May 2026 to raise at least $500 million at a $7.5 billion valuation in a round led by AMP, a vehicle founded by former Andreessen Horowitz GP Anjney Midha. | 高 | SV001, SV005 |
| CV006 | If the proposed $500 million 2026 round closes at $7.5 billion, Periodic Labs will have raised approximately $800 million in total capital in under 12 months of existence. | 中 | SV001, SV004 |
| CV007 | The 2026 Periodic Labs financing round is reportedly significantly oversubscribed and there are already discussions for a fast-follow round at an even higher valuation. | 中 | SV001 |
| CV008 | Bloomberg reported in March 2026 that Periodic Labs was in deal talks at approximately $7 billion valuation; Forbes reported in May 2026 that the figure had risen to $7.5 billion. | 中 | SV002, SV001 |
| CV009 | At the proposed $7.5 billion valuation, Periodic Labs' enterprise value would have increased nearly sixfold from the $1.3 billion seed mark in under eight months. | 高 | SV001, SV004 |
| CV010 | Periodic Labs was founded in May 2025 in San Francisco by Liam Fedus (former OpenAI VP of Research) and Ekin Dogus Cubuk (former Google Brain and DeepMind research scientist). | 高 | SV003, SV004 |
| CV011 | Isomorphic Labs raised $600 million in its first external funding round (Series A) in March 2025, led by Thrive Capital, with participation from GV and Alphabet. | 高 | SV007, SV011, SV017 |
| CV012 | Isomorphic Labs raised $2.1 billion in a Series B round in May 2026, led by Thrive Capital, bringing total external capital raised to approximately $2.7 billion. | 中 | SV028 |
| CV013 | Sakana AI raised $135 million in a Series B round in November 2025 at a $2.65 billion post-money valuation, led by Mitsubishi UFJ Financial Group with Khosla Ventures and other global VCs. | 高 | SV008, SV023 |
| CV014 | Anthropic was reported in May 2026 to be pursuing a funding round targeting approximately $900-965 billion valuation, briefly making it the world's most valuable private AI company ahead of OpenAI. | 中 | SV009, SV029 |
| CV015 | OpenAI was valued at approximately $852 billion in mid-2026 after closing a $122 billion funding round, representing the largest single private company financing in history. | 中 | SV009, SV029 |
| CV016 | Lila Sciences, a Flagship Pioneering-backed AI scientific superintelligence startup focused on autonomous science factories, raised over $500 million in 2025 across multiple tranches with participation from Nvidia NVentures. | 中 | SV012 |
| CV017 | Pathos AI raised $365 million in a Series D round in May 2025 at approximately $1.6 billion post-money valuation, developing AI oncology discovery platforms with clinical-stage assets. | 中 | SV012 |
| CV018 | Gartner forecasts global semiconductor revenue will exceed $1.3 trillion in 2026, representing 64% year-over-year growth—the highest in two decades—driven by AI processing demand. | 高 | SV014, SV015 |
| CV019 | AI semiconductors are projected to represent approximately 30% of total semiconductor revenue in 2026, equivalent to over $390 billion, per Gartner. | 高 | SV014, SV006 |
| CV020 | Global semiconductor revenue grew at double-digit rates in 2024, 2025, and is projected to in 2026, marking a third consecutive year of strong expansion per Gartner analysis. | 高 | SV014, SV015 |
| CV021 | AMP, the investment vehicle making the 2026 Series A lead for Periodic Labs, was founded by former Andreessen Horowitz general partner Anjney Midha. | 中 | SV001 |
| CV022 | SEC EDGAR shows a Form D filing on May 29, 2026 (CIK 0002122824) for AGC Wealt Periodic Labs I, a Series of AGC AI Nexus Fund LLC, a Sydecar-administered venture capital SPV, raising $4,743,803 from 7 investors. | 中 | SV022 |
| CV023 | Siemens acquired Dotmatics, a provider of AI-enabled R&D scientific software platforms including GraphPad Prism and SnapGene, for $5.1 billion in April 2025, completed July 2025. | 中 | SV012, SV015 |
| CV024 | In 2025, AI/ML drug discovery and licensing M&A reached $12.3 billion across 99 transactions, with the gap between headline and cash values of only $1 billion indicating mostly cash-dominant deal structures. | 中 | SV012 |
| CV025 | Finro's Q1 2026 AI company dataset of 575 companies shows LLM Vendors at a median 39.5x EV/Revenue, seed-stage AI companies at a median 20.2x, and infrastructure AI at 21.2x median EV/Revenue. | 中 | SV006 |
| CV026 | Finro Q1 2026 data shows seed-stage AI companies have 25th-75th percentile EV/Revenue range of 5.2x to 25.3x, with a median of 20.2x across 105 companies in the dataset. | 中 | SV006 |
| CV027 | Finerva reports that public Robotics and AI companies traded at a median EV/Revenue of 3.4x in Q4 2025, recovering from a 2.5x bottom in Q1 2025 but still far below the 6x peak of 2021. | 中 | SV018 |
| CV028 | Private AI company EV/Revenue multiples in 2026 range from 10x to 50x for most companies with category leaders reaching 40-100x, substantially above public company benchmarks, per Qubit Capital analysis. | 中 | SV010 |
| CV029 | Public Robotics and AI median EV/Revenue of 3.4x versus private AI seed company median of 20.2x represents a 6x private-to-public premium that could compress materially if AI hype cycle normalizes. | 中 | SV006, SV018 |
| CV030 | AI investment in 2025 reached a record high exceeding $225 billion globally, surpassing the prior 2021 peak of $74.6 billion by more than 3x, indicating a substantial potential for mean reversion. | 中 | SV018, SV029 |
| CV031 | Periodic Labs' closed-loop autonomous laboratory model generates proprietary experimental data—including negative results rarely published—that competitors relying on published scientific literature cannot access or replicate. | 中 | SV016, SV003 |
| CV032 | Periodic Labs has secured paying customers in the semiconductor industry as of late 2025, specifically assisting a manufacturer with chip heat dissipation research, making it commercially engaged at seed stage. | 中 | SV003, SV016, SV020 |
| CV033 | Periodic Labs has hired over 20 researchers from Meta, OpenAI, and DeepMind, many of whom left substantial unvested equity packages to join the startup. | 中 | SV001, SV003 |
| CV034 | The Periodic Labs founding team's credentials include co-creation of ChatGPT, co-authorship of the GNoME materials discovery paper, creation of the neural attention mechanism, and development of Microsoft's GenAI materials science tools MatterGen and MatterSim. | 高 | SV003, SV020 |
| CV035 | OpenAI announced in late 2025 the launch of an 'OpenAI for Science' unit to build AI-powered scientific discovery platforms, signaling hyperscaler entry into Periodic Labs' primary market. | 中 | SV003 |
| CV036 | Periodic Labs has no publicly disclosed revenue base; the $7.5 billion proposed valuation cannot be anchored to any reported ARR, contract value, or revenue run rate. | 高 | SV001, SV005, SV021 |
| CV037 | The proposed 2026 valuation of $7.5 billion represents a nearly sixfold increase from the $1.3 billion seed mark in under eight months with no publicly reported step-change in commercial scale, product capability, or laboratory infrastructure. | 中 | SV001, SV021 |
| CV038 | As of late 2025, Periodic Labs' robotic arm systems for autonomous laboratory experimentation were not yet fully operational; TechCrunch reported the robots 'will take a bit to train.' | 中 | SV003 |
| CV039 | An SEC Form D filing (CIK 0002122824, filed May 29, 2026) for a Sydecar-administered SPV named AGC Wealt Periodic Labs I confirms small-lot co-investment vehicles are providing retail-accessible entry into Periodic Labs' 2026 round. | 中 | SV022 |
| CV040 | TechCrunch noted in October 2025 that real-world usefulness of AI materials tools—beyond stability predictions—remains limited, with Periodic's wager requiring closed-loop experimentation to overcome existing constraints. | 中 | SV003 |
| CV041 | Periodic Labs is building a dedicated Menlo Park laboratory facility designed for physical robotic experiments at large scale, to complement its initial San Francisco operations. | 中 | SV024 |
| CV042 | The Periodic Labs scientific approach generates gigabytes of novel experimental data per run, including failed experiments that provide training signal unavailable from published scientific literature. | 中 | SV016, SV024 |
| CV043 | Periodic Labs is targeting the discovery of superconductors that function at higher temperatures as a primary moonshot application, with potential to transform power grids and transportation systems. | 中 | SV003, SV016 |
| CV044 | The 2025-2026 AI science exit landscape includes multiple pathways: M&A by hyperscalers and industrials (Siemens paid $5.1B for Dotmatics), strategic pharmaceutical partnerships (Isomorphic Labs $3B milestone commitments), and IPO (Caris Life Sciences $5.9B at IPO). | 中 | SV012, SV013 |
| CV045 | Isomorphic Labs has secured strategic partnerships with Eli Lilly and Novartis potentially generating up to $3 billion in combined milestone payments, demonstrating the commercial value achievable by AI science platforms in regulated industries. | 高 | SV007, SV011 |
| CV046 | XtalPi's milestone-heavy AI discovery partnership with DoveTree, valued at up to $6 billion across multiple therapeutic areas in August 2025, demonstrates multibillion-dollar commercial deal structures available to AI science platforms. | 中 | SV012 |
| CV047 | In 2025, AI/ML drug discovery venture funding reached $11 billion across 348 rounds, driven by platforms like Isomorphic Labs ($600M Series A), Pathos AI ($365M Series D), and Lila Sciences ($500M+), confirming deep institutional appetite for AI science at scale. | 中 | SV012 |
| CV048 | A down-round from the $7.5 billion proposed valuation would require material underperformance given the 2026 AI science market remains highly liquid; however, if autonomous lab commercialization is delayed past 2028, a below-mark round is plausible. | 中 | SV018, SV001 |
| CV049 | The competitive risk from hyperscaler autonomous lab programs (OpenAI for Science, Google DeepMind A-Lab follow-on) represents a structural threat to Periodic Labs' market position that could materialize within a 3-5 year horizon. | 中 | SV003, SV015 |
| CV050 | PwC's 2026 Technology Deals Outlook identifies proprietary data assets, scalable AI tooling, and specialized engineering talent as the highest-value M&A targets, all of which Periodic Labs possesses, making strategic acquisition by a hyperscaler a plausible exit. | 中 | SV015 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | TechCrunch | Former OpenAI and DeepMind researchers raise whopping $300M seed to automate science | Periodic Labs came out of stealth on Tuesday with a war chest of $300 million as a seed round backed by a tech industry whos who: Andreessen Horowitz DST Nvidia Accel Elad Gil Jeff Dean Eric Schmidt and Jeff Bezos. |
| SO002 | TechCrunch | Top OpenAI Google Brain researchers set off a $300M VC frenzy for their startup Periodic Labs | That investment didnt actually materialize however. OpenAI is not a backer of Periodic the founders confirmed to TechCrunch. |
| SO003 | Forbes | Former OpenAI Researcher To Raise $500 Million For AI Science Startup | Periodic Labs a startup building an AI scientist that can use automated labs to make discoveries is in advanced talks to raise at a $7.5 billion valuation in a round led by AMP an investment vehicle founded by former Andreessen Horowitz general partner Anjney Midha. |
| SO004 | Wilson Sonsini Goodrich and Rosati | Wilson Sonsini Advises Periodic Labs on $300 Million Seed Round | On September 30 2025 Periodic Labs announced a $300 million Seed round led by Andreessen Horowitz. Additional participation came from Felicis DST Global NVIDIA Accel and individual investors including Elad Gil Eric Schmidt Jeff Dean and Jeff Bezos. |
| SO005 | Periodic Labs | Periodic Labs Company Website | At Periodic we are building AI scientists and the autonomous laboratories for them to operate. |
| SO006 | TechCrunch | OpenAI exec leaves to found materials science startup | Liam Fedus OpenAIs VP of research for post-training is leaving the company to found a materials science AI startup. |
| SO007 | The AI Insider | Periodic Labs Emerges from Stealth with $300 Million Seed Round to Build AI Scientists | |
| SO008 | Observer | Jeff Bezos-Backed Startup Receives $300M Seed Round to Build an A.I. Scientist | The startup has already begun partnering with semiconductor makers to improve chip heat dissipation and its customer base also includes companies in space and defense. |
| SO009 | Datamation | Periodic Labs Powers Up for Scientific AI Advances | |
| SO010 | Analytics India Magazine | What Does Liam Fedus Departure Mean for AI Innovation? | |
| SO011 | eWeek | AI Scientists Just Got $300M and a Robot Army | |
| SO012 | TechFundingNews | Ex-OpenAI execs raise $200M at $1B valuation for AI materials science startup backed by a16z | |
| SO013 | TechFundingNews | Former OpenAI and DeepMind researchers seek $7B valuation to build AI scientists | |
| SO014 | The Outpost | Periodic Labs Raises $300M to Create AI-Powered Scientific Research Platform | |
| SO015 | Maginative | Periodic Labs launches with $300M to build an AI scientist | |
| SO016 | Labcritics | Dark Laboratories: AI Industry Veterans Launch New AI-Scientist Venture with Periodic Labs | |
| SO017 | Felicis Ventures | Felicis Seed in Periodic Labs: AI Models to Accelerate Materials Discovery | Dogus earned his PhD at Harvard and completed a postdoc at Stanford led Materials Science and Chemistry Research at DeepMind where he co-authored GNoME which identified millions of new stable crystals. |
| SO018 | Avantgarde News | Periodic Labs to Raise $500 Million for AI-Driven Scientific Research | |
| SO019 | Futurism | The More Scientists Work With AI the Less They Trust It | In 2024 scientists surveyed said they believed AI was already surpassing human abilities in over half of all use cases. In 2025 that belief dropped off a cliff falling to less than a third. |
| SO020 | Nature | Artificial intelligence tools expand scientists impact but contract sciences focus | |
| SO021 | MLQ.AI | Periodic Labs reportedly raises $300M for AI-powered materials science platform | |
| SO022 | Bloomberg | AI Science Startup Periodic Labs Is in Deal Talks at About $7 Billion Valuation | |
| SO023 | Forbes | When AI Takes Over Scientific Discovery | Yann LeCun Meta Chief AI Scientist and a Turing Award winner has long warned against mistaking pattern-matching for true intelligence. Current AI models cannot form the kind of mental models that underpin real-world reasoning or original discovery. |
| SO024 | 36Kr English | Former OpenAI VP Joins Forces with DeepMind Scientists to Launch Business: Over 20 Elite Scientists $300M Wager on AI for Science | |
| SO025 | Nextomoro | Liam Fedus Profile | Liam Fedus is an American computer scientist and machine-learning researcher. He is the co-founder and chief executive officer of Periodic Labs. |
| SM001 | TechCrunch | Former OpenAI and DeepMind researchers raise whopping $300M seed to automate science | Former OpenAI and DeepMind researchers have raised a $300M seed round to automate science with AI. |
| SM002 | Forbes | Former OpenAI Researcher To Raise $500 Million For AI Science Startup | The round is reportedly led by AMP, the new investment firm from ex-Andreessen Horowitz partner Anjney Midha. |
| SM003 | Bloomberg | AI Science Startup Periodic Labs Is in Deal Talks at About $7 Billion Valuation | Periodic Labs is in deal talks at about a $7 billion valuation. |
| SM004 | MIT Technology Review | The AI materials science startups attracting massive investment | |
| SM005 | ResearchAndMarkets (The Business Research Company) | AI in Materials Discovery Global Market Report 2026 | |
| SM006 | EIN News (press release) | Artificial Intelligence (AI) in Materials Discovery Market: Key Drivers, Regional Insights, Size Analysis 2026-2030 | |
| SM007 | DimensionMarketResearch | Autonomous Chemical Laboratory Market Size 2026–2035 | |
| SM008 | TowardsHealthcare | AI in Lab Automation Market to Grow at 18.44% CAGR by 2035 | |
| SM009 | The Business Research Company | Lab Automation Global Market Report 2026 | |
| SM010 | RealTimeDataStats | Autonomous Lab Robotics Market Share & Industry Trends 2032 | |
| SM011 | Cypris.ai | AI-Accelerated Materials Discovery in 2026: How Generative Models, Graph Neural Networks, and Autonomous Labs Are Transforming R&D | |
| SM012 | SandboxAQ | AQVolt26: Advancing AI-Driven Discovery for Next-Generation Solid-State Batteries | |
| SM013 | TechXplore | AI speeds up discovery of next-gen computer chips and electronic materials | |
| SM014 | Pangaea Ventures | AI & Materials Discovery - Part 1: Four Paths to the Frontier | |
| SM015 | AvantGardeNews | Startups Raise $1.3B for AI Scientists in Materials Discovery | |
| SM016 | Royal Society Publishing (Royal Society Open Science) | Autonomous 'self-driving' laboratories: a review of technology and applications | Autonomous self-driving laboratories are transitioning from pilot projects to broader adoption. |
| SM017 | Forbes | Overcoming Barriers To AI Adoption In 2026 | |
| SM018 | Deloitte | The State of AI in the Enterprise – 2026 AI Report | Only about a quarter of organizations have mature AI governance frameworks. |
| SM019 | Statista | Topic: Pharmaceutical Research and Development (R&D) | |
| SM020 | WIPO (World Intellectual Property Organization) | End of Year Edition – Despite the Odds, Global R&D Spending Grew in 2024 | WIPO estimates based on GII Database and data from Eurostat, OECD, RICYT, and UNESCO UIS. |
| SM021 | ComputeForecast | Enterprise AI Adoption Slower Than Forecast: The Real Barriers in 2026 | The enterprise AI adoption story of 2026 is not a story about technology falling short. It is a story about the implementation infrastructure needed to convert capability into production value taking longer to build than the technology took to arrive. |
| SM022 | IQVIA Institute | Global R&D Trends 2026 | |
| SM023 | Max Planck Institute for Iron Research (MPIE) | Accelerating battery innovation with AI-driven materials discovery | |
| SM024 | ChemDive | AI-Accelerated Material Discovery: What Will Happen in 2026 | |
| SM025 | Writer.com | Enterprise AI Adoption in 2026: Why 79% Face Challenges | |
| SM026 | R&D World (RDWorldOnline) | Global R&D Funding Forecast | |
| SM027 | AvantGardeNews | Periodic Labs Raising $500M for AI Science Startup | |
| SM028 | U.S. Securities and Exchange Commission | Periodic Labs Form D Filing (SEC EDGAR) | Form D filing confirms Periodic Labs' exempt offering registration with the SEC. |
| SM029 | DeepMind (Google) | Millions of new materials discovered with deep learning | GNoME identified 2.2 million stable materials, including 380,000 stable crystals. |
| SM030 | Nextomoro | Periodic Labs | |
| SP001 | Periodic Labs | Periodic Labs — Introductory Blog Post | At Periodic, we are building AI scientists and the autonomous laboratories for them to operate. |
| SP002 | TechCrunch | Former OpenAI and DeepMind researchers raise whopping $300M seed to automate science | It is not the only one working on AI scientists. AI as a tool to automate chemistry discoveries has been a topic of academic research since at least 2023. |
| SP003 | Forbes | Former OpenAI Researcher To Raise $500 Million For AI Science Startup | The aggressive fundraise marks a meteoric rise for the San Francisco-based startup, which emerged in September last year with a $300 million seed round at a $1.3 billion valuation. |
| SP004 | Tech Funding News | AI search engine for new materials nears $200M raise to cross $1B valuation: report | CuspAI calls its platform 'a search engine for the material world.' Users can enter the material properties they need, such as strength, conductivity, or thermal tolerance, and the system suggests possible chemical compositions up to ten times faster than traditional methods. |
| SP005 | QPillars | Self-Driving Labs in 2026 — What Actually Works vs. What's Still Hype | |
| SP006 | Drug Discovery and Development | SLAS 2026: Orchestration platforms, API-first instruments and the rise of semiautonomous labs | The lab OS wars: 15 companies vying to enable AI-enabled labs at SLAS 2026... the closed-loop lab is now a vendor selection decision, not a science fiction concept. |
| SP007 | U.S. Securities and Exchange Commission | Schrödinger Inc. — Exhibit 99.1, Q1 2026 Financial Results (8-K) | First Quarter ACV of $28 Million, Representing 12% Growth. Schrödinger to Launch Bunsen, an Agentic AI Co-Scientist, This Summer. |
| SP008 | Recursion Pharmaceuticals | Recursion Reports First Quarter Financial Results and Provides Business Update | |
| SP009 | Citrine Informatics | Citrine Informatics — Homepage | Citrine Informatics is the world leader in generative AI for materials and chemicals product development. |
| SP010 | Emerald Cloud Lab | Emerald Cloud Lab — Remote Controlled Life Sciences Lab | ECL® is equipped with over 200 different instrument models remotely controlled by a single unified software interface. |
| SP011 | Observer | Jeff Bezos-Backed Startup Receives $300M Seed Round to Build an A.I. Scientist | Periodic Labs is not alone in its mission. Tech giants like OpenAI and Google are pursuing similar goals... Smaller rivals include FutureHouse, a San Francisco nonprofit also working to create an autonomous A.I. scientist. |
| SP012 | Google DeepMind | Millions of new materials discovered with deep learning | We share the discovery of 2.2 million new crystals — equivalent to nearly 800 years' worth of knowledge. |
| SP013 | Tech Funding News | Former OpenAI and DeepMind researchers seek $7B valuation to build 'AI scientists' | Unlike many peers, the company is generating revenue. |
| SP014 | RobotToday | Laboratory Robotics in 2026: Technology, Companies and Vertical Maturity | The 'ChatGPT moment' for physical lab AI is a 2028–2030 event, not an imminent one. |
| SP015 | Business Wire / ResearchAndMarkets | Materials Informatics Global Market Report 2025-2035 | The core value proposition driving this growth is the dramatic reduction in materials development timelines. Traditional approaches typically require 10-20 years from concept to commercialization, whereas MI-enabled methods can potentially compress this to 2-5 years. |
| SP016 | MIT Technology Review | Google DeepMind's new AI tool helped create more than 700 new materials | GNoME can be described as AlphaFold for materials discovery... Thanks to GNoME, the number of known stable materials has grown almost tenfold, to 421,000. |
| SP017 | BioSpace | Schrödinger Reports First Quarter 2026 Financial Results | |
| SP018 | Startups Union | Periodic Labs Raised $300 Million-But Why?: Revolutionizing AI-Driven Materials Discovery | |
| SP019 | Avantgarde News | Periodic Labs Raising $500M for AI Science Startup | |
| SP020 | Tracxn | CuspAI — 2026 Company Profile, Team, Funding and Competitors | |
| SP021 | Andreessen Horowitz (a16z) | Investing in Periodic Labs | |
| SP022 | MIT Technology Review | AI materials science discovery startups investment 2025 | |
| SP023 | VIA News | Periodic Labs Faces $300M Burn Risk on Unproven AI Materials Timeline | Periodic Labs' $300M burn risk on an unproven AI-materials timeline—no clear commercial timeline for superconductor breakthroughs. |
| SP024 | Bloomberg | AI Science Startup Periodic Labs Is in Deal Talks at About $7 Billion Valuation | |
| SP025 | Recursion Pharmaceuticals (via MarketBeat) | Recursion Pharmaceuticals Details AI-Driven Drug Pipeline — Sanofi/Roche Milestones, Runway to 2028 | |
| SP026 | Investing.com | Schrödinger Q1 2026 slides: AI platform advances amid transition | |
| SP027 | Startupresearcher.com | CuspAI in Talks for $200M Round at Unicorn Valuation | |
| SI001 | Forbes | Former OpenAI Researcher To Raise $500 Million For AI Science Startup | "The round will be at least $500 million, two of the sources said. The round was 'significantly oversubscribed' and there are already talks for a fast-follow additional round at even higher valuation." |
| SI002 | TechCrunch | Former OpenAI and DeepMind researchers raise whopping $300M seed to automate science | "Periodic Labs came out of stealth on Tuesday with a war chest of $300 million as a seed round, backed by a tech industry who's who: Andreessen Horowitz, DST, Nvidia, Accel, Elad Gil, Jeff Dean, Eric Schmidt, and Jeff Bezos." |
| SI003 | ViaNews Markets | Periodic Labs faces $300M burn risk on unproven AI materials timeline | "Seed-stage companies rarely handle nine-figure rounds. The capital structure suggests investors expect rapid deployment into computational infrastructure, lab facilities, and talent acquisition. Monthly burn rates could reach $10-15 million before generating revenue." |
| SI004 | Andreessen Horowitz | Investing in Periodic Labs | "Periodic is already working with customers in space, defense, and semiconductors — sectors representing trillions in R&D spend. They're helping semiconductor manufacturers solve heat dissipation problems, training agents to automate simulations." |
| SI005 | Tech Funding News | Former OpenAI and DeepMind researchers seek $7B valuation to build 'AI scientists' | "What sets Periodic Labs apart is its early commercial traction. While many companies in this space remain in research mode, Periodic has already secured customers in the semiconductor industry. And unlike many peers, the company is generating revenue." |
| SI006 | SiliconAngle | Periodic Labs raises $300M to accelerate scientific research with AI | |
| SI007 | AI Insider | Periodic Labs Emerges from Stealth with $300 Million Seed Round to Build 'AI Scientists' | "The startup—co-founded by William Fedus and Ekin Dogus Çubuk—said the financing will fund hiring, scale out its laboratory infrastructure, and bring its first products to industry partners." |
| SI008 | AI Market Watch | Periodic Labs is in talks to raise at a ~$7B valuation — 5x its $1.3B seed from just 6 months ago | "The ~40-person startup already counts semiconductor customers generating revenue." |
| SI009 | UpsideList | Periodic Labs — Company Analysis | "Investors hold $300M in liquidation preferences ahead of common stock. In an exit at or below the current $1.2B valuation, common stock holders would see returns only after the initial $300M is returned to preferred shareholders." |
| SI010 | Nextomoro | Periodic Labs | |
| SI011 | Startups Union | Business model of Periodic Labs | "Revenue model not publicly disclosed yet (still in deep R&D phase). Likely future paths: 1) Material licensing: Discover breakthrough materials and license IP to manufacturers. 2) Contract research: Companies pay to access autonomous lab capabilities." |
| SI012 | Startups Union | Periodic Labs Raised $300 Million — But Why?: Revolutionizing AI-Driven Materials Discovery | |
| SI013 | Let's Data Science | Periodic Labs Seeks $500 Million Science Funding | |
| SI014 | Observer | Jeff Bezos-Backed Startup Receives $300M Seed Round to Build an A.I. Scientist | "The startup has already begun partnering with semiconductor makers to improve chip heat dissipation and is training agents to streamline research and engineering workflows. Its customer base also includes companies in space and defense." |
| SI015 | Avantgarde News | Periodic Labs to Raise $500 Million for AI-Driven Scientific Research | |
| SI016 | Emerald Cloud Lab | The Most Cost Effective Lab Space — Startup Comparison | |
| SI017 | PubMed Central (PMC) / PLOS Biology | Support academic access to automated cloud labs to improve reproducibility | "The cost to enter is high (>$250k for general access to Emerald Cloud Lab, or >$100k to automate and run a single method at Strateos), and the contract lengths are long (one year minimum)." |
| SI018 | arXiv | Boom, Bubble, or Bust? Dynamics of AI Investment in 2025–2026 | |
| SI019 | TechBuzz | Periodic Labs Raises Record $300M Seed to Build AI Scientists | |
| SI020 | Periodic Labs | Periodic Labs — Official Website | "We're also working to deploy our solutions with industry. As an example, we're helping a semiconductor manufacturer that is facing issues with heat dissipation on their chips. We're training custom agents for their engineers and researchers to make sense of their experimental data in order to iterate faster." |
| SI021 | U.S. Securities and Exchange Commission | Form D: AGC Wealt Periodic Labs I — a Series of AGC AI Nexus Fund LLC | |
| SI022 | Bloomberg | AI Science Startup Periodic Labs Is in Deal Talks at About $7 Billion Valuation | |
| SI023 | CBInsights | Periodic Labs Stock Price, Funding, Valuation, Revenue & Financial Statements | |
| SI024 | PitchBook | Periodic Labs 2026 Company Profile: Valuation, Funding & Investors | |
| SI025 | Nextomoro | Periodic Labs — Funding and Backers | "The seed round investor base is unusually concentrated in senior AI-industry figures. Andreessen Horowitz and DST Global led with venture-capital scale. Strategic investors included NVIDIA." |
| SE001 | Periodic Labs | Periodic Labs Official Launch Post — From Bits to Atoms | At Periodic, we are building AI scientists and the autonomous laboratories for them to operate. |
| SE002 | TechCrunch | Former OpenAI and DeepMind researchers raise whopping $300M seed to automate science | Cubuk led the materials and chemistry team at Google Brain and DeepMind, where one of his projects was GNoME. That tool discovered over 2 million new crystals in 2023. |
| SE003 | Andreessen Horowitz | Investing in Periodic Labs | The models will read literature, run quantum mechanical simulations, take action in the lab, and get feedback from nature itself. |
| SE004 | Forbes | Former OpenAI Researcher To Raise $500 Million For AI Science Startup | The round was 'significantly oversubscribed' and there are already talks for a fast-follow additional round at even higher valuation. |
| SE005 | Maginative | Periodic Labs launches with $300M to build an 'AI scientist' | |
| SE006 | AIM Media House | Can new AI venture Periodic Labs revolutionize discovery? | |
| SE007 | ViaNews | Periodic Labs Raises $300M Seed Round for AI Materials Discovery, Faces Decade-Long Validation Timeline | Traditional development cycles average 10-20 years from laboratory to commercial deployment worldwide. |
| SE008 | Startups Union | Periodic Labs Raised $300 Million—But Why? Revolutionizing AI-Driven Materials Discovery | |
| SE009 | Complete AI Training | Periodic Labs raises $300M to build autonomous AI labs for high-temperature superconductors | |
| SE010 | AI Insider | Periodic Labs Emerges from Stealth with $300 Million Seed Round to Build 'AI Scientists' | |
| SE011 | Periodic Labs via Ashby | Periodic Labs Jobs — Atoms and Bits open roles (accessed June 2026) | Atoms: Automation Engineer, Process Engineer Powder, Research Scientist Materials Synthesis; Bits: Distributed Training Engineer, ML Systems Engineer, Supercompute Engineer. |
| SE012 | Observer | Jeff Bezos-Backed Startup Receives $300M Seed Round to Build an A.I. Scientist | |
| SE013 | arXiv / Sakana AI | The AI Scientist-v2: Workshop-Level Automated Scientific Discovery via Agentic Tree Search | We introduce The AI Scientist-v2, an end-to-end agentic system capable of producing the first entirely AI generated peer-review-accepted workshop paper. |
| SE014 | GitHub / Sakana AI | SakanaAI/AI-Scientist-v2 — Workshop-Level Automated Scientific Discovery | |
| SE015 | TechCrunch | Top OpenAI, Google Brain researchers set off a $300M VC frenzy for their startup Periodic Labs | Robotic arms that could handle powder synthesis had recently proved themselves reliable. Machine learning simulations had become efficient and accurate enough to model complex physical systems. |
| SE016 | The Outpost AI | Periodic Labs Raises $300M to Create AI-Powered Scientific Research Platform | |
| SE017 | Avantgarde News | Periodic Labs to Raise $500 Million for AI-Driven Scientific Research | |
| SE018 | Nextomoro / AI Research Lab Intelligence | Periodic Labs — Company Profile and Strategic Overview | Open weights: None. The Periodic First Release model is closed; broader product strategy has not been disclosed. |
| SE019 | Datamation | Periodic Labs Powers Up for 'Scientific AI Advances' | |
| SE020 | Tech Funding News | Former OpenAI and DeepMind researchers seek $7B valuation to build 'AI scientists' | Unlike many companies in this space that remain in research mode, Periodic has already secured customers in the semiconductor industry. |
| SE021 | Bloomberg | AI Science Startup Periodic Labs Is in Deal Talks at About $7 Billion Valuation | |
| SE022 | Vogon Today / StartMag (translating New York Times) | Why top AI researchers are leaving OpenAI, Google, and Meta — NYT Report | More than 20 researchers who in recent weeks have left their jobs at Meta, OpenAI, Google DeepMind, and other major AI projects to join Periodic Labs. Many gave up tens of millions, if not hundreds of millions, in equity. |
| SE023 | Forbes | The AI 50 Brink List 2026 | His startup Periodic Labs is training models to accelerate scientific discovery in semiconductors, magnetism and superconductivity. |
| SE024 | AnalyzedNews (citing New York Times) | Top A.I. Researchers Leave OpenAI, Google and Meta for New Start-Up | |
| SE025 | ViaNews (second reference) | Periodic Labs risks and commercial-model analysis — decade-long validation timeline | AI applications in materials science show limited commercial success. Physical validation requirements create bottlenecks AI cannot eliminate—predictions require experimental confirmation regardless of computational speed. |
| SU001 | Periodic Labs | Periodic Labs — Official Homepage and Company Introduction | "We're also working to deploy our solutions with industry. As an example, we're helping a semiconductor manufacturer that is facing issues with heat dissipation on their chips. We're training custom agents for their engineers and researchers to make sense of their experimental data in order to iterate faster." |
| SU002 | Andreessen Horowitz (a16z) | Investing in Periodic Labs | "Periodic is already working with customers in space, defense, and semiconductors – sectors representing trillions in R&D spend. They're helping semiconductor manufacturers solve heat dissipation problems, training agents to automate simulations, and building systems that encode deep domain knowledge through mid-training and reinforcement learning." |
| SU003 | Forbes | Former OpenAI Researcher To Raise $500 Million For AI Science Startup | "The round was 'significantly oversubscribed' and there are already talks for a fast-follow additional round at even higher valuation." |
| SU004 | Observer | Jeff Bezos-Backed Startup Receives $300M Seed Round to Build an A.I. Scientist | "The startup has already begun partnering with semiconductor makers to improve chip heat dissipation and is training agents to streamline research and engineering workflows. Its customer base also includes companies in space and defense." |
| SU005 | Inc. Magazine | This Startup Emerged From Stealth With $300 Million to Create an 'AI Scientist' | "Fedus said that Periodic's ideal customers are engineers and researchers in advanced industries like space, defence, and semiconductors. These engineers and researchers 'don't really have particularly good tools,' said Fedus, 'and that is our opportunity. These are massive R&D budgets.'" |
| SU006 | SiliconANGLE | Periodic Labs raises $300M to accelerate scientific research with AI | |
| SU007 | Maginative | Periodic Labs launches with $300M to build an 'AI scientist' | "Periodic also cites work with an unnamed semiconductor manufacturer struggling with chip heat dissipation, where its agents are helping engineers interpret experimental data and test fixes faster. Critics note that real-world usefulness — beyond stability predictions — is still limited." |
| SU008 | Nextomoro | Periodic Labs — Company Profile and Analysis | "The company's product roadmap beyond the autonomous-lab platforms has not been disclosed. Whether Periodic Labs licenses its AI scientist capability, develops materials internally for direct commercialization, or operates as a research-services partner for existing industrial customers is an open question." |
| SU009 | AI Market Watch | Periodic Labs is in talks to raise at a ~$7B valuation | "The ~40-person startup already counts semiconductor customers generating revenue." |
| SU010 | Bromley Capital Partners | [Transactions Announced] – Periodic Labs – Jan/2026 | "Bromley Capital Partners (UK) successfully advised on a private placement into Periodic Labs. The multi-million dollar transactions were successfully concluded in Jan/2026." |
| SU011 | Deep Tech Week | Periodic Labs | Deep Tech Week Organization Profile | |
| SU012 | ViaNews Market | Periodic Labs faces $300M burn risk on unproven AI materials timeline | "Past AI-materials ventures show a 60-70% failure rate in reaching commercial production, even with promising lab results." |
| SU013 | AI Via News | Periodic Labs Faces Commercial Pressure After $300M Seed Round for AI Materials Discovery | "Traditional materials discovery from lab to market averages 10-20 years. AI acceleration may compress this timeline, but no companies have yet demonstrated commercial-scale success in AI-discovered materials." |
| SU014 | MIT Technology Review | AI materials discovery now needs to move into the real world | "So far there has been no 'eureka' moment, no ChatGPT-like breakthrough — no discovery of new miracle materials or even slightly better ones. By far the most time-consuming and expensive step in materials discovery is not imagining new structures but making them in the real world." |
| SU015 | UpsideList | Periodic Labs — Company Analysis | "Bear (25%): Periodic Labs faces significant challenges in commercializing its AI-discovered materials or scaling its autonomous labs, leading to slower-than-expected revenue growth." |
| SU016 | VentureRadar | Similar companies to Periodic Labs | |
| SU017 | The Outpost AI | Periodic Labs Raises $300M to Create AI-Powered Scientific Research Platform | |
| SU018 | Forbes Australia | Forbes AI 50 Brink List: 20 Startups Shaping the Future of AI | "His startup Periodic Labs is training models to accelerate scientific discovery in semiconductors, magnetism and superconductivity." |
| SU019 | Bloomberg | AI Science Startup Periodic Labs Is in Deal Talks at About $7 Billion Valuation | |
| SU020 | ZoomInfo | Periodic Labs — Company Profile | |
| SU021 | TechCrunch | Top OpenAI, Google Brain Researchers Set Off a $300M VC Frenzy for Their Startup, Periodic Labs | "Periodic Labs has already set up its lab, too, and is working with experimental data, simulations and testing some predictions... the robots — are not yet up and running." |
| SU022 | Felicis Ventures | Our Investment in Periodic Labs | "Liam said something to me on that walk that immediately resonated: 'In order to do science, you have to do real science.' They're combining compute with a physical lab to turn AI into an engine for discovery." |
| SU023 | Wilson Sonsini Goodrich & Rosati | Wilson Sonsini Advises Periodic Labs on $300 Million Seed Round | |
| SU024 | TechFundingNews | Former OpenAI and DeepMind Researchers Eye $7B Valuation for AI Startup Periodic Labs | "What sets Periodic Labs apart is its early commercial traction. While many companies in this space remain in research mode, Periodic has already secured customers in the semiconductor industry... And unlike many peers, the company is generating revenue." |
| SU025 | U.S. Securities and Exchange Commission (EDGAR) | Form D — AGC Wealt Periodic Labs I, a Series of AGC AI Nexus Fund LLC | |
| SR001 | TechCrunch | Periodic Labs Raises Whopping $300M Seed to Automate Science | |
| SR002 | Google DeepMind | Millions of new materials discovered with deep learning | |
| SR003 | ViaNews Market | Periodic Labs Faces $300M Burn Risk on Unproven AI Materials Timeline | |
| SR004 | Observer | Periodic Labs Launches $300M to Build Real Science AI | |
| SR005 | Startups Union | Periodic Labs Raised $300 Million — But Why? | |
| SR006 | arXiv | AI Hallucination in Scientific Text: Taxonomy and Detection | |
| SR007 | Via News AI | Periodic Labs Faces Commercial Pressure After $300M Seed Round for AI Materials | |
| SR008 | The AI Insider | Periodic Labs Emerges from Stealth with $300 Million Seed Round to Build AI Scientists | |
| SR009 | Science Reader | AI Materials Discovery: 5 Things to Know | |
| SR010 | US OSHA | OSHA Laboratory Safety Standards (29 CFR 1910.1450) | |
| SR011 | US OSHA | OSHA Robotics Safety Standards and Guidelines | |
| SR012 | Sterne Kessler | 2025 AI Intellectual Property Year in Review: Analysis and Trends | |
| SR013 | Arms Control Association | Regulatory Gaps in Benchtop Nucleic Acid Synthesis Create Biosecurity Vulnerabilities | |
| SR014 | CSIS | Opportunities to Strengthen US Biosecurity from AI-Enabled Bioterrorism | |
| SR015 | Nextwaves Insight | AI Materials Discovery: GNoME, MatterGen and the Road Ahead in 2026 | |
| SR016 | Arnold and Porter | Biosecurity Compliance and Risk Management | |
| SR017 | Lab Safety Institute | Lab Safety in Review: Major Changes in 2025 and What Is Ahead in 2026 | |
| SR018 | Nature | Advances in AI-driven materials design and synthesis | |
| SR019 | FAF.ae | Biosecurity and Oversight Gaps in Dual-Use Biotechnology: A Crisis of Governance | |
| SR020 | Osium AI | Understanding GNoME: Opportunities and Limitations in AI Materials Discovery | |
| SR021 | Nextomoro | Periodic Labs: Company Profile and Overview | |
| SR022 | MIT Technology Review | AI Materials Science Discovery Startups: Investment and the Road to Commercialization | |
| SR023 | PitchBook | Discovering New Materials with AI Has a Winding Road to VC Returns | |
| SR024 | US OSHA | OSHA Accident Investigation Search: Robotics Fatalities | |
| SR025 | US Chemical Safety Board | Chemical Safety Board Investigation Database | |
| SR026 | Microsoft Research | MatterGen: A New Paradigm of Materials Design with Generative AI | |
| SR027 | Success Quarterly | Periodic Labs Secures Record $300M for Autonomous AI Science | |
| SR028 | Information Technology and Innovation Foundation | Scaling Materials Discovery with Self-Driving Labs | |
| SR029 | Murgitroyd | Intellectual Property Trends and Developments Looking to 2026 | |
| SR030 | RAND Corporation | RAND Research Report on Autonomous AI Laboratory Biosecurity | |
| SR031 | Avantgarde News | Startups Raise $1.3B for AI Materials Discovery in 2026 | |
| SR032 | World Economic Forum | AI Materials Innovation: From Discovery to Design | |
| SR033 | Yahoo Finance | Exclusive: Orbital Industries Startup Using AI for Materials Manufacturing | |
| SV001 | Forbes | Former OpenAI Researcher To Raise $500 Million For AI Science Startup | Periodic Labs, which debuted on the Forbes AI 50 Brink list this year, will see its value increase nearly sixfold in less than eight months. |
| SV002 | Bloomberg | AI Science Startup Periodic Labs Is in Deal Talks at About $7 Billion Valuation | |
| SV003 | TechCrunch | Top OpenAI, Google Brain researchers set off a $300M VC frenzy for their startup Periodic Labs | The other seed investors include DST, Nvidia's venture capital arm NVentures, Accel, and angel backers like Jeff Bezos, Elad Gil, Eric Schmidt, and Jeff Dean. |
| SV004 | Wilson Sonsini Goodrich & Rosati | Wilson Sonsini Advises Periodic Labs on $300 Million Seed Round | On September 30, 2025, Periodic Labs announced a $300 million Seed round led by Andreessen Horowitz. |
| SV005 | Tech Funding News | Former OpenAI and DeepMind researchers seek $7B valuation to build 'AI scientists' | The AI startup Periodic Labs is in early discussions to raise hundreds of millions of dollars at a valuation of around $7 billion, as per Bloomberg. |
| SV006 | Finro Financial Consulting | AI Valuation Multiples (Q1 2026) | 575 Company Dataset | LLM Vendors: 27 companies, Avg EV/Rev 73.5x, Median EV/Rev 39.5x. Seed stage AI: 105 companies, Median 20.2x. |
| SV007 | TechCrunch | Alphabet's AI drug discovery platform Isomorphic Labs raises $600M from Thrive | Isomorphic Labs, the AI drug-discovery platform that was spun out of Google's DeepMind in 2021, has raised external capital for the first time. |
| SV008 | TechCrunch | Sakana AI raises $135M Series B at a $2.65B valuation to continue building AI models for Japan | Sakana AI has closed a ¥20 billion (approximately $135 million) Series B funding round, which values the company at $2.65 billion post-money. |
| SV009 | CNBC | Anthropic tops OpenAI as most valuable AI startup, nears $1T valuation | |
| SV010 | Qubit Capital | AI Startup Valuation Multiples: 10x-50x Range (2026) | |
| SV011 | PR Newswire (Isomorphic Labs) | Isomorphic Labs announces $600 million funding to further develop its next-generation AI drug design engine | Isomorphic Labs, an AI-first drug design and development company, today announced it has raised $600 Million in its first external funding round. |
| SV012 | DealForma | AI-ML Drug Discovery and Licensing R&D, M&A, Ventures and IPOs – 2025 Review | In 2025, venture funding for AI-ML drug discovery and licensing increased materially, with 348 financing rounds raising $11 billion. |
| SV013 | Vision Life Sciences | Biotech Funding and IPO Landscape 2026 | Recovery | |
| SV014 | Gartner | Gartner Forecasts Worldwide Semiconductor Revenue to Exceed $1.3 Trillion in 2026 | Global semiconductor revenue is projected to exceed $1.3 trillion in 2026, exhibiting the highest growth in the last two decades. |
| SV015 | PricewaterhouseCoopers | Technology: US Deals 2026 Outlook – AI-Fueled M&A | Strategic buyers will keep chasing hard-to-get AI capabilities to stay ahead. Expect more tuck-ins and acquihires focused on proprietary data, scalable tooling, and specialized engineering talent. |
| SV016 | Maginative | Periodic Labs launches with $300M to build an AI scientist | Periodic Labs' pitch is straightforward: AI models have exhausted the internet's finite troves of text and code. The next breakthrough requires giving AI the means to generate new knowledge itself. |
| SV017 | Isomorphic Labs | Isomorphic Labs announces $600m external investment round | |
| SV018 | Finerva | Robotics and AI: 2026 Valuation Multiples | After bottoming out at 2.5x in Q1 2025, valuation multiples staged a recovery throughout the last year. The median revenue multiple rose steadily from 2.5x in the first quarter to 3.4x by Q4 2025. |
| SV019 | She Talks AI | Periodic Labs Seeks $500 Million Funding at $7.5 Billion Valuation for AI Scientific Discovery | |
| SV020 | AI Insider | Periodic Labs Emerges from Stealth with $300 Million Seed Round to Build AI Scientists | Periodic Labs emerged from stealth with a $300 million seed round led by Andreessen Horowitz (a16z), a bet that AI scientists and autonomous labs can accelerate discoveries in materials and other physical sciences. |
| SV021 | AI Market Watch | Periodic Labs is in talks to raise at a ~$7B valuation — 5x its $1.3B seed from just 6 months ago | Periodic Labs is in talks to raise at a ~$7B valuation — 5x its $1.3B seed from just 6 months ago. |
| SV022 | U.S. Securities and Exchange Commission (EDGAR) | Form D: AGC Wealt Periodic Labs I, a Series of AGC AI Nexus Fund LLC (CIK 0002122824) | AGC Wealt Periodic Labs I a Series of AGC AI Nexus Fund LLC; Venture Capital Fund; Amount Sold: $4,743,803; Total number of investors: 7 |
| SV023 | Sacra | Sakana AI valuation, funding and news | In November 2025, Sakana AI raised 20 billion yen (US$135 million) in its Series B, establishing a post-money valuation of approximately 400 billion yen (US$2.635 billion). |
| SV024 | ETCentric | AI Startup Periodic Labs Raises $300M for Scientific Research | The company's mission is to pair large AI models with physical laboratories, leveraging autonomous systems to conduct experiments and generate proprietary new data, especially in the physical sciences. |
| SV025 | Nextomoro | Periodic Labs | |
| SV026 | Crowdfund Insider | Tech Exits In H1 2025: AI M&A Surges, IPOs Loom, Secondaries Gain Traction | |
| SV027 | TechBuzz AI | Periodic Labs Raises Record $300M Seed to Build AI Scientists | |
| SV028 | LetsDataScience | Isomorphic Labs Raises $2.1 Billion Series B for AI Drug Discovery | |
| SV029 | AgentMarketCap | Three Labs Raised $172B in Q1 2026. Here's What That Means for Everyone Else. | |
| SV030 | CXO Digital Pulse | Former OpenAI and Google Brain Researchers Launch Periodic Labs with $300 Million Seed Funding |