OpenAI
由使命治理的前沿 AI 平台与应用公司
OpenAI 在前沿 AI 的体量和需求上都拿得出真凭实据,但 8520 亿美元的入场估值还是跑在了公开披露的前面。
封面要素
公司概况
OpenAI 是一家由使命治理的前沿 AI 公司,通过非营利组织控制的公益公司把研究商业化。它兼具罕见的产品触达和融资能力,也背着不寻常的治理复杂度、正在进行的诉讼,以及必须单独承销的基础设施依赖。
- 成立时间
- 2015-12-01
- 创始人
- Sam Altman, Greg Brockman
- 创立地点
- San Francisco, California, USA
- 总部
- San Francisco, California
- 产品
- OpenAI 通过 ChatGPT、API、Azure OpenAI Service 和面向企业的产品商业化前沿模型,覆盖范围越来越像一套同时服务消费者、开发者和企业工作流的统一栈。
- 客户
- 消费者、开发者、企业,以及与 Azure 相连的合作伙伴渠道。
- 商业模式
- 订阅、按用量计费的 API 收入、企业合同,以及伙伴渠道带来的分发。
- 阶段
- late-stage private
- 融资情况
- 私募融资公司;最近一次公开融资是 2026 年 3 月完成的 1,220 亿美元融资,投后估值 8,520 亿美元。
执行摘要
主要优势
- ChatGPT、API、企业三条渠道一齐拿出超常的需求证据。
- 分发与资金通路依旧顶级,Microsoft 加上后续大额轮次的投资人持续把产能往上推。
- 产品栈在消费者、开发者、企业三类工作流之间收敛,而不是分裂。
主要风险
- 版权、隐私、治理三条线上的法律与监管暴露仍然偏高。
- 算力、云和融资集中在少数合作方手里,合作方经济性一变,下行风险就会集中释放。
- 8520 亿美元入场价下,毛利率、烧钱、合同条款与融资权利的公开披露仍然不够看。
未决问题
- 经审计的分部收入、毛利率、烧钱速度与现金余额仍未公开。
- 2026 年那轮融资的优先股堆叠、清算权利、棘轮条款以及任何老股交易机制,全都没有公开。
- 各家基础设施合作方的承诺消费、预付与照付不议(take-or-pay)排程仍未披露。
- 客户黏性指标——净留存率(NRR)、流失率、合同期限、头部客户集中度——也都还缺。
目录
01公司概览
1.1 身份、使命与运营模式
OpenAI 将自己定义为一家 AI 研究与部署公司,使命是确保通用人工智能惠及全人类。在现有架构下,这一使命由两个相连实体承载:非营利的 OpenAI Foundation,以及负责把公司模型和产品商业化的营利性公益公司 OpenAI Group PBC。Foundation 治理 Group,而不只是被动持股;这和传统 VC 支持的创业公司有实质差异,后续章节也应把它作为核心尽调事实反复使用。 运营模式之所以重要,是因为 OpenAI 同时是研究实验室、消费级产品公司、API 平台和企业基础设施依赖。公开材料显示,公司靠 ChatGPT、API、企业产品和 Azure 相关渠道分发模型,同时保留一套使命治理结构,意在把商业化和更广泛的公共利益目标绑在一起。这套结构不是门面:OpenAI 称 Foundation 任命 OpenAI Group 董事会并可更换董事,而 Group 本身按公益公司组织,需要考虑更广泛的利益相关方。结果是,分析 OpenAI 时必须同时把它看作高速扩张的软件平台,也看作对治理高度敏感的机构。 [CO001, CO002, CO003, CO004, CO005, CO006]
1.2 治理、领导层与关键人依赖
治理仍是最重要的承销变量之一。官方材料显示,现任董事会成员包括 Bret Taylor、Adam D’Angelo、Sue Desmond-Hellmann、Zico Kolter、Paul Nakasone、Adebayo Ogunlesi、Nicole Seligman 和 Sam Altman。OpenAI 2024 年 3 月的审查材料很关键,因为它把 2023 年董事会危机从传闻变成可复用的规范事实:WilmerHale 审查称,裂痕源于前董事会与 Altman 之间的信任崩塌,而不是产品安全、安全、融资或合作伙伴披露问题。该结论降低了一类生存性担忧,但另一类担忧仍在:领导权集中和治理执行风险。 关键人地图仍然很窄。Sam Altman 仍是核心高管和公共面孔;Greg Brockman 升任总裁,继续贴近技术关键路径;Fidji Simo 后来加入并负责 Applications,补强了商业侧执行深度;2024 和 2025 年新增董事也被明确包装为加强治理。即便有这些变化,证据仍指向一家公司:战略、融资、算力伙伴关系,以及外部利益相关方信任,仍异常集中在少数人手里。对前沿 AI 领导者来说,这种集中不一定构成否决项,但它是后续产品、风险和估值工作中的重大尽调事项。 [CO007, CO008, CO009, CO010, CO011, CO012]
| 人员 | 角色 | 背景 | 创始人-市场匹配或职能覆盖 | 关键人物依赖 |
|---|---|---|---|---|
| Sam Altman | CEO 兼董事会成员 | 联合创始人;主要负责战略、融资和外部信任叙事 | 连接使命叙事、投资者关系、产品方向和公共定位 | 高 |
| Greg Brockman | President | 联合创始人;仍贴近技术关键路径和公司战略 | 连接前沿研究执行和公司建设的重要连续性人物 | 高 |
| Bret Taylor | 董事会主席 | 2023 年危机后加入的独立治理负责人 | 稳定董事会流程和外部利益相关方信心的关键人物 | 中 |
| Fidji Simo | Applications CEO | Instacart CEO、前董事会成员,被引入以扩大应用业务 | 为商业化和执行增加资深运营深度 | 中 |
| Adebayo Ogunlesi | 董事 | 来自 GIP/BlackRock 的基础设施和金融资深人士 | 增加与 AI 规模化相关的资本市场和基础设施监督能力 | 中 |
本表有意聚焦当前对治理和执行依赖最重要的创始人及资深领导者。
[CO007, CO008, CO009, CO010, CO012, CO013]1.3 资本基础、利益相关方地图与公开规模
OpenAI 的资本画像已经自成一档。公开报道和公司材料支撑了一条清晰序列:从 2023 年 1 月 Microsoft 合作伙伴关系延长,到 2024 年 10 月以 1,570 亿美元投后估值完成 66 亿美元融资,再到 CNBC 和 Bloomberg 报道的 2026 年 3 月融资,规模 1,220 亿美元、投后估值 8,520 亿美元。同一批材料也指向一套利益相关方结构:资本重组后,OpenAI Foundation 持有 OpenAI Group 约 26%,Microsoft 约 27%;Amazon、Nvidia、SoftBank、Thrive 及其他财务投资人则对 OpenAI 未来融资能力和谈判筹码产生实质影响。 公开规模指标亮眼,但支撑力度并不均匀。Forbes 将 OpenAI 总部列为旧金山,称截至 2025 年 12 月有 6,400 名员工,并报告 2025 年营收 37 亿美元。同一公司档案称,ChatGPT 周活用户约 5 亿,付费企业用户约 200 万。这些数字有助于判断成熟度和规模,但公开报道仍未给出完全对齐的股权结构表、精确的累计合并融资总额,也没有金融承销人想看的已实现定价、利润率或 cohort 质量颗粒度。因此,后续章节应把这些数字作为公开参照点,而不是把它们当成完整数据室资料的替代品。 [CO015, CO016, CO020, CO021, CO022, CO023]
| 指标 | 数值/状态 | 日期 | 置信度 | 缺口 |
|---|---|---|---|---|
| 创立年份 | 2015 | 2015 | 中 | |
| 总部 | San Francisco, California | 2025-12 | 中 | |
| 运营结构 | OpenAI Foundation 控制 OpenAI Group PBC | 中 | ||
| 最新公开估值(USD B) | 852 | 2026-03-31 | 中 | |
| Foundation 持股比例(%) | 26 | 2025-10-28 | 中 | |
| Microsoft 持股比例(%) | 27 | 2025-10-28 | 中 | |
| 员工 | 6400 | 2025-12 | 中 | |
| 2025 年收入(USD B) | 3.7 | 2025-12 | 中 | |
| ChatGPT 周活用户 | 500000000 | 2025-03 | 中 | |
| 付费企业用户 | 2000000 | 2025-03 | 中 | |
| 累计融资额 | 2026-03-31 | 中 | 公开来源支持重大融资轮次,但在资本重组和后续银行渠道参与之后,仍无法完整对齐累计总额。 | |
| 办公地点 | 低 | 本章公开来源能支持总部信息,但不能支持完整办公室/地点版图。 |
公开规模和资本数据最适合作为当前外部参照点,不能替代尽调资料室里的内部指标。
[CO002, CO003, CO021, CO022, CO023, CO025]| 利益相关方 | 角色 | 控制权或经济重要性 | 尽调问题 |
|---|---|---|---|
| OpenAI Foundation | 控制性非营利组织 | 控制 OpenAI Group,持有约 26% 股权,并通过股权加 warrant 机制分享上行 | 确认董事会保留事项,以及 Foundation 权利如何与未来融资互动。 |
| Microsoft | 战略算力和资本伙伴 | 资本重组后约持有 27%,并根据 2023 年延伸协议担任独家云服务商 | 澄清算力和收入分成协议中的当前独家性、定价和终止经济性。 |
| Amazon | 2026 年融资锚定方 | Bloomberg 和 CNBC 称,Amazon 是 2026 年融资中最大参与方,金额为 $50B | 判断承诺中有多少取决于 IPO/AGI 里程碑,以及附带哪些治理权。 |
| Nvidia | 2026 年融资锚定方 | 据报道是 2026 年融资中 $30B 参与方,也是战略相关的算力供应商 | 区分纯财务暴露和任何硬件或供应链商业承诺。 |
| SoftBank | 2026 年融资锚定方 | 据报道是 2026 年融资中 $30B 参与方,也是 OpenAI 最新规模融资的主要赞助方 | 确认董事会、信息权或后续跟投权。 |
| Thrive Capital 和此前 syndicate | 2024 年财务支持方 | TechCrunch 识别出 Thrive 及广泛的 2024 年 syndicate,帮助 OpenAI 过渡到后续 mega-round | 将早期轮次的优先权、pro rata 和 clawback 条款与资本重组后的经济性对齐。 |
公开 cap table 不完整;本图谱捕捉最重要的已披露控制和融资利益相关方,而不是每一名持有人。
[CO015, CO016, CO020, CO021, CO022, CO023]| 日期 | 事件 | 类型 | 金额/估值/状态 | 参与方/来源 | 含义 |
|---|---|---|---|---|---|
| 2015 | 以造福 AGI 的使命创立为非营利组织 | 创立 | OpenAI 结构页面 | 使命优先的起点仍是当前治理设计的核心。 | |
| 2019 | 在 Foundation 控制下设立营利业务 | 治理 | 设立营利业务 | OpenAI 合作和结构页面 | 设立了后来资本重组为 OpenAI Group PBC 的载体。 |
| 2022-05-05 | 领导团队更新任命 Greg Brockman 为 president、Mira Murati 为 CTO | 治理 | 高管角色变化 | OpenAI 领导团队更新 | 显示公司在 ChatGPT 级商业化之前,开始正式搭建资深领导层。 |
| 2022-11-30 | ChatGPT 以研究预览形式发布 | 产品 | 研究预览发布 | OpenAI ChatGPT 发布文章 | 标志着商业化拐点,改变了 OpenAI 的规模和可见度。 |
| 2023-01-23 | OpenAI 与 Microsoft 扩展合作关系 | 合作 | 多年、数十亿美元投资 | OpenAI 和 Microsoft 文章 | 锁定资本、Azure 算力和分发杠杆。 |
| 2023-03-14 | GPT-4 研究发布 | 产品 | 旗舰前沿模型里程碑 | OpenAI GPT-4 研究页面 | 增强产品可信度和企业采用潜力。 |
| 2024-03-08 | WilmerHale 审查完成,董事会扩容 | 治理 | Altman 和 Brockman 获确认;新增董事 | OpenAI 审查和董事会文章 | 在 2023 年董事会危机后重置治理,并成为解释动荡的标准公开说法。 |
| 2024-10-02 | OpenAI 完成 $6.6B 融资 | 融资 | $6.6B,投后估值 $157B | TechCrunch | 在后续资本重组周期之前,提供了清晰的公开估值跃升。 |
| 2025-06-05 | OpenAI 公开反击 NYT 数据保留要求 | 负面 | 法律保留和隐私辩护回应 | OpenAI 诉讼回应 | 把隐私、版权和取证负担抬升为核心尽调议题。 |
| 2025-10-28 | 与 California 和 Delaware AG 办公室沟通后完成资本重组 | 监管 | OpenAI Foundation 26%;Microsoft 约 27% | OpenAI 结构和重组文章;Axios | 让罕见的 Foundation 控制 PBC 结构变得具体,并经过监管审视。 |
| 2025-12 | 公开规模快照达到 6,400 名员工和 $3.7B 收入 | 规模 | Forbes 2025 快照 | Forbes 公司画像 | 确认 OpenAI 已经以连大型软件公司都少见的私营公司规模运营。 |
| 2026-03-31 | 迄今公开最大融资轮完成 | 融资 | $122B,投后估值 $852B | CNBC 和 Bloomberg | 发出非凡融资能力信号,但也强化了市场对执行和最终流动性的期待。 |
本时间线是报告其他部分可复用的单一公开里程碑记录。
[CO003, CO004, CO010, CO011, CO013, CO014]公开可支撑的快照指标显示规模异常强,但还不是完全对账的私营公司 data room。
[CO021, CO022, CO023, CO028, CO029, CO030]1.4 里程碑、法律摩擦与可靠性背景
公司概览必须保留一条后续章节可引用、无需重新争论核心事实的时间线。公开记录支撑的顺序很清楚:2015 年以非营利组织创立,2019 年创建营利性业务,2022 年 5 月领导角色调整,2022 年 11 月推出 ChatGPT,2023 年 1 月 Microsoft 续签多年投资,2023 年 3 月发布 GPT-4,2024 年 3 月完成治理审查并扩充董事会,2024 年 10 月融资,2025 年 6 月公开回应 New York Times 数据保留要求,2025 年 10 月资本重组,2026 年 3 月完成融资。这是当前来源集下最好的一条单一记录时间线。 同一条时间线也说明,OpenAI 不是一个干净的“不惜代价增长”故事。状态历史页面显示,ChatGPT、API、登录、文件处理和特定模型服务反复出现服务事件;这很重要,因为 OpenAI 现在是企业和开发者的基础设施,而不再只是新奇应用。法律风险也不是历史问题,而是仍在发生:OpenAI 自己的诉讼回应页面、JD Supra 对 New York Times 证据保全争议的分析、Loeb 的版权诉讼说明,以及 Musk v. Altman 案件卷宗,都显示版权、隐私、非营利使命和治理争议仍是活跃尽调主题。结论不是业务已经坏掉,而是规模、治理和法律复杂度已经成为公司核心身份的一部分。 [CO013, CO015, CO017, CO018, CO023, CO031]
OpenAI 的公开记录从非营利起点走到 Foundation 控制的 PBC 和巨额轮融资,治理与诉讼仍在同一条时间线里。
[CO003, CO004, CO015, CO017, CO023, CO025]使命治理、资本伙伴、产品触点以及法律/可靠性约束共同喂给同一套操作系统。
[CO002, CO005, CO015, CO016, CO023, CO031]1.5 展示项
02市场分析
2.1 市场边界与规模测算口径
OpenAI 的相关市场不是整个 AI 技术栈。公开产品页显示,公司有三块直接变现界面:面向员工的 ChatGPT Business 和 Enterprise 套餐,面向开发者和软件团队、按 token 定价的 API 用量,以及面向大学的独立教育产品。这个边界很重要,因为 Gartner 对 2025 年生成式 AI 6,440 亿美元的预测,是一个宽口径、自上而下的支出口径,包含设备、服务器、软件和服务等类别;相比之下,Menlo 的 138 亿美元企业支出和 46 亿美元应用层支出,更接近真正流向基础模型应用和 copilots 的预算。投资判断中,实用市场边界应包括按席位收费的企业助手、按用量计费的开发者和 agent 工作流,以及高校等机构部署;应排除通用硬件更新、hyperscaler 基础设施 capex,以及不直接变现模型或助手的非 GenAI 软件模块。 McKinsey 的价值池研究仍有用,但更适合作为外层天花板,而不是直接收入池。其每年 2.6 万亿至 4.4 万亿美元的估算解释了为什么这个类别能拿到溢价估值,但实际支出仍远低于该上限,并且高度受组织变革、集成和治理制约。对 OpenAI 的关键含义是:公司受益于真实庞大的需求背景,但直接 SAM 应通过企业应用预算和模型用量预算来界定,而不是套用整个头条式 AI 经济。[CM001, CM002, CM004, CM008, CM009, CM016]
| 细分/类别 | 纳入支出 | 排除支出 | 买方/付款方 | 相关性 |
|---|---|---|---|---|
| 企业员工助手 | ChatGPT Business 和 Enterprise 的按席位订阅,加上工作流 agent 和知识工作 copilot | 通用办公 SaaS、设备更新和非 GenAI 生产力套件 | CIO、COO、中央 IT,以及为员工采购的业务线负责人 | OpenAI 企业助手界面的核心直接市场 |
| 开发者和 API 平台 | 按 token 定价的推理、编程 copilot、agent 工作流和应用层模型调用 | 超大云厂商服务器 capex、存储和无关开发者工具 | CTO、工程 VP、平台团队和软件预算 | OpenAI API 和编程产品的核心直接市场 |
| 高等教育部署 | 通过 ChatGPT Edu 为学生、教师、研究人员和运营团队提供校园级 AI 访问 | 通用 LMS 支出、校园硬件和非 AI 软件合同 | 教务长、CIO、采购和大学行政部门 | 商业企业销售之外的独立机构预算线 |
| 垂直工作流应用 | 支持、搜索、数据抽取、摘要和受监管工作流中的应用层支出 | 没有模型驱动功能的传统企业软件模块 | 职能负责人,加上 IT 和采购 | 间接但重要的需求渠道,因为许多应用构建在模型提供商之上 |
| 消费者和 prosumer 订阅 | 个人订阅和自助使用,培育熟悉度和工作流习惯 | 广告支持搜索、社交媒体和通用消费者互联网类别 | 个人专业人士和极小团队 | 有助于打开认知,但耐久性弱于托管式企业预算 |
| 广义 AI 基础设施相邻市场 | 广义市场报告计入的全球 GenAI 设备、服务器、服务和软件 | 当基础设施支出不流经 OpenAI 时的 OpenAI 直接产品收入 | 设备 OEM、超大云厂商、基础设施和服务买家 | 重要 TAM 背景,但大多落在 OpenAI 直接 SAM 之外 |
承销边界应跟着可直接流向 OpenAI、或通过基础模型应用需求流向 OpenAI 的货币化界面走,而不是把更广泛 AI 硬件和服务栈里的每一美元都算进去。
[CM001, CM002, CM003, CM004, CM005, CM008]| 来源 | 年份 | 地理范围 | 数值 | CAGR | 方法 | 置信度 | 局限 |
|---|---|---|---|---|---|---|---|
| Gartner | 2025 | Worldwide | 644 | 全球生成式 AI 支出的自上而下预测 | 中 | 广义市场镜头,包含设备、服务器、软件和服务,而不是只看 OpenAI 收入池。 | |
| VentureBeat 对 Gartner 的摘要 | 2025 | Worldwide | 76.4 | 对 Gartner 全球 GenAI 支出同比增长的二手摘要 | 中 | 增长率来自二手报道,且仍反映 Gartner 的广义定义。 | |
| McKinsey 价值池 | 2023 | Global | 2600-4400 | 跨 63 个用例的年度经济价值估算 | 中 | 价值池不等于已货币化的软件支出。 | |
| Menlo 企业支出 | 2024 | 美国企业样本 | 13.8 | 从受访美国企业领导者观察到的企业支出 | 中 | 基于调查,且仅限美国企业。 | |
| Menlo 应用层支出 | 2024 | 美国企业样本 | 4.6 | 企业 GenAI 预算内观察到的应用层支出 | 中 | 更贴近 OpenAI 的收入界面,但仍是样本口径且偏美国。 | |
| McKinsey 银行业价值 | 2023 | Global banking | 200-340 | 用例全部落地时的年度经济价值区间 | 中 | 行业价值上限,不是当前支出。 | |
| McKinsey 零售/消费品价值 | 2023 | 全球零售 / CPG | 400-660 | 用例全部落地时的年度经济价值区间 | 中 | 行业价值上限,不是当前支出。 | |
| IDC 公开摘要尝试 | 2025 | Worldwide | 访问日查看公开 URL 时返回文章不可用响应 | 低 | 没有可用公开访问或授权报告,无法用于交叉验证。 |
广义预测、观察到的企业预算和价值池估算都有用,但它们测的是市场的不同层次。因此,本章把它们当作不同镜头,而不是可互换的 TAM 数字。
[CM016, CM020, CM036, CM037, CM038, CM049]宽口径经济价值池远大于当前真正流向模型提供商和应用厂商的企业预算。
这张图是一组口径叠层,不是严格的 TAM-SAM-SOM 级联,因为底层来源衡量的是不同层级的价值和支出。
[CM016, CM020, CM049, CM051]公开价值池估算支撑 OpenAI 相关工作流的巨大天花板,但垂直行业区间差异显著。
中点是已发布低/高区间的算术中心,仅用于让比较更清楚。
[CM051, CM053, CM054]2.2 买方分层、预算与采用路径
公开证据显示,OpenAI 同时卖进多套买方地图。第一类是广义知识工作市场:OpenAI 明确把 Business 和 Enterprise 套餐作为员工工具销售,而 Microsoft 2025 Work Trend Index 显示,领导者优先关注客服、营销和产品开发,并预期数字劳动力会扩充员工队伍。第二类是开发者市场,OpenAI 的 API 和面向开发的定价,与 Menlo 发现的企业采用率最高用例一致:代码 copilots 的采用率达到 51%。第三类是机构教育,OpenAI 的 ChatGPT Edu 材料把学生、教师、研究人员和校园运营纳入集中化大学采购动作。 预算归属同样关键。Menlo 数据显示,企业需求不再局限于沙盒式创新团队:60% 支出仍来自创新预算,但 40% 已来自常设预算,而且这部分经常性支出中超过一半来自既有预算重分配。按部门看,支出集中在 IT、产品与工程、数据科学,但客服、销售、营销、HR、财务和法务也握有有意义的预算。这一模式适配 OpenAI 的 GTM 界面:公司既能通过中央 IT 横向销售,也能靠职能 ROI 案例纵向切入。采用路径通常仍是从实验到特定职能试点,再到安全和数据集成,最后才进入经常性预算和广泛部署。[CM002, CM003, CM009, CM010, CM012, CM014]
| 细分 | 买方 | 用户 | 付款方/工作流 | 预算负责人 | 采用触发因素 |
|---|---|---|---|---|---|
| 企业员工 | CIO、COO 或转型负责人 | 普通知识工作者和管理者 | 公司出资的席位部署,用于生产力、搜索和跨职能辅助 | 中央 IT 加共享转型预算 | 需要在守住安全和治理的同时补上产能缺口 |
| 开发者和工程 | CTO、工程 VP、平台负责人 | 软件工程师、DevOps 和技术项目团队 | 工程预算支持的 copilot、agent 工作流和按 token 定价 API | 产品 + 工程和平台预算 | 编程 copilot 和核心技术工作流自动化具有高 ROI |
| 客户支持 | 支持运营或 CX 负责人 | 联络中心坐席和客户服务经理 | 支持自动化、检索和 chatbot 工作流 | 支持与运营预算,并与 IT 合作 | 需要 24/7 响应能力和降低每次联系成本 |
| 销售和营销 | CRO、CMO 或 rev-ops 负责人 | 销售、营销人员和 campaign 团队 | 内容生成、潜客开发和 pipeline 工作流 | 销售和营销预算 | 需要更快产出内容,并提高一线团队生产力 |
| 高等教育 | 教务长、CIO、采购负责人 | 学生、教师、研究人员和行政人员 | 面向教学、研究和校园运营的机构级 AI 部署 | 大学中央行政和 IT | 需要可负担的安全访问,以及学术和行政生产力 |
| 受监管专业职能 | 总法律顾问、CFO、合规负责人,或临床/金融运营负责人 | 律师、财务人员、临床人员和专业运营人员 | 法律、金融和医疗中重文档或重判断的工作流 | 带治理监督的职能预算 | ROI 潜力强,但规模化前要经过更重的隐私、准确性和监管审查 |
OpenAI 的实际买方图谱横跨中央 IT、职能负责人和机构买家;采用通常从一个工作流开始,只有在治理和 ROI 证明之后才扩张。
[CM002, CM003, CM009, CM010, CM012, CM014]OpenAI 近期最好的细分市场同时具备高 ROI 可见度和可管理的治理负担;受监管的专业职能扩张更慢。
[CM009, CM012, CM014, CM021, CM022, CM030]2.3 增长驱动、采用约束与估值相关性
需求侧驱动力很强。Microsoft 报告称,82% 的领导者认为 2025 年是关键战略年份,82% 预期未来 12 至 18 个月会增加数字劳动力。PwC 2025 Jobs Barometer 用可衡量的劳动力市场证据强化了这一信号:具备 AI 技能的员工享有 56% 薪资溢价,受 AI 影响的岗位继续增长,AI 暴露度最高的行业在人均收入增长上显著快于暴露度较低的行业。McKinsey 还给出一张功能地图,显示约 75% 的经济上行集中在客户运营、营销与销售、软件工程和 R&D。OpenAI 的助手、API 和 agent 模式,正好能在这些工作流里争取预算。 制动力同样实质。Deloitte 显示,大多数组织仍预期近期只有一小部分实验能完全规模化,尽管最先进项目的 ROI 往往为正。监管合规成为 Deloitte 报告中排名第一的障碍,治理项目推进缓慢,Menlo 的失败试点数据则显示,实施成本、数据隐私、ROI 落差和幻觉仍是持续失败模式。这就是为什么 OpenAI 的隐私承诺具有商业意义:信任、保留控制和企业身份功能不是装饰性卖点,而是预算从试点走向生产的前提。欧盟 AI Act、OECD 问责框架和 NIST 风险管理指南都强化同一结论:OpenAI 所处市场需求真实且在扩大,但在受监管或高后果工作流中,估值应对较慢规模化打折,直到治理和价值证明能逐个客户跑通。[CM006, CM007, CM011, CM012, CM013, CM014]
| 驱动/约束 | 方向 | 时间 | 含义 | 尽调问题 |
|---|---|---|---|---|
| 数字劳动力需求 | 上行 | 近期 | 支撑助手和 agent 预算扩张 | 要求 OpenAI 拆分 pipeline 中席位扩张与 agentic 工作流部署的比例。 |
| 生产力和工资信号 | 上行 | 近期 | 强化企业为高价值 AI 工具付费的意愿 | 要求客户 ROI 案例,并提供劳动力替代或收入提升证据。 |
| 预算常态化 | 上行 | 12-24 个月 | 暗示 AI 支出正在进入经常性运营预算 | 要求证明 OpenAI 合同正在替代既有软件或服务支出。 |
| Code-copilot 采用 | 上行 | 当前 | 有利于开发者 API、编程 agent 和技术席位增长 | 要求 OpenAI 在工程和编程工作负载中相对竞争对手的份额。 |
| 监管合规 | 下行 | 当前且加重 | 放慢受监管行业和高后果工作流的部署 | 要求产品合规路线图、逐地区准入安排和客户审计要求。 |
| 治理建设 | 下行 | 12+ 个月 | 即使试点显示 ROI,也会推迟全面部署 | 要求实施服务负担、治理模板和安全审查周期。 |
| 隐私和信任 | 下行 | 当前 | 信任功能是从试点扩张到生产环境的前置条件 | 要求客户侧隐私控制、保留设置和数据边界例外的证明。 |
| 实施成本和幻觉 | 下行 | 当前 | 抬高服务负担,并收窄值得规模化的用例范围 | 要求部署成本曲线、人工审核率和生产错误指标。 |
| 多模型切换 | 混合 | 当前 | 维持较高买方权力,并削弱单一供应商锁定 | 要求按工作负载和细分拆分的留存、胜率和互操作性数据。 |
估值真正要回答的,不是需求有没有,而是 OpenAI 能不能在治理要求和竞争侵蚀买方动能之前,把试用转成可复制、可信、能留存的生产支出。
[CM006, CM007, CM012, CM016, CM017, CM018]公开证据指向一条可复制路径:探索预算先催生试点;只有数据集成、治理和 ROI 证明到位后,生产部署才会跟进。
[CM006, CM007, CM013, CM017, CM018, CM026]2.4 展示项
03竞争格局
3.1 版图:直接实验室、套件型巨头、基础设施聚合商和自建替代方案
OpenAI 竞争的不是一两家前沿模型实验室。公开版图已经覆盖直接前沿同行 Anthropic,Google 和 Microsoft 这类套件型巨头,Amazon Bedrock 这类多模型基础设施聚合商,以及围绕 Llama 等开放权重模型或 Mistral 等隐私导向供应商搭建的替代方案。Menlo 2025 企业报告尤其重要,因为它显示买方替代集合在实践中扩宽,而不是只在理论上扩宽:企业 AI 用例中 76% 现在选择外购而非内部自建,横向 copilots 已占 84 亿美元支出,基础模型份额也已在 Anthropic、OpenAI 和 Google 之间发生实质重分配。 这改变了 OpenAI 的“竞争者”定义。直接同行集合仍包括 API 和编码市场里的 Anthropic,但实际交易集合也包括 Google Workspace 和 Microsoft 365 预算、Amazon 的多模型路由层,以及把开放权重模型与云工具结合起来的内部自建。OpenAI 的核心优势仍是旗舰消费产品、广泛使用的 API 和企业信任控制的组合。弱点在于,现在有多条竞争路径可以解决同一类任务:买方可以购买套件内助手,可以在多家供应商之间路由流量,也可以在控制权比前沿质量更重要时基于开放模型自建。[CP027, CP028, CP029, CP030, CP033, CP034]
独立的 2025 年市场数据显示,OpenAI 仍是一线竞争者,但不是毫无争议的企业市场领导者。
[CP027, CP030, CP037]3.2 竞争者类别画像:OpenAI 哪里最强,哪里对手更锐利
Anthropic 是最清晰的直接产品威胁。Anthropic 官方材料显示,Claude 横跨商业和企业界面销售,并通过 Amazon Bedrock 和 Google Cloud Vertex AI 分发;独立报道则显示 Claude 赢得企业编码叙事,并从 OpenAI 手中拿走企业用量份额。Google 的竞争方式不同:Gemini 不只是模型 API,也是更大的 Workspace 和 Google Cloud 采购界面的一部分,已有合规、安全和终端管理功能。Microsoft 的威胁更偏分发。它的 Copilot 界面嵌在 Microsoft 365 和 Teams 里,Azure OpenAI 则提供一条并行的开发者和企业路线。Amazon 的姿态不是押一款旗舰助手,而是卖可选性:Bedrock 让买方在 Anthropic、Meta、Mistral 和 Amazon 模型之间选择,Amazon Q 则提供明确按席位收费的内部知识工作助手。 Meta Llama 和 Mistral 即便不主导闭源模型份额,也仍然重要。Llama 给企业和构建者一条开放权重后路,并带有长上下文多模态能力;Mistral 则卖隐私和数据主权叙事,在欧洲的说服力强于许多美国对手。合在一起,这些替代方案限制了专有实验室单靠模型独占性建护城河的上限。[CP007, CO083, CP009, CP010, CP011, CP012]
| 竞争对手 | 类别 | 规模/融资 | 目标客群 | 差异化 | 短板 |
|---|---|---|---|---|---|
| OpenAI | 直接前沿实验室 + 平台 | 公开认知中的领先者,覆盖消费者和 API;Menlo 2025 显示其占企业 LLM 支出的 27% | 开发者、企业、教育、专业消费者 | ChatGPT + API 广度 + 隐私控制 | 企业份额正受 Anthropic 和 Google 挤压;实际成交价并未完全公开 |
| Anthropic | 直接前沿实验室 | CNBC 报道 2025 年 3 月估值 615 亿美元;Menlo 2025 显示其占企业 LLM 支出的 40% | 企业买家和代码密集型团队 | 代码能力、安全品牌、经由 Bedrock 和 Vertex 的多云分发 | 企业合同价格在公开层面仍大多不透明 |
| Google Gemini | 既有套件厂商 + 超大云 | Menlo 2025 显示其占企业 LLM 支出的 21%,并嵌入 Workspace/Cloud 分发 | Workspace 账户、开发者、IT 主导的部署 | 捆绑能力、合规姿态、Google 基础设施 | 企业 Gemini 合同的确切价格仍是定制/联系销售 |
| Microsoft Copilot / Azure OpenAI | 既有套件厂商 + OpenAI 渠道 | Microsoft 365 预算内领先的横向 copilot 阵营 | 标准化在 Microsoft 体系上的企业和 Azure 开发者 | 工作应用里的 Copilot、agent builder、Azure OpenAI 路径 | 已审阅的商业页面没有公开完整价格表 |
| Amazon Bedrock / Amazon Q | 基础设施聚合器 + 助手 | Bedrock 支持多模型,Amazon Q 明确公布按用户席位定价 | 以 AWS 为中心的构建者和内部知识工作团队 | 模型选择、更低批处理价格、治理护栏 | 在广泛终端用户心智中,助手品牌弱于 ChatGPT 或 Copilot |
| Meta Llama | 开放权重替代品 | 按采用度看是企业开源领导者,但开源份额仍只有 11% | 追求控制权、自托管或降低依赖的构建团队 | 开放权重部署、多模态长上下文、成本效率 | 企业采用闭源模型的速度仍快于开放权重 |
| Mistral | 面向隐私和主权的替代方案 | CNBC 报道 2025 年 9 月估值 140 亿美元 | 以欧洲为重点、重视隐私的企业和构建者 | 企业隐私、数据所有权、部署灵活性 | 与更大对手相比,公开采用份额和精确价目表信息更薄 |
| 基于开放模型 + 云工具的内部自建 | 现状方案 / 替代品 | 76% 的用例仍是外购,但在控制要求很高的工作流里,内部自建依然可行 | 内部 AI 工程能力强且数据受限的团队 | 最大控制权和供应商可选性 | 相比购买成品,见效更慢、集成负担更重 |
各行覆盖了严肃 OpenAI 买家在 2026 年可以选择的主要直接、既有、相邻、替代和内部自建方案。
[CP001, CP003, CP005, CP007, CO083, CP009]| 购买标准 | OpenAI | Anthropic | Microsoft | Amazon | 开放权重 / 主权栈 | |
|---|---|---|---|---|---|---|
| 前沿通用模型深度 | 强 | 强 | 强 | 中 | 中 | 中 |
| 代码专精 | 强 | 强+ | 中 | 中 | 未知 | 中 |
| 企业管理 / 隐私姿态 | 强 | 中 | 强 | 强 | 中 | 不定 |
| 多模型可选性 | 低 | 中 | 中 | 低 | 强 | 强 |
| 原生生产力套件分发 | 中 | 低 | 强 | 强 | 中 | 低 |
| 开放部署 / 自托管灵活性 | 低 | 低 | 中 | 中 | 强 | 强 |
序位标签概括的是已审阅的公开证据,而非隐藏的产品遥测;当来源集合不足以支撑更硬判断时,单元格标为未知或不定。
[CP002, CP003, CP007, CP009, CP010, CP011]OpenAI 在分发和模型能力上都接近前沿,但套件型巨头在分发端更靠右;Amazon 和开放权重栈则把买方灵活度拉得更高。
分数是基于公开产品界面、市场份额变化和捆绑 / 分发证据得出的序数判断,而不是厂商直接披露的指标。
[CP007, CP010, CP014, CP018, CP021, CP024]OpenAI 和 Anthropic 在核心模型深度上领先,既有巨头在套件分发上领先,Amazon 与开放权重栈则在买方选择空间上领先。
单元格是对已审阅证据的序数概括;公开来源覆盖不完整处,刻意保留未知或可变。
[CP002, CP003, CP007, CP009, CP010, CP014]3.3 切换成本、锁定、多家并用与护城河耐久性
OpenAI 仍有真实竞争耐久性,但这种耐久性有条件,并非自动成立。公司受益于 ChatGPT 熟悉度、API 广度,以及默认不使用业务数据训练等企业隐私承诺;这些因素帮助它同时拿下开发者预算和知识工作者预算。不过,同一批公开证据也说明,商品化风险已经是核心尽调事项。Menlo 2025 数据中,Anthropic 已在企业 LLM 支出和编码份额上超过 OpenAI;Google 和 Microsoft 能把 AI 采购藏进更大的套件续约;Amazon Bedrock 则把多模型路由变成一流运营模式,而不再是定制工程项目。 因此,这个市场的结构性锁定低于经典 SaaS。买方可以多家并用模型,把席位预算留给既有套件,也可以把敏感工作负载转向开放权重或主权导向替代方案。Menlo 发现只有 16% 的企业部署算真正 agent,这一点也重要:许多现实系统仍足够简单,价格、治理或可靠性变化时可以迁移到别的供应商。对 OpenAI 来说,护城河不再主要取决于它是不是唯一可信的模型实验室,而在于能否同时在质量、开发者体验、信任和分发上持续明显领先。[CP002, CP003, CP004, CP020, CP030, CP031]
| 供应商 | 公开套餐 | 价格/单位/合同模式 | 包含能力 | 折扣或未知项 | 启示 |
|---|---|---|---|---|---|
| OpenAI | GPT API + ChatGPT Business / Codex 产品线 | GPT-5.5 API 每 100 万 tokens 输入 $5 / 输出 $30;Business 打包在 2026 年 4 月调整;Codex 席位按用量计价 | 前沿 API、缓存输入折扣、商业管理/隐私控制 | 已审阅定价页没有完整公开当前自助席位价格表;企业条款为定制 | OpenAI 在 API 经济性上透明,但大型企业实际成交价透明度较低 |
| Anthropic | Claude 套餐 + 企业合同 | claude.ai 列出套餐层级,但已审阅公开材料显示,企业定价仍是定制且按模型而异 | Claude 产品入口,加上经 Bedrock 和 Vertex 的分发 | 企业席位标价和折扣未在保留来源中公开 | 公开价格透明度较低,但采用度仍强 |
| Gemini API + Workspace Enterprise 产品线 | 按 token 计价的 Gemini API,按模型分层;Workspace Enterprise 为联系销售/定制 | Gemini API,加上企业套件控制和端点/合规功能 | 边际 AI 成本可埋在更大的 Workspace 合同里 | Google 既能打直接模型经济性,也能打套件捆绑 | |
| Microsoft | Microsoft 365 Copilot + Azure OpenAI | 取决于附加购买和许可证路径;已审阅公开商业页面没有完整价格表 | Microsoft 应用里的 Copilot、工作 agent、Azure OpenAI 路径 | 区域许可证和捆绑折扣在保留来源中不可见 | 即使 AI 单价难以拆出,Microsoft 也能赢单 |
| Amazon | Amazon Q Business + Bedrock | Q Business Lite 为 $3/用户/月;Pro 为 $20/用户/月;Bedrock 批量推理在支持模型上比按需低 50% | 按席位计费的助手,加多模型服务层 | 底层 Bedrock 模型经济性随供应商和用量模式变化 | Amazon 把明确的助手定价和基础设施可选性绑在一起 |
| Meta / 开放权重 | Llama 4 及托管/开放部署 | 开放权重控制权、长上下文、自托管或第三方托管路径 | 总成本取决于基础设施、优化和运维,而不是单一供应商标价 | 开放权重降低买方对单一专有 API 的依赖 | |
| Mistral | Studio + API / 企业部署 | 存在按模型使用量计价和企业合同,但已审阅官方文档无法完整抽取精确费率 | 重隐私的平台和企业部署 | 公开价目表细节比 OpenAI 或 AWS Q 更薄 | Mistral 更多靠主权和部署灵活性竞争,而不是最清晰的公开标价 |
不受支持或不透明的单元格有意保留为 null 或描述为定制,而不是猜测。
[CP001, CP005, CP006, CP018, CP019, CP038]| 护城河主张 | 威胁 | 严重性 | 缓释/尽调问题 |
|---|---|---|---|
| OpenAI 的消费者 + API 切入点带来持久心智和分发 | Anthropic 和 Google 正在拿走企业份额,知名度不等于企业留存 | 高 | 要求提供企业净留存、赢单/输单原因,以及消费者起步与 CIO 主导落地之间的拆分 |
| 企业信任控制帮助 OpenAI 通过受监管买家 | Google、Microsoft、AWS 和 Anthropic 都在宣传竞争性信任或合规控制 | 中 | 按受监管垂直行业比较审计材料、留存控制和采购成交率 |
| 前沿模型质量是一道可防守的护城河 | Bedrock 式多模型路由和快速模型替换会把基础模型层商品化 | 高 | 测试工作负载在性能、成本、可靠性和治理之间如何分流 |
| OpenAI 能比小型实验室更强分发 | Microsoft 和 Google 可以把 AI 卖进买家本来就预算好的续约动作里 | 高 | 检查捆绑附加率,以及 OpenAI 被既有套件整合替代的频率 |
| 随着 agent 铺开,转换成本会上升 | 目前只有 16% 的企业部署算真正 agent,许多系统仍可迁移 | 中 | 审阅合同期限、数据导出工具、prompt 可迁移性和模型路由抽象 |
| 开放权重替代品仍居次要地位 | Llama 和内部自建路径持续改进;当控制权重于最高前沿质量时,它们会变得重要 | 中 | 按自托管或采用开放权重模型的意愿切分买方群体,尤其关注欧洲和受监管行业 |
登记表聚焦最可能改变承销假设的耐久性问题,而不是每一场战术产品战。
[CP002, CP003, CP004, CP005, CP006, CP021]3.4 展示项
04财务
4.1 收入模式与公开牵引力:变现面广,但结构披露不完整
OpenAI 现在不是靠单一 SaaS 套餐变现,而是通过几块不同界面赚钱。公开公司材料显示,收入面包括 ChatGPT Plus 和 Pro 消费订阅、自助式 ChatGPT Business 席位、按用量计费的 Codex 席位、定制 Enterprise 合同、按 token 定价的 API 用量,以及由合作伙伴中介的分发或基础设施关系。这个广度很重要,因为它降低了对单一买方分层的依赖;但也意味着公开记录混在一起的是席位定价、token 定价、订阅和谈判得来的商业条款,彼此不能直接比较。最干净的定价证据仍是消费者和 API 标价;Business 和 Enterprise 在已审阅公开材料中只部分透明。 公开牵引力足以证明需求,但不足以对齐收入质量。OpenAI 自己 2025 年 12 月的企业报告称,ChatGPT 服务超过 8 亿周活用户;TechCrunch 报道称,企业消息量自 2024 年 11 月以来增长 8 倍,API reasoning-token 用量同比增长 320 倍,自定义 GPT 用量增长 19 倍并占企业消息的 20%。Forbes 公司档案还给出 2025 年营收 37 亿美元这一公开参照点。关键承销细节是,TechCrunch 也报道 OpenAI 大部分收入仍来自消费者订阅;这意味着企业采用正在快速上升,但尚未完全取代消费者收入成为经济重心。[CI001, CI002, CI003, CI004, CI005, CI006]
| 收入流 | 机制 | 单位 | 当前价值/状态 | 质量 | 尽调问题 |
|---|---|---|---|---|---|
| 消费者订阅 | ChatGPT Plus 和 Pro 订阅 | 每月订阅用户 | Plus $20;Pro $200 | 标价可见性高;实际 ARPU 可见性低 | 提供按月订阅用户数、流失率,以及 Plus 与 Pro 的占比。 |
| 商业席位 | 标准 ChatGPT Business 席位 | 每席位每月 | 公开帮助中心审阅确认自 2026-04-02 起每月降价 $5,但审阅文本未公开一套规范化全球标价 | 中 | 提供当前区域价目表、最低席位数和席位使用率 cohort。 |
| Codex 用量 | 按用量计费的 Codex 席位和额度 | 按用量的席位 / token 支出 | 无固定席位费;按用量即付 | 中 | 提供每个 workspace 的 Codex 实际支出和毛利率画像。 |
| 企业合同 | 联系销售的企业协议 | 定制合同 | Enterprise 通过联系销售出售,公开标价仍不透明 | 低 | 提供主服务协议样本、最低承诺和折扣政策。 |
| API 平台 | 按 token 计价的模型用量 | 100 万 tokens | GPT-5.5 输入 $5 / 输出 $30;GPT-5.4 输入 $2.5 / 输出 $15 | 标价可见性高;实际收益率可见性低 | 提供客户 token 组合、缓存/批处理占比,以及按模型家族拆分的实际净收入。 |
| 合作伙伴/渠道经济性 | Azure 和合作伙伴中介的商业路径 | 协商合同 / 收入分成 | 分发存在,但当前抽成率和结算机制不公开 | 低 | 提供当前 Microsoft 收入分成和算力定价表,以及其他渠道经济性。 |
本表区分可见标价和不透明的实际经济性。历史融资轮次仍放在 Company Overview;本表只关注当前变现表面。
[CI001, CI002, CI003, CI004, CI005, CI006]| SKU 或合同 | 价格/单位/合同 | 标价 vs 实际成交价 | 折扣/未知项 |
|---|---|---|---|
| ChatGPT Plus | $20/用户/月 | 公开标价可见 | 无公开流失、追加销售或 cohort 留存数据 |
| ChatGPT Pro | $200/用户/月 | 公开标价可见 | OpenAI 称高级模型需要显著更多算力,因此利润率敏感度可能很高 |
| ChatGPT Business 标准席位 | 部分可见;2026 年 4 月价格每月下调 $5 | 已审阅文本只有部分公开可见性 | 当前区域标价、最低量和实际折扣仍不清楚 |
| Codex 席位 / workspace 用量 | 按用量即付;无固定席位费 | 公开计费逻辑可见 | 实际客户支出取决于用量画像和消耗的额度 |
| ChatGPT Enterprise | 定制/联系销售 | 标价不透明 | 无公开价目表、最低承诺或实施费细节 |
| GPT-5.5 API | 每 100 万 tokens 输入 $5 / 输出 $30 | 公开标价可见 | 实际价格取决于组合、缓存率和折扣 |
| GPT-5.4 API | 每 100 万 tokens 输入 $2.50 / 输出 $15 | 公开标价可见 | 同上,实际定价未知项仍在 |
OpenAI 现在公开的定价多于许多私有 AI 厂商,但公开记录仍过度集中在标价,而不是实际合同经济性。
[CI002, CI003, CI005, CI006, CI008, CI011]| 指标 | 值/null | 置信度 | 为什么重要 |
|---|---|---|---|
| 公开 2025 年收入 | 3.7 | 中 | 给出一个外部收入规模锚点,但不能说明分部质量或利润率 |
| 公开收入结构信号 | 大部分收入仍来自消费者订阅 | 中 | 显示企业增长尚未清楚成为主导经济基础 |
| 企业消息量增长 | 自 2024 年 11 月以来增长 8x | 中 | 是工作场景采用的有用落地扩张需求代理 |
| API 推理 token 增长 | 同比增长 320x | 中 | 表明技术使用更深,但也暗示算力强度很高 |
| Custom GPT 深度 | 增长 19x;占企业消息的 20% | 中 | 表明工作流集成,而不是浅层试用 |
| 员工 ROI 代理指标 | 每天节省 40-60 分钟 | 中 | 是公开层面最好的效率信号,但不等同于回本数据 |
| 按产品面的毛利率 | 低 | 模型/API/订阅耐久性的核心承销指标仍未披露 | |
| CAC / 回本期 / 销售周期 | 低 | 评估企业 GTM 效率和资本需求所必需 | |
| 净收入留存 / 扩张 | 低 | 承销企业账户经常性收入质量的关键指标 |
公开证据对采用代理指标较好,对真正单位经济性较差。这里的 null 是有意标出的尽调阻断点,不是格式缺口。
[CI020, CI021, CI022, CI023, CI024, CI043]OpenAI 靠多个变现界面把使用量转成收入,但能留下多少毛利,取决于仍只部分公开的合作伙伴经济性和算力强度。
这张桥接图是定性判断,因为公开材料只暴露标价和采用代理指标,没有按收入流披露实际利润率。
[CI001, CI002, CI003, CI008, CI011, CI012]公开证据能支撑需求和漏斗顶部定价可见度,但到 CAC、回本期和毛利率可以量化之前,证据链就断了。
这张桥接图只使用公开采用和定价信号;公开记录止步之处,下游单位经济输出刻意保持未解。
[CI005, CI011, CI012, CI013, CI014, CI015]4.2 GTM 路径与收入质量:自助速度、企业上行销售和定价不透明
已审阅证据指向一种混合 GTM 模式。OpenAI 的 Business 页面同时推动立即试用和联系销售两种动作;旧版 ChatGPT Team 发布文,现已更新为 Business,也确认自助式员工产品与更早的 Enterprise 发布并行。帮助中心材料显示,Business 将标准订阅席位和按用量计费的 Codex 席位结合在一起;OpenAI 还在 2026 年 4 月 2 日调整 Business 定价,将标准席位月费下调 5 美元,同时引入仅 Codex 的灵活席位。这个结构在财务上很重要:它说明 OpenAI 愿意在同一个 workspace 里混合可预测的席位收入和弹性的开发者式用量收入。 明确的隐私控制有利于收入质量,但最大合同所在位置的定价不透明又削弱了可判断性。OpenAI 称默认不使用业务数据训练,企业客户也可以控制保留期限,这支撑企业在敏感工作流中扩大使用。与此同时,公开记录没有披露标准化 Enterprise 价目表、典型折扣、最低承诺、销售周期长度、CAC、回本期或 NRR。因此,最好的公开销售效率代理指标只能是间接的:OpenAI 企业研究中的员工报告称每天节省约 40 至 60 分钟,Enterprise logos 加上广泛业务路由也显示出真实部署动能。这些是有用的需求信号,但不能替代 cohort 经济性。[CI004, CI005, CI006, CI007, CI008, CI009]
4.3 成本结构与资本强度:算力主导承销故事
OpenAI 的公开财务图景中,主导因素不是工资或销售费用,而是算力和伙伴经济性。OpenAI 自己在 ChatGPT Pro 发布中明确说,更高级模型需要显著更多算力;这是官方材料中最干净的解释,说明为什么单看标价几乎看不出利润率。公司当前的基础设施姿态,通过一组异常庞大的生态承诺变得可见:Stargate 项目计划四年投资 5,000 亿美元,并立即部署 1,000 亿美元;CoreWeave 另行宣布一份最高 119 亿美元的 OpenAI 基础设施合同,以及 OpenAI 对其 3.5 亿美元股权投资;Oracle 也多次把主要 backlog 增长和融资计划与 OpenAI 相关 AI 需求联系起来。 Oracle 披露尤其重要,因为它显示相当一部分 capex 负担由供应商和融资伙伴中介,而不是透明地落在 OpenAI 自己已发布的资产负债表上。Oracle 2025 年 3 月称已与 OpenAI 签署云协议,并预计拿到第一份 Stargate 合同,同时当年把数据中心容量翻倍。到 2026 年 2 月,Oracle 称计划融资 450 亿至 500 亿美元,以为包括 OpenAI 在内的客户扩建 OCI 容量;到 2026 年 3 月,它又称大规模 AI 合同 backlog 的大部分跃升,可以由客户预付款或客户提供 GPU 支撑。这并不让 OpenAI 变成轻资本公司;它意味着部分融资负担转移进合作伙伴资产负债表、供应商合同和预付款结构,而公开投资人仍无法把这些项目完整净额化。[CI012, CI025, CI026, CI027, CI028, CI029]
| 项目 | 当前价值/状态 | 启示 | 尽调问题 |
|---|---|---|---|
| Stargate 四年基础设施计划 | 四年目标 $500B;即时部署 $100B | OpenAI 规模的基础设施需求更接近工业资本开支,而不是普通软件托管 | 要求提供 OpenAI 在 JV 融资中的份额、里程碑计划,以及任何最低采购义务。 |
| CoreWeave 专用算力合同 | 最高 $11.9B 合同,外加 OpenAI $350M 股权投资 | 正通过明确的长期财务承诺锁定供应商产能 | 要求提供付款计划、终止权和最低用量条款。 |
| Oracle 供应商融资 | Oracle 计划在 2026 年融资 $45B-$50B,以扩建 OCI,客户包括 OpenAI | 合作伙伴资产负债表正被用来满足与 OpenAI 有关的需求 | 要求提供转嫁定价、预付款机制,以及 OpenAI 是否担保任何产能。 |
| Oracle AI 合同融资结构 | Oracle 称大量设备由客户预付款或客户提供的 GPU 资助 | 部分资本开支负担被上游转移,但最终付款方结构不清楚 | 要求提供预付款台账、客户出资 GPU 计划,以及任何相关资产负债表负债。 |
| Microsoft 经济条款 | CNBC 报道,2026 年重置对 Microsoft 的收入分成付款设上限 | 相比旧经济条款,可能提高 OpenAI 留存毛利,但幅度未公开披露 | 要求提供当前收入分成、算力价目表和独占时间线。 |
| OpenAI 现金 / 烧钱 / 跑道 | 公开证据不足以直接评估流动性 | 要求提供月度现金桥、季度烧钱、契约包,以及基准/高算力场景下的跑道。 |
历史融资时间线放在 Company Overview;本表关注前瞻融资负担、供应商融资和未解的流动性问题。
[CI025, CI026, CI027, CI028, CI035, CI036]围绕 OpenAI 的公开资本和基础设施承诺,已经从单一供应商合同延伸到以数千亿美元计的多年项目。
所有数值都是公开承诺或融资金额,单位为十亿美元;它们并不等同于 OpenAI 单期现金流出。
[CI025, CI026, CI027, CI035]公开记录显示,OpenAI 的资本开支负担有一部分转移给了供应商和合作伙伴,但时点和剩余义务仍难以承销。
单元格标签是公开证据质量的序数概括,不是隐藏的内部遥测。
[CI025, CI026, CI027, CI028, CI035, CI036]4.4 财务结论:变现改善,但关键承销指标仍未公开
正面案例很直接。OpenAI 有多条变现向量、明确 API 定价、消费者愿意为每月 20 美元和 200 美元套餐付费、自助式 Business 产品、定制 Enterprise 上行销售路径,以及企业用量深度正在上升的清晰证据。供应商和合作伙伴披露也显示,生态愿意围绕 OpenAI 以异常规模融资建设容量;相比必须自筹每一次重大建设的公司,这降低了近期执行风险。 负面案例同样清楚。公开证据仍未披露现金余额、净 burn、runway、已实现企业定价、产品级毛利率、headline 报道之外的伙伴分成机制,或标准化销售效率数据。CNBC 报道称 Microsoft 经济性被重置,以封顶收入分成付款,这可能有助于未来利润率留存;但已审阅公开材料没有给出量化该收益所需的完整排程。因此,承销结论是混合的:OpenAI 的收入质量在改善,需求基础真实,但利润率路径和资本充足性仍取决于关于算力合同、客户预付款、折扣和内部现金生成的私有信息。本章能支撑强需求叙事;没有管理层数据,它不能完成一份传统财务尽调。[CI009, CI010, CI011, CI012, CI020, CI024]
| 缺失的私有指标 | 影响 | 精确尽调路径 |
|---|---|---|
| 按消费者、商业、企业、API 和渠道拆分的收入结构 | 无法承销各收入流的集中度、季节性或耐久性 | 要求按产品面提供月度收入桥和递延收入滚动表。 |
| 企业合同价格、折扣和最低承诺 | 仅有标价不足以评估 ASP、收入质量或利润留存 | 审阅代表性企业 MSA、订单表和折扣政策备忘录。 |
| 按产品面拆分的毛利率和算力 COGS | 只靠公开定价无法建模利润率路径 | 要求提供 COGS 瀑布,拆分推理、训练、第三方收入分成和支持。 |
| 手头现金、净烧钱、跑道和融资契约 | 资本充足性仍是公开信息承销中最大的硬阻断点 | 获取资金管理仪表盘、现金报表、债务协议和董事会跑道情景。 |
| 按分部拆分的销售效率 | 无公开 CAC、回本期、销售周期或 NRR 数据 | 要求提供漏斗指标、按分部 CAC、cohort NRR 和企业实施成本数据。 |
| 横跨 Microsoft、Oracle、CoreWeave 和 Stargate 的供应商义务 | 无法把直接义务与合作伙伴出资产能和预付款净额抵销 | 要求提供合并承诺支出计划,纳入 take-or-pay、预付款和 GPU 采购责任。 |
这些缺口是把公开研究章节变成可投资财务模型所需的最低私有数据请求。
[CI020, CI035, CI036, CI039, CI042, CI043]4.5 展示项
05产品与技术
5.1 以客户工作流理解产品界面
OpenAI 的产品界面现在是一个组合,而不是单一聊天机器人。公司公开 Business 页面把 ChatGPT Business 和 Enterprise 描述为员工产品,提供无限聊天、高级模型访问、共享 workspace 管理,以及包括 Codex 和 workspace agents 在内的专门 agent。帮助中心材料显示,Business 产品在 2026 年 4 月 2 日发生实质变化:现在同时支持标准 ChatGPT 席位和按用量计费的 Codex 席位,说明 OpenAI 正在变现不同工作流,而不只是打包一个通用席位。定价界面也强化了这一理解。ChatGPT 公开定价页覆盖 Free、Go、Plus、Pro、Business 和 Enterprise;API 平台则单独公布旗舰模型的 token 价格和上下文窗口限制。 对买方来说,运营含义是 OpenAI 有三种主要交付模式。第一,ChatGPT 作为终端用户 workspace,承载广义知识工作任务。第二,deep research 和 Codex 等专用 agent,把研究和软件开发中的更高自主性工作流产品化。第三,API 平台和 Responses API 给开发者一条可编程路径,接入同一模型家族。这个组合扩大了 OpenAI 可触达的预算所有者,但也让尽调更依赖具体 SKU。评估 ChatGPT Business 的团队应关心管理控制和席位经济性;评估 API 的开发团队应关心模型定价、上下文和工具访问;评估 Codex 的软件团队应关心任务自主性和互联网权限。 公开路线图还显示,OpenAI 愿意快速扩张,也愿意快速下线产品界面。Deep research 和 Codex 在 2025 与 2026 年快速从发布走向更广泛权益;Sora 则反向移动,消费者体验在 2026 年 4 月关闭,API 计划于 2026 年 9 月 sunset。这个组合让 OpenAI 的产品栈显得雄心很大、商业覆盖很广,但也意味着不能只凭发布动能假定产品会长期存在。[CE001, CE003, CE004, CE005, CE006, CE007]
| 模块/产品线 | 主要用户 | 状态/成熟度 | 有证据支持的能力 | 差异化 | 尽调缺口 |
|---|---|---|---|---|---|
| ChatGPT Business / Enterprise 套餐 | 知识工作者和管理员 | 已 GA;Business 套餐在 2026 年 4 月重新定价并重新分层 | 无限聊天、高级模型和工具、共享 workspace 控制,以及包括 Codex/workspace agents 在内的专门 agent | 把前沿模型、管理控制和 agent 产品面打包进用户 workspace | 公开来源未披露按套餐拆分的 SLA、支持响应指标或实际企业折扣。 |
| API 平台 / Responses API | 开发者和 ISV | 已 GA;自 2025 年 5 月起成为核心 agent builder 产品面 | 前沿模型、token 定价、旗舰模型 1.05M 上下文、远程 MCP 支持、图像生成、Code Interpreter 和文件搜索 | 把模型访问与第一方编排和工具调用结合起来 | 公开材料未区分 OpenAI 直接交付与合作伙伴渠道量或经济性。 |
| Codex | 软件工程师和工程管理者 | 2025 年 5 月发布;2025 年 6 月扩大访问 | 基于云的 codex-1 agent 可并行处理许多任务;任务执行期间可启用互联网访问 | 把代码工作变成专门的 agent 产品,而不是通用聊天工作流 | 公开层面的独立生产力基准和实际席位经济性仍很薄。 |
| Deep research | 分析师、研究员和高管 | 2025 年 2 月发布;2026 年 2 月重大控制更新 | 多步研究 agent,支持 MCP/应用连接、可信站点限制、实时进度跟踪和中断/细化流程 | 把研究综合变成产品化、带来源意识的 agent 工作流 | 公开来源未按套餐或领域量化准确率、成本或完成率。 |
| Sora | 创作者和媒体试验者 | 正在停止;应用/web 于 2026 年 4 月停用,API 已公告将在 2026 年 9 月停用 | OpenAI 在关停期间仍提供导出和删除指引 | 显示 OpenAI 能快速测试相邻模态,也能快速退出 | 替代视频战略和迁移路径未公开定义。 |
OpenAI 的产品组合很宽,但成熟度不均。公开层面最耐久的产品面是 ChatGPT、API 平台和 agentic 开发者工具;Sora 最清楚地说明,相邻产品可能被迅速降级优先级。
[CE003, CE004, CE005, CE006, CE007, CE008]ChatGPT 工作区和 API 是目前最耐久的公开界面;Codex 和 deep research 正在快速上行,而 Sora 明显被降权。
[CE006, CE015, CE018, CE020, CE021, CE022]5.2 架构、工具与差异化
OpenAI 的架构在公开材料中越来越明确。公司 2025 年 5 月发布文和开发者参考资料,把 Responses API 定位为 agentic applications 的核心 API primitive,支持远程 MCP servers,也内置 image generation、Code Interpreter 和 file search 等工具。实际含义是,OpenAI 卖的不只是模型推理;它还在卖一层编排层,减少构建者需要自行拼装的工具量。在这一层之上,是 ChatGPT、deep research 和 Codex 等终端用户产品;在其下,是 GPT-5.5、GPT-5.4、GPT-4.1、o1 和 GPT-4o 等前沿模型家族,每个模型通过文档、定价页和发布说明的组合暴露出来。 差异化既来自原始模型质量,也来自包装。公开叙事中,GPT-4.1 相比 GPT-4o 改进了编码、指令遵循和长上下文。Codex 把编码变成一个专用的云端软件工程 agent,可以并行处理多项任务。Deep research 把联网综合能力产品化为 agent,并在 2026 年 2 月加入受信网站限制、MCP 或 app 连接、实时进度跟踪,以及可打断后继续追问精修。这些是工作流 primitive,而不只是模型 benchmark。它们重要,是因为它们让 OpenAI 更接近拥有研究、编码和 agent 构建中的经常性 jobs-to-be-done,而不是只在抽象模型质量上竞争。 最好的公开技术证明仍然不完整。OpenAI 维护第一方 Python 和 Node SDK,模型/系统卡记录覆盖 GPT-4o 和 o1。NIST 对 o1 的预部署评估提供了少见的政府作者检查点。即便如此,公开来源没有完整披露模型路由逻辑、废弃政策,或直接 OpenAI 交付与 Azure 等伙伴渠道之间的确切运营分工。因此,技术故事在界面广度和工具集成上很强,但在企业承销人想要更深部署证据的位置仍不完整。[CE002, CE004, CE015, CE016, CE017, CE018]
| 用户任务 | 当前工作流痛点 | OpenAI 方案 | 可量化的公开收益 | 部署证据 | 局限 |
|---|---|---|---|---|---|
| 全团队知识工作 | 文档起草、分析和问答仍分散在消费级聊天工具和割裂的企业系统里 | 带高级模型、工具和管理控制的 ChatGPT Business / Enterprise 工作区 | 公开资料披露了不限量聊天,以及按套餐层级开放的高级模型和工具;客户案例中心称已有 100 万家企业使用 OpenAI | OpenAI 商业页、价格页和客户案例中心 | 公开资料尚未把企业留存设置、席位下限或支持承诺披露到采购可直接判断的颗粒度。 |
| Agent 构建者 | 开发者原本需要分别拼接模型调用、工具执行和上下文管线 | 支持远程 MCP 并内置工具的 Responses API | 旗舰 API 模型公开 105 万上下文;Batch 定价比标准处理低 50% | OpenAI API 平台、价格页和 2025 年 3 月 Responses 发布 | 公开资料对标价披露更多,对迁移路径和大批量合同下的实际生产经济性披露较少。 |
| 软件开发 | 手工搜代码、打补丁和反复跑测试,会吃掉大量上下文切换成本 | ChatGPT 套餐内的 Codex 云端 agent | OpenAI 称 Codex 可并行处理多项任务;Business FAQ 现已支持按用量计费的专用 Codex 席位 | Codex 发布资料及 ChatGPT Business 帮助文章 | 公开证据对验收测试、合并成功率和工程师净节省时间的披露仍有限。 |
| 研究与综合 | 人类分析师仍要手工浏览、比对来源并产出初稿 | ChatGPT 内的 deep research agent | 到 2026 年 2 月,OpenAI 增加了可信站点限制、实时进度跟踪,以及中断/细化控制 | 官方 deep research 文章及 TechCrunch 发布报道 | 公开证据尚未按领域、来源集合或套餐层级量化完成质量。 |
| 零售门店信息检索 | 门店员工查找流程和运营信息会浪费时间 | 合作伙伴渠道部署的 Azure OpenAI | Microsoft 称 Worten 每年节省 11,000 小时 | Microsoft 关于 Worten 的客户案例 | 这是合作伙伴渠道证据,并非 OpenAI 原生部署经济性或合同条款的直接披露。 |
工作流证据最强的部分在知识工作、开发者工具和合作伙伴渠道的生产部署。公开 ROI 证据方向上积极,但仍比产品宽度暗示的要薄。
[CE004, CE006, CE008, CE015, CE016, CE017]| 层级/组件 | 作用 | 有证据支撑的实现 | 依赖 | 公开风险 |
|---|---|---|---|---|
| 体验层 | 终端用户工作区和权益界面 | ChatGPT 套餐覆盖从消费级到 Business、Enterprise 层级,高级工具和 agent 访问通过套餐设计开放 | 身份、管理控制和套餐权益 | 公开套餐页对功能披露充分,但合同级 SLA 细节不足。 |
| Agent 层 | 把重复任务产品化为更高自主度的 agent | Codex 和 deep research 将代码与研究工作流打包成专用 agent | 模型可用性、工具权限和按套餐开放的访问 | OpenAI 扩 agent 能力很快,买方假设可能按季度变化。 |
| API 编排层 | 面向 agentic 应用的编程接口 | Responses API 被定位为核心 API 原语,配套 MCP,以及用于图像生成、Code Interpreter 和文件搜索的内置工具 | 工具可用性、模型目录和 API 兼容性 | Assistants 迁移到 Responses 的细节演进速度,通常快过大型企业愿意承受的节奏。 |
| 模型层 | 推理、代码、多模态和长上下文能力底座 | 公开模型家族包括 GPT-5.5、GPT-5.4、GPT-4.1、GPT-4o,以及 o1 相关 system-card 证据 | 算力访问、模型路由和安全门控 | 单模型路由、降级和退役政策并未完全公开。 |
| SDK / 开发者工具层 | 降低应用团队的集成摩擦 | OpenAI 维护官方 Python 和 Node SDK,并提供专门 API 参考 | SDK 发布节奏和向后兼容性 | 生产团队仍需密切跟踪库变化和 API 生命周期迁移。 |
| 信任 / 运营层 | 数据处理、安全、安全性和正常运行时间披露 | 业务数据、安全/隐私、安全性和状态页共同形成可见控制面 | 内部控制成熟度和信任门户材料 | 公开信任披露方向上有用,但许多买方级材料仍受限或缺失。 |
OpenAI 的架构最好理解为一套分层运营栈:顶部是工作区,中间是 agent,Responses/工具负责编排,模型是能力引擎,信任/运营在底层提供控制。
[CE001, CE002, CE006, CE007, CE010, CE014]| 日期 | 功能或里程碑 | 状态 | 影响 |
|---|---|---|---|
| 2025-02-02 | deep research 发布 | 已发布 | ChatGPT 从对话助手扩展为多步骤研究 agent。 |
| 2025-04-14 | GPT-4.1 系列在 API 发布 | 已发布 | OpenAI 强调代码、指令遵循和长上下文增益,作为 API 差异化。 |
| 2025-05-16 | Codex 发布 | 已发布 | 代码从聊天功能的一部分,变成专用云端 agent 工作流。 |
| 2025-05-21 | Responses API 增加 MCP 和更多内置工具 | 已发布 | Agent 构建者获得了更完整的一方编排层。 |
| 2025-06-03 | Codex 扩展到 Plus,并开放可选互联网访问 | 发布更新 | 更广分发和更自主的代码工作流提升了产品触达,也扩大了风险面。 |
| 2026-02-10 | deep research 增加可信站点、MCP/app 连接、实时进度和中断能力 | 发布更新 | 研究工作流对企业用例更可控。 |
| 2026-04-02 | ChatGPT Business 引入标准 ChatGPT 席位和按用量计费的 Codex 席位 | 商业更新 | OpenAI 将员工协作定价与代码 agent 用量拆开。 |
| 2026-04-26 | Sora 网页和 app 体验停止 | 已停用 | OpenAI 释放信号:并非每个相邻模态都会长期留在战略内。 |
| 2026-09-24 | Sora API 已安排下线 | 已宣布下线 | Sora API 客户需要在披露的截止日前规划迁移。 |
OpenAI 2025-2026 年产品记录最准确的描述是快速扩张、选择性修剪。发布节奏强化了 agent 战略,Sora 则说明相邻实验可能很快被退出。
[CE008, CE009, CE015, CE016, CE017, CE018]OpenAI 的产品栈把用户工作区、原生智能体、编排 API、模型家族、开发者工具和信任控制叠成一个操作系统。
[CE001, CE006, CE010, CE015, CE016, CE017]常见 OpenAI 工作流从选择套餐或 API 开始,再经过模型和工具选择,进入智能体执行任务,并设置监控和控制检查点。
[CE006, CE007, CE010, CE015, CE017, CE018]OpenAI 的产品体验靠模型可用性、API / 工具编排、SDK 集成和信任控制撑住;合作伙伴渠道扩大了部署触达,也增加了不透明度。
[CE010, CE015, CE016, CE018, CE021, CE034]5.3 可靠性、信任控制与路线图风险
信任和运营是产品的一部分,不只是后台职能。OpenAI 的业务数据页面称,ChatGPT Enterprise、ChatGPT Business、ChatGPT Edu、ChatGPT for Healthcare、ChatGPT for Teachers 和 API 平台中的组织数据归客户所有,并且默认不用于模型训练。面向消费者的安全页面称,用户可以选择数据是否用于训练和模型改进,并突出 Advanced Account Security 作为账户保护功能。安全材料还把 OpenAI 的运营循环描述为 teach、test、share。合在一起,这些材料对尽调有方向性帮助,但仍未给出许多企业买方在没有单独 trust portal 访问时会要求的完整认证与控制包。 可靠性已经公开到足以影响判断。2026 年 5 月 4 日,状态页面显示一起影响 ChatGPT workspace-connector 写入动作的活跃事件。同一页面披露了 2026 年 2 月至 5 月期间的汇总 uptime 快照:API 为 99.99%,ChatGPT 为 99.83%,Codex 为 99.98%,FedRAMP 为 99.93%;同时也提示,单个客户可用性可能因套餐和模型而异。这比许多 AI 创业公司披露得更好,但仍不能替代特定套餐 SLA 条款或支持响应承诺。 下行背景也来自外部来源。SecurityWeek 在 2025 年末报道称,研究人员攻破了 ChatGPT memory 和 web-search 功能,凸显新便利功能会多快扩大攻击面。The Verge 记录了 Sora 如何从发布热度快速走向申请关闭,随后 OpenAI 自己的帮助页正式确认了停用路径。结果是一家公司既有快速移动的路线图和越来越可信的信任控制,也可能在几个季度内改变套餐设计、产品可用性和运营假设。[CE010, CE011, CE012, CE013, CE014, CE025]
| 控制或信号 | 公开状态 | 范围 | 缺口 |
|---|---|---|---|
| 组织数据默认不用于训练 | 有效承诺 | ChatGPT Enterprise、ChatGPT Business、ChatGPT Edu、ChatGPT for Healthcare、ChatGPT for Teachers 和 API 平台 | 具体留存开关、审计映射和合同例外并未完全公开。 |
| 客户所有权/保密措辞 | 有效承诺 | 同一组商业产品加 API 平台 | 公开章节来源无法替代完整 DPA、信任门户导出件或客户专属安全附件。 |
| 用户可控训练设置 | 有效控制 | 消费者和登录用户体验 | 公开材料没有完全打通消费者设置与所有企业部署模式之间的关系。 |
| Advanced Account Security | 有效控制 | 账户保护和恢复 | 企业身份提供商映射和强制策略细节公开有限。 |
| Teach / test / share 安全流程 | 公开陈述的框架 | 模型开发和部署生命周期 | 相比买方尽调对红队范围和例外处理的需求,公开框架仍偏高层。 |
| 状态页正常运行时间和事故披露 | 可公开观察 | APIs、ChatGPT、Codex 和 FedRAMP 组件 | 汇总正常运行时间不等于客户专属 SLA 或支持承诺。 |
| 独立或政府技术审查 | 存在但有选择性 | o1 评估和已发布 system cards | 不是每个当前 OpenAI 界面都有同等深度的公开评估。 |
| 关于 memory/web-search 攻击面的公开负面信号 | 外部负面报告 | ChatGPT memory 和 web search 功能 | 公开补救细节不如漏洞叙事本身完整。 |
公开信任证据明显强于纯营销材料,但仍不是一套可完整下载的尽调包。买方应把公开页面当作筛选证据,再索取有门户材料支撑的文件和合同附件。
[CE010, CE011, CE012, CE013, CE014, CE027]5.4 展示项
06客户
6.1 客户基础横跨员工席位、API 构建者和伙伴主导的企业项目
OpenAI 的客户基础已经不再是单一的“给知识工作者用 ChatGPT”故事。公开记录现在支撑三条不同收入路线。第一,OpenAI 有直接员工席位业务:其客户案例 hub 称 100 万家企业使用 OpenAI;2025 年 11 月里程碑文章称超过 100 万名企业客户直接使用 OpenAI;VentureBeat 报道截至 2025 年 6 月有 300 万付费商业用户;2025 企业报告称 OpenAI 服务超过 700 万个 ChatGPT 工作场所席位。合起来看,这既意味着广度,也意味着速度:OpenAI 正在吸引大量组织,同时也在这些组织内部加深权益。 第二,OpenAI 有 API 驱动的构建者分层,且用量越来越接近生产级。企业报告称,超过 9,000 家组织已处理逾 100 亿 tokens,近 200 家组织已超过 1 万亿 tokens,单个组织平均 reasoning-token 消耗量同比约增长 320 倍。API 使用也不局限于传统软件公司。OpenAI 自己的报告称,客户支持和内容生成合计约占 API 活动的 20%,非科技公司 API 使用量同比增长 5 倍。Mercado Libre 的 Verdi 平台进一步强化了开发者/平台买方地图:OpenAI 称 17,000 名开发者可在其上构建,覆盖 30,000 多个微服务。 第三,渠道和伙伴正在放大企业触达。TechCrunch 基于 Ramp 的快照显示,OpenAI 正在美国商业 AI 支出中赢得份额;Microsoft 客户案例和 Accenture 2025 年 12 月公告则显示,OpenAI 正通过 Azure 和咨询主导 rollout 被拉进更大的转型预算。这对尽调很重要,因为买方、用户和付款方并不总是同一方。有些账户由中央 IT 或数字化负责人购买 ChatGPT 席位,用于广泛员工使用;有些由产品或平台团队为 API 用量买单;还有一些由服务或云伙伴塑造部署路径。因此,OpenAI 看起来不像窄应用供应商,更像一个多界面企业平台,同时跑着几套 GTM 动作。[CU001, CU002, CU003, CU004, CU009, CU010]
| 客群 | 买方 / 用户 / 付费方 | 主要用例 | 有证据支撑的规模 | 收入 / 战略价值 | 缺口 |
|---|---|---|---|---|---|
| 工作场景席位型知识工作 | 买方:职能负责人或 IT;用户:员工;付费方:部门或企业预算 | 在 ChatGPT 工作区内做起草、分析、搜索和任务辅助 | 超过 100 万家企业客户;超过 700 万个工作场景席位;300 万付费企业用户 | 最大的可见直销动作,也最容易从初始落地切入更大账户 | 公开资料没有按队列拆分 Business、Enterprise 和 Edu 的席位组合或 ARPU。 |
| 开发者 / 平台团队 | 买方:CTO、产品或平台负责人;用户:开发者;付费方:工程或产品预算 | 通过 API 构建产品内助手、搜索、客户支持、自动化和自定义应用 | 超过 9,000 个组织消耗超过 100 亿 tokens;近 200 个组织超过 1 万亿 tokens;Mercado Libre 支撑 17,000 名开发者和 30,000+ 微服务 | 战略上重要,因为 API 使用把 OpenAI 嵌进客户产品,并抬高切换成本 | 直接 OpenAI 与 Azure 关联 API 的份额不公开。 |
| 受监管企业知识工作 | 买方:数字、运营或合规负责人;用户:顾问和银行员工;付费方:企业采购 | 金融场景中的研究、摘要、内部知识和工作流自动化 | Morgan Stanley 报告顾问团队采用率 98%;BBVA 报告周活使用率 83%,并创建 20,000+ 个自定义 GPT | 高价值证明:OpenAI 能穿过受监管环境的治理门槛 | 合同规模、续约时间和定价仍未披露。 |
| 客户支持自动化 | 买方:支持负责人;用户:支持坐席和终端客户;付费方:支持或运营预算 | 自动化聊天、电话和多语言客户服务 | Klarna 首月处理 230 万次对话;Intercom 平均对话解决率 41%,端到端电话解决率 53% | ROI 导向强,劳动力和延迟收益可衡量 | OpenAI 不公布支持行业按客户队列划分的整体留存或利润率。 |
| 零售辅助 | 买方:数字或门店运营负责人;用户:购物者和门店员工;付费方:零售 IT / 运营预算 | 产品发现、项目建议和员工赋能 | Lowe's 称 Mylow Companion 已在 1,700+ 家门店上线,Mylow 将建议与产品发现连接起来 | 说明 OpenAI 能同时支撑面向客户和面向员工的零售工作流 | 公开资料未披露 GMV、转化率或客单价提升。 |
| 合作伙伴牵头的企业转型 | 买方:CIO 或转型赞助人;用户:客户团队和顾问;付费方:企业转型预算 | OpenAI 采用通过咨询项目和 Azure 部署加速 | Accenture 将装备数万名专业人员;Microsoft 客户案例展示了零售和银行场景中的 Azure OpenAI | 在不只依赖直销的情况下,扩大进入大型企业和受监管部署的触达 | 合作伙伴来源收入、客户所有权和毛利分成不公开。 |
公开证据最能支撑的分群方式是上市路径和工作流,而不是收入带。证据指向一个混合客户群:直接席位、API 构建者和合作伙伴中介的企业部署并存。
[CU002, CU004, CU009, CU010, CU016, CU017]| 指标 | 数值 | 日期 | 置信度 | 影响 | 缺失分母 |
|---|---|---|---|---|---|
| 直接企业客户 | 1000000 | 2025-11-05 | 中 | OpenAI 已经拥有广泛组织渗透,而不只是少数标杆客户。 | 未知按产品、层级和支出带的拆分。 |
| 付费企业用户 | 3000000 | 2025-06-04 | 中 | 2025 年内商业席位增长明显加速。 | 用户数不等于客户数,也不等于纯企业 logo 数。 |
| ChatGPT 工作场景席位 | 7000000 | 2025-12-08 | 中 | OpenAI 的已安装工作场景基数,在企业 AI 平台里已经异常庞大。 | 没有按 Business、Enterprise 或 Edu 拆分的席位组合。 |
| 企业席位增长 | 同比 9 倍 | 2025-12-08 | 中 | 账户在扩大权益,而不只是签初始试点。 | 未披露起始席位基数或流失抵消。 |
| Enterprise 周消息量 | 自 2024 年 11 月以来 8 倍 | 2025-12-08 | 中 | 使用强度随部署深度一起放大。 | 没有按账户分布或付费/免费使用拆分。 |
| Custom GPTs 和 Projects 周用户 | 年初至今 19 倍 | 2025-12-08 | 中 | 账户越来越多在搭建可复用内部工作流,而不只使用通用聊天。 | 没有 GPT 重度账户数量或续约影响。 |
| 通过 Custom GPT / Project 发送的 Enterprise 消息 | 20% | 2025-12-08 | 中 | 结构化工作流正在已部署账户内变得重要。 | 没有按行业或产品层级拆分。 |
| 单组织平均推理 token 消耗 | 同比 320 倍 | 2025-12-08 | 中 | 构建者客户正在转向更高级的推理工作负载。 | token 增长无法直接映射到净留存或收入。 |
| 超过 100 亿 tokens 的组织数 | 9000 | 2025-12-08 | 中 | 生产 API 使用已经具备广泛基础。 | token 阈值不揭示合同规模或盈利能力。 |
| 超过 1 万亿 tokens 的组织数 | ~200 | 2025-12-08 | 中 | 已经存在一条有意义的超大部署长尾。 | 未披露头部客户集中度或收入贡献。 |
公开轨迹在使用量和席位增长上很强,但分母仍弱,投资人难以把采用转化为续约质量和收入估计。
[CU002, CU004, CU005, CU006, CU007, CU008]| 扩张驱动 | 集中度风险 | 影响 | 尽调路径 |
|---|---|---|---|
| 初始落地后的席位扩张 | 公开资料没有展示 headline growth 之下的席位组合、净扩张或流失 | 高 | 索取按套餐和客群拆分的席位数、付费席位净新增和扩张。 |
| Custom GPTs 和 Projects 带来的工作流嵌入 | 没有公开证据把 GPT 重度使用与续约、ACV 提升或扩张耐久性连接起来 | 中 | 索取结构化工作流账户与通用聊天重度账户的队列留存和支出扩张对比。 |
| API 生产渗透 | 公开资料没有区分直接 OpenAI API 消耗与经 Azure 中介的消耗 | 高 | 索取直接 API 与合作伙伴渠道之间的收入、利润率和客户数拆分。 |
| 服务和云渠道杠杆 | 合作伙伴牵头交易可能削弱直接客户所有权,并可能通过少数渠道集中支持或收入 | 高 | 索取合作伙伴来源 ARR、利润率、支持义务和续约所有权。 |
| AI-native 收入耐久性 | 续约质量和头部账户敞口可见度弱,可能在强劲采用标题下掩盖集中度风险 | 高 | 索取前 10 大客户 ARR、渠道集中度、合同期限和续约排期。 |
扩张案例真实存在,且按公开标准异常强;但集中度案例仍披露不足。因此本表把正向扩张证据与承销耐久性所需的具体缺失披露放在一起。
[CU004, CU007, CU008, CU009, CU010, CU021]OpenAI 客户采用通常从广泛认知和席位实验开始,再沿着结构化工作流、生产集成和伙伴驱动扩张不断加深。
[CU002, CU004, CU007, CU008, CU009, CU016]公开证据支撑这样一条漏斗:从广泛商业采用进入更深工作流使用,再进入生产和渠道扩张。
[CU002, CU004, CU005, CU007, CU008, CU009]6.2 具名客户证明最强处在受监管知识工作、客服自动化和零售辅助
最好的公开客户证明并不平均分布在所有分层,而是集中在几个反复出现的 jobs-to-be-done 上。受监管知识工作是最清晰的一类。Morgan Stanley 称 AI @ Morgan Stanley Debrief 在取得客户同意后起草笔记、行动项、邮件和 Salesforce 更新;该公司还称 98% 的 Financial Advisor 团队已采用 AI @ Morgan Stanley Assistant。BBVA 是另一个高信号金融服务案例:OpenAI 2025 年 11 月案例研究称,员工每周节省约 3 小时,周活使用率达到 83%,员工创建超过 20,000 个自定义 GPT,其中约 4,000 个经常使用。合在一起,这两个引用说明,OpenAI 可以在重监管环境中走出试点,而这类环境尤其看重工作流质量和信任。 客服自动化是第二个主要证明集群。Klarna 2024 年 2 月发布称,其 OpenAI 驱动助手首月处理 230 万次对话,覆盖三分之二客服聊天,将重复询问减少 25%,并把解决时间从 11 分钟降至不到 2 分钟。Intercom 给出的是同一故事更像企业软件的版本。Intercom 称 Fin 在数千名客户中平均解决 41% 的对话;OpenAI 2025 企业报告称 Fin Voice 将延迟降低 48%,端到端解决 53% 的电话,并让转人工电话快 40%。这些不是泛泛的“AI 助手”推荐语,而是直接绑定客服运营的工作流指标。 零售辅助是第三个集群。Lowe’s 称 Mylow 为客户提供实时项目指导和产品发现,Mylow Companion 已覆盖 1,700 多家门店的员工。Digital Commerce 360 独立确认,Mylow 已在 Lowe’s 网站和移动 app 上向登录购物者上线;OpenAI 还称,员工工具运行在与面向客户顾问相同的基础上。Morgan Stanley、BBVA、Klarna、Intercom 和 Lowe’s 合在一起形成的模式很重要:OpenAI 的公开客户引用,在买方能指向具体工作流、可衡量节省时间或明确服务水平改善的场景里最强,而不是靠模糊的创新叙事撑着。[CU029, CU030, CU031, CU032, CU033, CU034]
| 客户 | 客群 | 部署 / 用例 | 生产还是试点 | 成效 | 限制 |
|---|---|---|---|---|---|
| Morgan Stanley | 财富管理 / 金融 | 顾问知识检索,以及会议复盘和跟进工作流 | 生产 | 98% 的顾问团队采用了该助手;一名顾问称每次会议约节省 30 分钟 | 公开证据对采用和工作流收益很强,但没有合同规模、续约或错误率。 |
| Lowe's | 零售 | 面向客户的 Mylow 顾问,以及面向员工的 Mylow Companion | 生产 | Mylow 已面向购物者上线,Companion 覆盖超过 1,700 家门店 | 公开资料显示部署广度,但未披露转化、客单价或劳动生产率提升。 |
| BBVA | 银行 | 全企业内部生产力和自定义 GPT 推广 | 生产 | 每名员工每周节省 3 小时;周活使用率 83%;创建 20,000+ 个 GPT | 参与度证据很强,但没有按业务单元公开续约或支出数据。 |
| Klarna | 金融科技 / 支持自动化 | 面向退款、退货、支付和服务问题的客户服务助手 | 生产 | 首月 230 万次对话;覆盖三分之二聊天;重复咨询减少 25%;不到 2 分钟解决 | 发布期数据有说服力,但仍早,不能证明多年留存。 |
| Intercom | 企业软件 / 支持自动化 | Fin 和 Fin Voice 跨渠道自动化支持,包括电话 | 生产 | Fin 客户平均解决率 41%;端到端电话平均解决率 53%;延迟降低 48% | 成效由供应商发布,逐客户差异不公开。 |
这张表刻意聚焦当前公开证明中工作流最具体、成效最可衡量的案例,覆盖受监管知识工作、零售、金融科技支持和企业软件。
[CU029, CU030, CU031, CU032, CU033, CU035]| 指标 | 数值 | 客群 | 置信度 | 尽调追问 |
|---|---|---|---|---|
| 顾问团队采用 | 98% | Morgan Stanley 财富管理 | 高 | 确认这是周活使用、logo 采用还是权益覆盖,并索取按办公室 / 队列划分的续约情况。 |
| 周活使用 | 83% | BBVA 企业推广 | 中 | 索取更广员工上线后的月活趋势、队列衰减和席位扩张。 |
| 重复咨询变化 | -25% | Klarna 支持自动化 | 中 | 测试该降幅是否延续到发布期之后,并是否跨市场成立。 |
| 平均对话解决率 | 平均 41%;最高 50% | Intercom Fin 客户群 | 中 | 按 ICP 索取 Fin 队列的留存、扩张和毛利贡献。 |
| 端到端电话解决率 | 平均 53% | Intercom Fin Voice | 中 | 按行业、问题复杂度和人工升级率拆分。 |
| 公开 NRR / GRR / 流失披露 | OpenAI 整体企业客户群 | 低 | 按产品和头部队列索取 logo 留存、席位留存、GRR、NRR、流失。 | |
| 公开合同期限 / 续约条款披露 | OpenAI 整体企业客户群 | 低 | 按套餐索取标准期限、续约结构、席位下限和折扣区间。 |
OpenAI 的公开记录最强的是使用量和工作流成效,而不是经典 SaaS 留存披露。数值列为 null 的地方,缺失指标是实质性尽调缺口,不是格式遗漏。
[CU025, CU026, CU030, CU039, CU044, CU046]OpenAI 最强的公开客户证明,在部署状态和工作流具体性上已经成熟,但在续约和合同可见度上较弱。
[CU030, CU033, CU035, CU038, CU039, CU040]6.3 扩张在用量代理指标上很强,但仍缺少 SaaS 式耐久性披露
判断留存韧性,最有力的证据不是公开披露的 NRR,而是 OpenAI 自己的参与度数据已经显示出多深的工作流嵌入。2025 企业报告称,Enterprise 每周消息数增长 8 倍,Custom GPT 和 Projects 的周活用户年初至今增长 19 倍,约 20% 的 Enterprise 消息已经跑在这些结构化工作流里,工作场景席位超过 700 万。这些都是有用的“落地后扩张”信号,因为它们说明客户不只是买席位然后闲置。报告还称,75% 受访员工表示输出质量或速度提升,75% 表示现在能完成过去做不了的任务,企业用户估计每个活跃日节省 40–60 分钟。合在一起看,OpenAI 进入一个账户后,继续向更多职能扩张这件事是说得通的。 即便如此,公开材料仍明显弱于成熟企业软件尽调所需要的证据包。OpenAI 的企业隐私页面方向上有帮助:客户拥有并控制业务数据,OpenAI 默认不会用业务数据训练,部分企业计划还允许客户控制保留期限。这些披露大概率有助于敏感环境里的采购和扩张。但它们没有回答更硬的商业问题:公开材料里仍没有 NRR、GRR、流失率、标准合同期限、最低席位要求、折扣区间,也没有头部客户集中度表。 这块披露缺口之所以重要,是因为外部基准提醒人谨慎。ChartMogul 的 AI 流失报告警告,部分 AI 合约带有三个月退出条款,退出率达 70–80%;实验性收入或高度按用量计费的收入,可能比实际更像可持续收入。High Alpha 的 2025 基准从上行角度给出相反提示:公司规模做大后,扩张收入会成为主要增长引擎,结果对齐的定价也能支撑更好的留存。合理的综合判断是,OpenAI 的扩张动能显然很强,但公开披露仍没有证明,今天的采用能否转化为上市公司级企业软件通常要证明的续约和 cohort 韧性。[CU004, CU005, CU006, CU007, CU008, CU012]
| 留存视角 | 公开代理指标 | 可用数值 | 限制 | 尽调追问 |
|---|---|---|---|---|
| 席位采用深度 | ChatGPT 工作场景席位 | 7000000 | 已安装席位不是留存指标,也不显示续约。 | 按队列和续约批次索取月度付费席位。 |
| 席位扩张速度 | 企业席位增长 | 同比 9 倍 | 如果漏斗顶部新增足够大,高流失也能与增长并存。 | 按套餐索取毛新增、毛流失和净新增。 |
| 工作流粘性代理 | 通过 Custom GPT / Project 发送的 Enterprise 消息 | 20% | 结构化工作流使用意味着嵌入更深,但不等于续约率。 | 索取 GPT 重度账户与通用聊天账户的留存对比。 |
| 日常用户价值代理 | 单个活跃日节省时间 | 40-60 分钟 | 自报生产力不等于合同耐久性。 | 按实测生产力结果拆分索取续约和扩张。 |
| 客服重复使用代理 | Klarna 重复咨询变化 | -25% | 单一客户运营指标不是跨客户队列指标。 | 索取支持自动化客户的重复使用和续约指标。 |
| 已发布留存队列图 | 没有公开的按时间桶留存百分比,无法诚实支撑计划中的队列图。 | 按客群索取月度或季度 logo 留存、席位留存、token 留存、GRR 和 NRR。 |
这张表刻意替代原计划的留存 / 重复队列图。公开资料提供了有用的耐久性代理指标,但没有针对 OpenAI 客户队列的真实月度或年度留存百分比。
[CU004, CU005, CU008, CU013, CU044]6.4 渠道依赖和集中度仍是客户尽调的主要盲点
OpenAI 的企业增长覆盖面很广,但公开证据也显示,客户关系并不总是由 OpenAI 直接掌握。Microsoft 的客户案例显示,Azure OpenAI 已用于 Worten、Discovery Bank 等零售和银行部署;Accenture 则称,会让数万名专业人员使用 ChatGPT Enterprise,并为金融服务、医疗、公共部门和零售构建基于 OpenAI 的项目。这些材料是正向信号,因为它们证明生态杠杆有效;但它们也意味着,部分企业扩张可能由云厂商或服务伙伴居中完成,而不是完全靠 OpenAI 自己的直销和客户成功体系拿下并服务。 客户集中度风险也因此很难只靠公开证据判断。公开材料没有说明,客户增长中有多少来自直接席位销售,又有多少来自 Azure 相关用量、伙伴主导实施或大型标杆客户。公开材料也没有披露最大客户对 ARR、续约排期或支持负担的贡献。这块缺口很关键,因为一家公司可以同时呈现出色的顶线采用,又仍然依赖少数超大型企业关系或渠道。 因此,尽调姿态应当是双向的。正面看,OpenAI 似乎已经建出一台客户机器,可以面向知识工作者、开发者、零售商、银行、金融科技公司和企业软件厂商销售,并且扩张信号异常快。风险侧看,公开集中度和合同数据缺位,意味着客户故事仍不完整;管理层还需要拿出按 cohort 的留存、头部客户暴露、伙伴来源 ARR,以及把采用转成耐久经常性收入的商业条款。[CU021, CU022, CU024, CU053, CU054, CU055]
6.5 图表要点
07风险
7.1 法律和监管风险仍是严重度最高的一组风险
OpenAI 最大的风险桶仍是法律和监管,而不是单纯的产品竞争。New York Times 诉讼之所以重要,是因为它已经把公司拉进取证和数据保留问题,直接触及用户隐私承诺。OpenAI 表示,影响消费者 ChatGPT 和 API 内容的非常规保留令已于 2025 年 9 月结束,标准删除做法已经恢复,这是实质缓释。但公司自己的诉讼页面也显示,该案仍活跃到需要专门的辩护网站;CourtListener 也确认底层联邦案卷仍在进行。也就是说,隐私负担可以下降,但诉讼负担并没有消失。 第二组法律风险是治理。即便 OpenAI 已在 2024 年 3 月重置董事会,Musk v. Altman 仍把非营利使命和控制权问题留在法庭上。OpenAI 的审查材料有价值,因为它收窄了一类担忧:2023 年的裂痕被定性为信任破裂,而不是隐藏的安全或融资丑闻。即便如此,该结构仍把董事任命权集中在 Foundation,公益公司结构也不能消除未来治理争议的风险。欧洲审查又叠了一层。EDPB ChatGPT taskforce 报告和欧盟委员会 2025 AI Act 落地推进都意味着,前沿模型提供商在欧洲要面对比普通软件公司更重的合规、文档和隐私要求。[CR001, CR002, CR003, CR004, CR005, CR006]
| 规则/许可/案件 | 司法辖区 | 状态 | 可能性 | 严重性 | 缓释 | 剩余敞口 | 尽调路径 |
|---|---|---|---|---|---|---|---|
| New York Times 版权与证据开示诉讼 | 美国 | 诉讼进行中;非常规留存令已于 2025-09-26 结束,但实体问题和证据开示仍在继续 | 中 | 高 | 标准删除已恢复,披露了有限的法律/安全访问,业务客户隐私控制仍然有效 | 高 | 获取外部律师关于剩余请求、保存范围和客户通知义务的备忘录。 |
| Musk v. Altman 治理和使命偏离诉讼 | 美国 | 联邦诉讼进行中 | 中 | 高 | 董事会扩容、WilmerHale 审查和后续结构更新降低但未消除治理敞口 | 中高 | 审阅诉状、所请求救济、监管往来,以及任何非营利法敞口分析。 |
| 欧盟隐私和 AI Act 审查 | 欧盟 | 执行中且可能形成先例的审查 | 高 | 高 | DPA、企业留存控制和成文隐私承诺降低了客户侧风险,但没有降低合规负担 | 中高 | 梳理 GPAI 合规负责人、训练数据记录,以及逐司法辖区的隐私姿态。 |
| 公开条款下的版权/IP 分配 | 美国/欧盟/全球 | 诉讼叠加合同分配风险 | 中 | 高 | OpenAI 发布了商业条款和 DPA,但公开材料并未完整披露谈判形成的赔偿和 SLA 附表 | 中高 | 承保受监管或内容密集型用例前,先审查赔偿除外、客户限制和来源控制。 |
各行按剩余严重性排序,而非按时间排序。公开证据足以给主要诉讼和合规桶排序,但不足以声称覆盖所有司法辖区的完整案件清单。
[CR001, CR002, CR003, CR004, CR005, CR006]7.2 运营可靠性和安全风险已经是平台风险,不再是边缘问题
OpenAI 已经不是一个宕机只会让人烦躁的新奇应用。状态历史现在显示,ChatGPT、API、登录、文件处理和特定模型服务都反复出现事故;2026-05-04 的连接器写入事故正好说明,可靠性为什么已经变成核心尽调变量。公开的 uptime 数字方向上很强,但 ChatGPT 99.83% 的 uptime 仍足以在 OpenAI 嵌入研究、编程、支持和受监管工作流后造成可见中断。因此,一个前沿 AI 平台可以总体可用性很高,同时对那些一个连接器或一个模型路由坏掉就停工的买家来说,仍然存在运营风险。 安全风险也是同样的结构。SecurityWeek 关于 memory 和网页搜索功能漏洞的报道显示,便利功能扩大攻击面的速度可以非常快。OECD.AI 事故记录和 OpenAI 自己关于 Axios 入侵事件的回应表明,软件供应链问题已不是假设风险。OpenAI 在公开层面确实有实质缓释:安全和隐私页面、漏洞报告流程、bug bounty 计划。但这些是控制信号,不等于公共用户和企业客户已经看到足够的事后复盘、红队证据或合同支持的服务承诺,可以有把握地判断剩余风险。[CR016, CR017, CR018, CR019, CR020, CR021]
| 故障模式 | 发生概率 | 严重性 | 缓释成熟度 | 剩余敞口 | 待补缺口 |
|---|---|---|---|---|---|
| 跨产品反复事故和连接器中断 | 高 | 高 | 中 | 中高 | 公开可用性快照不能替代客户级 SLA、支持承诺或事故复盘披露。 |
| 记忆、浏览和 agent 工具拉大功能攻击面 | 中 | 高 | 中低 | 高 | 需要看到浏览、记忆、连接器和 agent 权限的红队结果与发布控制。 |
| 开发者工具或软件供应链被攻破 | 中 | 高 | 中 | 中高 | 需要公开叙事之外的软件签名、SBOM 和修复时间线证据。 |
| 信任资料包透明度落后于买方尽调需求 | 高 | 中 | 中 | 中 | 公开控制措施方向上较强,但完整认证、控制映射和支持承诺仍有一部分被门控或缺失。 |
运营风险更多来自反复可用性事故、快速扩大的攻击面和不完整的公开信任材料的组合,而不是某一个灾难性事件。
[CR016, CR017, CR018, CR019, CR020, CR021]| 角色/职能 | 依赖或缺口 | 发生概率 | 严重性 | 缓释 | 尽调路径 |
|---|---|---|---|---|---|
| CEO/战略资本关系 | Sam Altman 仍是治理信任、伙伴谈判、融资和公共叙事的核心 | 中 | 关键 | 应用层领导班子和更宽的董事会边际上降低集中度 | 要求提供继任计划、授权矩阵和关键人员保护。 |
| 董事会/使命治理 | Foundation 控制让董事会任命权在董事会更新后仍保持集中 | 中 | 高 | 独立董事和 PBC 架构提升流程纪律 | 审查保留事项、监管沟通,以及 Foundation 与 Group 之间的升级路径。 |
| 商业运营跨度 | OpenAI 同时跑消费者、企业、API、伙伴和基础设施几条线 | 高 | 中高 | 专门的应用层负责人显示角色分工在改善 | 测试产品、销售、支持和基础设施经济性是否由独立记分卡和负责人管理。 |
| 安全与可靠性归属 | 功能快速发布拉大运营表面,速度超过公开控制文件化 | 高 | 高 | 已有漏洞赏金、漏洞报告和公开状态披露 | 要求提供具名控制负责人、事故复盘和高风险功能的发布治理证据。 |
执行风险的重点不是 OpenAI 能不能发货,而是以当前规模能否让治理、基础设施、GTM 和安全流程保持一致。
[CR008, CR010, CR011, CR016, CR020, CR021]| 风险 | 可监测触发因素 | 阈值/事件 | 行动含义 |
|---|---|---|---|
| 诉讼/隐私 | 留存或证据开示负担再次扩大 | 新法院命令、监管行动或客户通知要求推翻既有删除承诺 | 暂停企业信心假设,并立即提高合规成本折价。 |
| 可靠性/产品质量 | 持续服务降级 | 季度可用性明显低于当前披露水平,或重大事故多次持续数日 | 下调增长和利润率假设;把它视为产品-基础设施执行失手,而不是噪音。 |
| 平台集中度 | 交易对手经济性恶化 | Microsoft 重置不利、Oracle 融资承压,或关键项目出现可见容量延迟 | 提高要求回报、限制仓位,并要求更完整的合同披露。 |
| 资本强度 | 伙伴融资的产能跑在商业化前面 | 紧急融资、建设延期,或有证据显示固定承诺跑赢使用量转化 | 除非管理层提供清晰的承诺支出桥,否则从可跟踪风险升级为击穿投资论点。 |
| 治理/领导层 | 领导层或董事会再次破裂 | 高管离职、董事会冲突,或使命治理诉讼出现重大不利裁定 | 视为暂停并重做承保的事件,直到治理稳定重新建立。 |
这些标准设计成可由公开新闻、状态页、法院进展和管理层尽调材料监测,而不是凭直觉。
[CR002, CR016, CR018, CR019, CR026, CR032]7.3 依赖、资本强度和执行跨度,会把风险传导到利润率和估值
OpenAI 的伙伴栈既是护城河,也是集中度问题。Microsoft 从 2023 年合作延伸以来仍是基础性伙伴,公开记录也继续把 Azure 视为研究、API 和产品的结构性核心。Oracle 和 Stargate 更适合解读为部分多元化,而不是独立:它们增加了产能和融资选项,但也增加了第二层交易对手,而这些交易对手自己的资本计划现在已经明确引用 OpenAI 需求。CoreWeave 则增加了第三个集中节点:最高 $11.9B 的基础设施协议,再加上 OpenAI 参股。任一关系如果重新定价、停滞或交付不足,损害都可能传导到产品可用性、毛利率和融资需求,而不是只留在采购环节。 财务模型风险在于,公开需求代理指标是真实的,但公开的耐久性和义务数据不完整。OpenAI 的 2025 企业报告显示,企业消息增长 8 倍、推理 token 消耗增长 320 倍,这意味着宕机、合规限制或渠道冲击现在会打到比一年前大得多的已安装工作流基础。但 ChartMogul 关于短退出期 AI 合约、以及收入在真正耐久前看似经常性的警告,是必要的校正。领导层广度正在改善,尤其是 Fidji Simo 接任 Applications 角色;但公司仍要求投资者相信,它能同时跑好研究、消费者、企业、API 和伙伴多条战线,却没有公开足以闭环的继任、P&L 或承诺支出细节。[CR026, CR027, CR028, CR029, CR030, CR031]
| 依赖 | 交易对手 | 角色 | 集中度 | 失效场景 | 严重性 | 缓释 | 剩余敞口 |
|---|---|---|---|---|---|---|---|
| 核心云与经济栈 | Microsoft | 长期云、分发和经济交易对手 | 极高 | 商业重新定价、关系破裂或容量摩擦会同时打击可用性、毛利率和谈判筹码 | 关键 | Oracle/Stargate 和其他伙伴带来一定分散,2026 年重置也显示条款可再谈 | 高 |
| 大规模建设产能 | Oracle / Stargate | 扩张产能和融资伙伴 | 高 | 数据中心容量或融资延迟会拖慢路线图,或迫使 OpenAI 采购更昂贵算力 | 高 | 多站点和分阶段部署拓宽了选项,但 Oracle 自身也在针对需求举借大额融资 | 中高 |
| 专用基础设施合同 | CoreWeave | 专用 AI 基础设施供应商 | 高 | 供应商执行、定价或集中度压力会损害 OpenAI 的算力可用性或成本结构 | 高 | 股权绑定和多供应商策略有帮助,但披露合同相对正常软件外包仍然体量巨大 | 中高 |
| 企业销售通路 | Accenture 和 Azure 相关伙伴 | 伙伴主导的实施和获客渠道 | 中 | OpenAI 拿到使用量,却失去直接客户归属、续约议价权或毛利率可见度 | 中高 | 直营产品和 API 销售降低单一渠道依赖,但公开渠道组合仍未披露 | 中 |
依赖问题不是单点故障,而是一套相互咬合的算力、融资和渠道栈;其经济性只有部分公开。
[CR026, CR027, CR028, CR029, CR030, CR031]OpenAI 最高的剩余风险集中在法律 / 监管敞口和伙伴集中度;尽管已有可见缓释,运营和执行风险仍然重要。
该热力图采用有来源支撑的序位评分,而非编造概率;目标是给剩余敞口排序,不是假装精确。
[CR002, CR008, CR016, CR020, CR026, CR028]OpenAI 的主要风险沿少数通道传导:法律与隐私负担、平台集中度、可靠性事件,最终都会落到客户信任、利润率、融资需求和估值上。
该图是有来源支撑的定性图:展示因果方向,不编合成概率权重。
[CR001, CR016, CR026, CR028, CR030, CR032]OpenAI 的依赖网络已不再是一条双边关系,而是跨云、基础设施和渠道伙伴的多节点栈;但这些节点仍足够集中,足以影响判断。
该依赖关系图只表示方向、不做量化;它标出合作伙伴失灵会在哪里传导进运营或经济账。
[CR026, CR027, CR028, CR030, CR032, CI042]7.4 缓释措施看得见,但投资判断仍需要明确否决标准
和许多高速增长的 AI 公司相比,OpenAI 确实有更多公开的缓释脚手架。2026 DPA 和 Services Agreement 为企业和开发者客户定义了法律框架。企业隐私材料称,客户拥有并控制数据,业务数据默认不用于训练,关键企业计划可以控制保留期限。安全报告通道和运行中的 bug bounty 也显示,公司没有把信任当作边缘问题。治理也已有改善:WilmerHale 审查、董事会补强和 Fidji Simo 的运营角色,都降低了所有关键职能压在一个人或一套临时流程上的概率。 但在这一章里,缓释措施降低风险,并不消灭风险。缺失项恰恰是投资者最关心的东西:合并口径的承诺支出表、完整客户合同条款、SLA 和赔偿附件、伙伴来源 ARR、头部客户暴露和继任材料。正确姿态因此应是有条件的,而不是宿命式的。如果诉讼扩大、可用性恶化、Microsoft 或 Oracle 经济条件转向不利,或者领导层治理再次破裂,这些都不是普通执行失误,而是投资 thesis 的击穿触发项。反过来,如果管理层能拿出合同包、集中度表和支出承诺,并且比公开记录暗示的更干净,剩余风险可能明显压缩。[CR010, CR014, CR015, CR023, CR024, CR025]
7.5 图表要点
08估值
8.1 建议:质量真实,但入场价仍跑在公开披露前面
OpenAI 显然已经不再按投机性原型定价。第三方报道称,公司 2025 年年化收入超过 200 亿美元,到 2026 年 2 月底超过 250 亿美元;OpenAI 自己也称 ChatGPT 每周服务超过 8 亿用户,且超过 100 万企业客户直接使用其产品。这些都是顶级需求信号,也解释了为什么资本仍愿意在前所未有的规模上涌向公司。 话虽如此,当前决策仍应对价格敏感,而不是对崇拜敏感。CNBC 的 2026 年 3 月融资报道把入场估值定在 8520 亿美元,Reuters 随即把挑战描述为在这个估值上找到焦点。公开证据支持溢价倍数,但不支持把这个价格视为显然便宜。市场实际上是在被要求提前承销 run-rate 收入转化、伙伴融资的基础设施扩张和未来公开市场接纳度,而在此之前,投资者还看不到申报级别的利润率、烧钱、优先权或现金流数据。这组条件足以支持“继续深入研究”的判断,但不足以在当前估值给出买入建议。[CV001, CV002, CV004, CV005, CV006, CV025]
建议的核心,是强需求证据碰上偏紧价格和披露不足;合作伙伴经济性是主要摇摆因素。
[CV001, CV004, CV025, CV026, CV033, CV036]只有 OpenAI 能守住高收入倍数,或把收入拉到远高于当前公开年化收入水平,当前估值才显得站得住。
收入门槛只是用最新公开融资估值做 EV/收入的简单桥接,不是贴现现金流模型输出。
[CV004, CV022, CV023, CV024]OpenAI 在市场拉力和产品验证上得分极高,但经济性可见度和价格支撑明显低得多。
分数是为投委会讨论基于公开证据综合出的 0-10 分序位判断。
[CV004, CV019, CV025, CV026, CV033, CV036]8.2 价格背景:OpenAI 相对上市巨头偏贵,但相对前沿 AI 稀缺性并不荒谬
最有用的估值框架不是单一比率,而是一组比率。用最新公开 run-rate 证据看,OpenAI 8520 亿美元估值意味着,按 2026 年 2 月底数字约为年化收入的 34.1 倍,按 2025 年底数字约为 39.8 倍。这些水平远高于 Microsoft、Oracle 等上市平台,用当前市值和收入数据看,后两者大约分别是 10.1 倍和 7.7 倍。单看这个镜头,OpenAI 是偏紧的。 但同一组数据也说明,当前价格并没有明显脱离前沿 AI 私募市场行为。Anthropic 2026 年 2 月融资,对 Reuters 后来归因给它的年化收入数字,隐含约 42.2 倍;Palantir 在公开市场约按过去收入的 77.3 倍交易。因此,OpenAI 落在一个不舒服的中间位置:明显贵于多元化上市巨头,但又没有明确贵于能找到的最佳前沿模型或 AI 溢价可比公司。承销问题在于,OpenAI 缺少公开申报文件,投资者无法判断它到底应该落在哪个可比桶里。[CV004, CV005, CV008, CV010, CV011, CV012]
| 论据 | 方向 | 改变判断的因素 |
|---|---|---|
| 年化变现水平对一家私营软件公司已属顶级。 | 投资论点 | 经证实的 ARR 转化放缓,或企业使用量反转,会削弱溢价理由。 |
| 消费者和商业使用共同形成异常广的变现分发面。 | 投资论点 | 如果这些用户不能转化为持久确认收入或盈利 cohort,单有规模还不够。 |
| 伙伴融资的基础设施能比自筹资产负债表更快支撑增长。 | 投资论点 | 如果这些合同重新定价或制造隐藏义务,同一结构就会变成利润率拖累。 |
| 当前估值筛选后仍远高于上市既有巨头。 | 反论点 | 若走向申报级披露,或确认收入大幅提高,该溢价会更容易防守。 |
| 公开证据在优先权、利润率和现金生成上太薄,无法给出买入判断。 | 反论点 | 披露经审计经济性和 2026 年融资栈,会显著提升可承保性。 |
这些论据刻意对价格敏感;它们描述的是当前估值已经假设了什么,而不只是产品本身是否成立。
[CV004, CV005, CV006, CV012, CV015, CV019]| 可比对象 | 指标 | 倍数/估值/状态 | 相关性 | 局限 |
|---|---|---|---|---|
| Microsoft | 市值/过去十二个月收入 | ~10.1x | 衡量规模化 AI 分发、云和企业变现的最佳公开基准。 | 多元化巨头;AI 只是业务之一。 |
| Oracle | 市值/过去十二个月收入 | ~7.7x | 可用来观察大型既有云平台如何变现 AI 基础设施需求。 | 多元数据库和云厂商,不是前沿模型实验室。 |
| Palantir | 市值/过去十二个月收入 | ~77.3x | 显示公开市场愿意为稀缺 AI 溢价软件敞口付多少钱。 | 业务组合不同,已是盈利上市公司,且政府业务敞口明显。 |
| Anthropic | 私募估值/年化收入 | 基于 3800 亿美元估值和约 90 亿美元年化收入,约 42.2 倍 | 2026 年披露信息中最接近的前沿模型私营可比公司。 | 私募估值和 run-rate 收入不能直接拿来和经审计公开指标相比。 |
| Mistral | 私募估值状态 | 2025 年 9 月估值 140 亿美元 | 显示下一档已披露前沿模型公司仍比 OpenAI 低很多。 | 地域、规模和披露深度不同;本资料集没有可逐项对照的收入指标。 |
可比组刻意混合,因为没有一个公开同行能干净匹配 OpenAI 的消费者、企业、模型和基础设施敞口。
[CV008, CV009, CV010, CV011, CV012, CV013]8.3 情景逻辑:上行存在,但下行路径更容易用公开证据解释
多头情景并不难描述。如果 OpenAI 能继续把前沿模型领先转化为企业支出,守住消费者规模,并在伙伴经济性改善的同时,把 run-rate 收入变成耐久的确认收入,公司就有可能复利长成足以支撑甚至超过当前估值的收入基数。Microsoft 收入分成上限重置在这里很重要,因为收入基数已经以数百亿美元计,留存经济性哪怕边际改善,也会产生很大差异。 基准和空头情景更容易由公开证据支撑。OpenAI 正同时尝试把消费者使用、企业工作流、开发者需求和基础设施级伙伴关系货币化,同时又计划承担按软件标准看极其庞大的算力支出和供应商承诺。Stargate 5000 亿美元雄心、Oracle 为 OCI capacity 制定的 450 亿至 500 亿美元融资计划,以及 CoreWeave 119 亿美元 OpenAI 合同,都指向同一个事实:估值压缩不需要需求崩塌。只要收入转化、利润率捕获或伙伴杠杆低于 8520 亿美元估值已经内含的预期,就足够了。[CV003, CV004, CV007, CV019, CV027, CV028]
| 场景 | 假设 | 估值/回报逻辑 | 关键风险 | 概率信号 |
|---|---|---|---|---|
| 乐观 | OpenAI 将当前 run-rate 转化为 600 亿美元收入,留存经济性改善,同时继续拿到前沿模型稀缺性溢价。 | 按 25 倍收入计算,股权价值可接近 1.5 万亿美元,摊薄前约为当前估值的 1.8 倍。 | 需要产品领先持续、伙伴配合,并且没有重大治理或监管断裂。 | 中低 |
| 基准 | 收入扩大到约 400 亿美元,公私市场投资者给予 18 倍倍数,接近高端但不是亢奋的平台估值。 | 隐含价值约 7200 亿美元,低于当前 8520 亿美元入场估值。 | 即便执行不错,也未必跑赢倍数正常化。 | 中 |
| 悲观 | 收入停在当前约 250 亿美元 run-rate 附近,市场按更成熟的基础设施/软件平台给 OpenAI 12 倍估值。 | 隐含价值约 3000 亿美元,相比最新轮估值有严重下行。 | 即使需求没有崩,利润率或伙伴经济性不及预期也足以触发压缩。 | 中 |
之所以采用收入倍数情景,是因为公开证据对利润和现金流转化太弱,撑不起 DCF 质量模型。
[CV004, CV019, CV022, CV023, CV024, CV039]| 触发因素 | 证据/阈值 | 对投资论点的传导 | 行动含义 |
|---|---|---|---|
| 收入转化不及预期 | run-rate 停在约 250 亿美元附近,没能向支撑溢价所需的约 340 亿-470 亿美元区间增长。 | 当前估值失去支撑 25 倍-18 倍框架所需的增长桥。 | 不追加资金;按悲观情景重新承保下行。 |
| 伙伴经济性恶化 | Microsoft 收入分成重置未能改善留存经济性,或新供应商条款更严苛。 | 毛利率和自由现金流转化仍不足以支撑溢价倍数。 | 将立场从偏贵下调为昂贵。 |
| 产能变成资产负债表毒性 | Stargate、Oracle 或 CoreWeave 承诺从伙伴支持转为 OpenAI 的硬性义务,却没有匹配的商业化。 | 估值从软件稀缺性故事变成资本开支故事。 | 任何新增承保前,先要求承诺支出排期。 |
| 治理或法律冲击重启控制权问题 | 新的董事会或非营利控制权破裂会重定价执行信心。 | 财务数据追上前,稀缺性溢价先被压缩。 | 暂停投资工作,直到治理条款厘清。 |
这些是杀死投资论点的触发器,不是泛泛风险。每一项都会直接损害支撑当前估值所需的具体假设。
[CV003, CV019, CV022, CV023, CV024, CV027]| 主题 | 缺失证据 | 重要性 | 负责人或尽调路径 |
|---|---|---|---|
| 2026 年融资栈 | 1220 亿美元轮次中的优先权顺位、清算权、估值调整条款、最惠国条款(MFN)以及任何老股交易条款。 | 这些条款决定头部估值对新钱是否经济上真实。 | 管理层 + 律师会议;融资文件。 |
| 收入质量 | 按分部打通年化 run-rate 到确认收入的桥,并补充流失率、NRR 和头部客户集中度。 | 当前公开证据对 run-rate 很强,对耐久性偏弱。 | CFO 尽调包;收入 cohort 导出。 |
| 利润率留存 | 按产品线拆分的毛利率,以及扣除 Microsoft/Oracle/CoreWeave 经济性后的毛利率。 | 不知道 OpenAI 留住多少收入,就无法判断当前倍数。 | 财务 + 基础设施采购审查。 |
| 承诺支出 | 算力和数据中心伙伴中的照付不议、预付款和最低承诺义务。 | 下行不只来自需求弱,也可能来自义务跑赢商业化。 | 采购和司库排期。 |
| 治理转换路径 | 任何 IPO 相关治理变化、非营利控制权让步或董事会权利变化。 | 是否具备公开市场准备度,取决于治理变得可读,也取决于增长继续强劲。 | 董事会材料和外部律师备忘录。 |
| 现金和资金跑道 | 当前现金余额、烧钱和融资时点假设,按乐观/基准/悲观需求情景拆分。 | 即便私募估值巨大,只要资本需求持续跑赢内部生成,估值也会压缩。 | 司库模型和董事会批准预算。 |
这些要求刻意收窄到投资关键项;每一项都可能改变建议或可接受价格。
[CV003, CV004, CV019, CV027, CV028, CV029]公开证据能支撑很宽的估值区间,因为倍数选择几乎和增长本身一样重要。
区间是基于情景的收入倍数测算,服务于投委会讨论,不是管理层指引。
[CV004, CV019, CV022, CV023, CV024, CV039]8.4 退出就绪度和最终尽调:公开证据方向强,但还不完整到能让投资者下判断
公开证据已经足以让人认真看待 OpenAI 作为未来流动性候选人的可能。CNBC 把最新融资直接和 IPO 预期挂钩,Reuters 也描述 OpenAI 和 Anthropic 都在为最终上市铺路。需求、品牌触达和融资能力已经不再是门槛。门槛在披露质量。Microsoft、Oracle 和 Alphabet 都有披露密集的投资者门户;OpenAI 没有,因为它仍是由 Foundation 控制、通过 OpenAI Group PBC 组织的私营公司。 在 8520 亿美元估值上,这块缺口比普通成长期估值下更重要。投资者仍无法公开查看把好感转化为可承销 conviction 所需的精确 2026 年优先权栈、ratchet、清算权、经审计分部收入、客户集中度、确认毛利率、承诺算力流出或现金消耗轨迹。管理层拿出这套材料之前,正确姿态是有纪律的好奇:保持跟进,定义清晰否决触发项,并愿意在价格重置或披露包开始像上市公司级尽调资料室时,重新评估判断。[CV001, CV003, CV030, CV031, CV032, CV033]
| 决策字段 | 当前判断 | 决策含义 |
|---|---|---|
| 建议 | 继续研究 | 保持跟踪,但在没有私人尽调前,不按当前估值承保新增资金。 |
| 置信度 | 中 | 公开证据能支撑需求判断,但不足以支撑申报级经济性。 |
| 风险评级 | 高 | 倍数压缩可能通过算力成本、伙伴经济性或治理冲击传导。 |
| 估值态度 | 偏贵 | 当前价格要求高端收入转化和持续稀缺性倍数。 |
| 持有/退出姿态 | 只有私人尽调补齐缺口,才适合持有 3-5 年 | 短期上行主要靠情绪;基本面支撑需要时间和执行。 |
| 价格纪律 | 不能在 8520 亿美元估值下不看价格买入 | 要么要求明显更好条款,要么要求披露显著改善。 |
这个判断反映的是当前入场价格,并非泛泛认为 OpenAI 是弱公司。
[CV001, CV019, CV022, CV023, CV024, CV033]8.5 图表要点
免责声明
本报告是基于公开证据的尽调快照,不构成投资建议。重要的财务、法律、技术和合同事实仍未公开;作出任何投资决定前,应直接向管理层并通过一手文件核验。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | OpenAI describes itself as an AI research and deployment company whose mission is to ensure AGI benefits all of humanity. | 中 | SO001, SO003 |
| CO002 | OpenAI currently consists of the nonprofit OpenAI Foundation and the for-profit OpenAI Group PBC, with the Foundation governing the Group. | 中 | SO001, SO002 |
| CO003 | OpenAI was founded in 2015 as a nonprofit. | 中 | SO002, SO025 |
| CO004 | OpenAI's for-profit business was established in 2019 under foundation control. | 中 | SO005, SO025 |
| CO005 | The OpenAI Foundation appoints all members of the OpenAI Group board and can replace directors at any time. | 中 | SO002 |
| CO006 | OpenAI Group is organized as a public benefit corporation that is required to advance its stated mission and consider broader stakeholder interests. | 中 | SO002, SO025 |
| CO007 | OpenAI's current board includes Bret Taylor, Adam D'Angelo, Sue Desmond-Hellmann, Zico Kolter, Paul Nakasone, Adebayo Ogunlesi, Nicole Seligman, and Sam Altman. | 中 | SO002, SO022 |
| CO008 | OpenAI added Sue Desmond-Hellmann, Nicole Seligman, and Fidji Simo to the board in March 2024 and returned Sam Altman to the board. | 中 | SO010, SO011 |
| CO009 | Adebayo Ogunlesi joined OpenAI's board on January 14, 2025. | 中 | SO022 |
| CO010 | Greg Brockman became OpenAI's President in the May 5, 2022 leadership team update. | 中 | SO009 |
| CO011 | Mira Murati became OpenAI's Chief Technology Officer in the May 5, 2022 leadership team update. | 中 | SO009 |
| CO012 | Fidji Simo was announced as OpenAI's CEO of Applications in May 2025 and reports directly to Sam Altman. | 中 | SO023 |
| CO013 | The WilmerHale review found the 2023 board crisis was precipitated by a breakdown in trust between the prior board and Sam Altman. | 中 | SO010 |
| CO014 | The WilmerHale review said the prior board's action did not arise from product safety or security, the pace of development, OpenAI's finances, or statements to investors, customers, or partners. | 中 | SO010 |
| CO015 | On January 23, 2023, OpenAI and Microsoft announced a multiyear, multibillion-dollar extension of their partnership following Microsoft's earlier 2019 and 2021 investments. | 中 | SO005, SO006 |
| CO016 | The January 2023 partnership materials say Azure remains the exclusive cloud provider for OpenAI workloads across research, API, and products. | 中 | SO005, SO006 |
| CO017 | OpenAI launched ChatGPT as a research preview on November 30, 2022. | 中 | SO007 |
| CO018 | OpenAI's GPT-4 research page is dated March 14, 2023. | 中 | SO008 |
| CO019 | The OpenAI Foundation says it made an initial $50 million commitment in 2025 to nonprofits and mission-focused organizations working at innovation and public good. | 中 | SO004 |
| CO020 | OpenAI says the Foundation holds equity in the for-profit currently valued at about $130 billion and can receive additional ownership if valuation milestones are met. | 中 | SO002, SO025 |
| CO021 | OpenAI's structure materials say the OpenAI Foundation holds a 26% equity stake in OpenAI Group. | 中 | SO002, SO016, SO025 |
| CO022 | OpenAI's structure materials say Microsoft holds roughly 27% of OpenAI Group after recapitalization. | 中 | SO002, SO016 |
| CO023 | CNBC and Bloomberg reported that OpenAI closed a $122 billion funding round at an $852 billion post-money valuation on March 31, 2026. | 中 | SO014, SO015 |
| CO024 | Bloomberg and CNBC reported that Amazon, Nvidia, and SoftBank were the three largest participants in the March 2026 funding round, at $50 billion, $30 billion, and $30 billion respectively. | 中 | SO014, SO015 |
| CO025 | TechCrunch reported that OpenAI raised $6.6 billion at a $157 billion post-money valuation on October 2, 2024. | 中 | SO013 |
| CO026 | TechCrunch reported that OpenAI's total capital raised reached $17.9 billion after the October 2024 round. | 中 | SO013 |
| CO027 | TechCrunch identified Thrive Capital, Microsoft, Nvidia, SoftBank, Khosla Ventures, Altimeter Capital, Fidelity, and MGX among backers of the October 2024 round. | 中 | SO013 |
| CO028 | Forbes listed OpenAI as headquartered in San Francisco with Sam Altman as CEO and 6,400 employees as of December 2025. | 中 | SO017 |
| CO029 | Forbes listed OpenAI's 2025 revenue at $3.7 billion. | 中 | SO017 |
| CO030 | Forbes said ChatGPT had roughly 500 million weekly users and about 2 million paying enterprise users. | 中 | SO017 |
| CO031 | OpenAI's status history shows recurring incidents affecting ChatGPT, the API, login, file uploads, and specific models across 2025 and 2026. | 中 | SO012 |
| CO032 | OpenAI says it learned of the New York Times lawsuit on December 27, 2023 and disputes the copyright claims. | 中 | SO018 |
| CO033 | OpenAI said on June 5, 2025 that limited April-September 2025 user data would be retained under legal hold and accessed only by a small audited legal and security team. | 中 | SO024 |
| CO034 | JD Supra's analysis said a May 13, 2025 preservation order required OpenAI to retain ChatGPT conversation logs affecting over 400 million users worldwide. | 中 | SO019 |
| CO035 | Loeb's June 2025 note says OpenAI-related copyright litigation includes multidistrict and related disputes touching New York Times and other journalism claims. | 中 | SO020 |
| CO036 | CourtListener shows Elon Musk filed a complaint on August 5, 2024 against Altman, Brockman, and multiple OpenAI entities. | 中 | SO021 |
| CO037 | Axios reported that OpenAI completed a controversial recapitalization in October 2025 that split the organization into a nonprofit foundation controlling OpenAI Group PBC. | 中 | SO016 |
| CO038 | OpenAI says the Foundation will initially focus a $25 billion commitment across health and AI resilience work. | 中 | SO025 |
| CO039 | OpenAI's partnership materials show the company commercializes its models through ChatGPT, its API, Azure OpenAI Service, and enterprise-facing products. | 中 | SO005, SO007, SO008 |
| CO040 | OpenAI says the plan includes essential admin controls, SAML SSO, and MFA. | 中 | SM002 |
| CO041 | OpenAI says enterprise privacy commitments cover ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu, ChatGPT for Healthcare, ChatGPT for Teachers, and the API platform. | 中 | SM004 |
| CO042 | OpenAI API pricing offers batch processing at a 50% discount relative to standard processing. | 中 | SM003 |
| CO043 | OpenAI launched ChatGPT Enterprise with enterprise-grade security and privacy. | 中 | SI001 |
| CO044 | OpenAI described ChatGPT Enterprise as an AI assistant for work customized for an organization. | 中 | SI001 |
| CO045 | OpenAI says ChatGPT Team was renamed ChatGPT Business in August 2025 without changing features or pricing. | 中 | SI002 |
| CO046 | OpenAI launched ChatGPT Team as a self-serve plan after ChatGPT Enterprise. | 中 | SI002 |
| CO047 | Microsoft says the Work Trend Index combines survey responses from 31,000 people in 31 countries with productivity signals. | 中 | SO028 |
| CO048 | Menlo Ventures reports that 72% of enterprise decision-makers anticipate broader generative AI adoption in the near future. | 中 | SM007 |
| CO049 | Menlo Ventures reports that 47% of enterprise generative AI solutions are built in-house versus 53% bought from vendors. | 中 | SM007 |
| CO050 | Menlo Ventures reports that buyers most often select generative AI tools for measurable value at 30% and industry-specific customization at 26%, while price is cited by only 1% of respondents. | 中 | SM007 |
| CO051 | Menlo Ventures reports that the LLM layer commands $3.5 billion of enterprise investment. | 中 | SM007 |
| CO052 | Menlo Ventures reports that closed-source foundation models hold 81% enterprise market share versus 19% for open-source alternatives. | 中 | SM007 |
| CO053 | Menlo Ventures reports that fine-tuning appears in only 9% of production models. | 中 | SM007 |
| CO054 | Deloitte reports that 78% of respondents expect to increase overall AI spending in the next fiscal year. | 中 | SM008 |
| CO055 | Deloitte reports that 26% of organizations are exploring autonomous agent development to a large extent and 42% to some extent. | 中 | SM008 |
| CO056 | Deloitte’s fourth-wave survey covered 2,773 director-to-C-suite respondents across 14 countries and 6 industries. | 中 | SM008 |
| CO057 | McKinsey estimates current generative AI and adjacent technologies could automate work activities that absorb 60% to 70% of employees’ time today. | 中 | SM011 |
| CO058 | Stack Overflow reports that more than 65,000 developers responded to its 2024 survey. | 中 | SO031 |
| CO059 | Anthropic maintains an Economic Index focused on understanding AI’s effects on the economy. | 低 | SO032 |
| CO060 | Anthropic’s Privacy Center explicitly groups API, Console, Team, and Enterprise plans under commercial customers. | 中 | SO034 |
| CO061 | Anthropic’s responsible disclosure policy says the security of its systems and user data is a top priority and was last updated on 2025-02-14. | 中 | SO035 |
| CO062 | Google’s Gemini API pricing documentation highlights long context, structured outputs, function calling, and agent support in the Gemini 3 family. | 中 | SO037 |
| CO063 | TechCrunch reported that OpenAI had committed $1.4 trillion to infrastructure commitments over the next few years. | 低 | SI013 |
| CO064 | TechCrunch reported on August 7, 2025 that GPT-5 had become OpenAI’s new flagship model line. | 中 | SO040 |
| CO065 | TechCrunch reported in May 2025 that OpenAI’s enterprise adoption appeared to be accelerating relative to rivals. | 中 | SU001 |
| CO066 | Forbes framed OpenAI’s 2026 revenue model as moving beyond ChatGPT toward a broader enterprise platform story. | 低 | SO041 |
| CO067 | OpenAI’s 1 million business customers post named Amgen, Commonwealth Bank, Booking.com, Cisco, Lowe’s, Morgan Stanley, T-Mobile, Target, and Thermo Fisher Scientific as customers. | 中 | SU002 |
| CO068 | OpenAI’s 2025 enterprise report says BBVA regularly uses more than 4,000 GPTs. | 中 | SU003 |
| CO069 | OpenAI’s 2025 enterprise report says the median sector expanded by more than 6x year over year. | 中 | SU003 |
| CO070 | OpenAI’s 2025 enterprise report says technology was the fastest-growing sector at 11x year-over-year customer growth. | 中 | SU003 |
| CO071 | OpenAI’s 2025 enterprise report says healthcare was the second-fastest-growing sector at 8x year-over-year customer growth. | 中 | SU003 |
| CO072 | OpenAI’s 2025 enterprise report says manufacturing was the third-fastest-growing sector at 7x year-over-year customer growth. | 中 | SU003 |
| CO073 | OpenAI’s 2025 enterprise report says Australia, Brazil, the Netherlands, and France were growing faster than the global average in paying business customers, at 187%, 161%, 153%, and 146% respectively. | 中 | SU003 |
| CO074 | VentureBeat reported that OpenAI launched business connectors and Record Mode alongside the 3 million paying business-user milestone. | 中 | SU004 |
| CO075 | Klarna said its AI assistant was available 24/7 in 23 markets and communicated in more than 35 languages. | 中 | SU016 |
| CO076 | Klarna estimated that the AI assistant would drive a $40 million profit improvement in 2024. | 中 | SU016 |
| CO077 | Intercom says Fin is available in 45 languages. | 中 | SU018 |
| CO078 | OpenAI’s 2025 enterprise report says Fin is already saving customers hundreds of millions of dollars annually. | 中 | SU003 |
| CO079 | OpenAI’s Mercado Libre case says Verdi handled 10% of disputes on a major site within months. | 中 | SU019 |
| CO080 | OpenAI’s New York Times litigation page said the court also rejected claims brought by Ziff Davis. | 中 | SO018 |
| CO081 | MarketScreener reported that Anthropic climbed to roughly $9 billion in annualized revenue. | 中 | SV003 |
| CO082 | CNBC reported in April 2026 that Anthropic was discussing a new funding round at a $900 billion valuation. | 中 | SO044 |
| CO083 | CNBC reported that Anthropic was valued at $61.5 billion after its March 2025 funding round. | 中 | SP018 |
| CO084 | Oracle reported $553 billion of Q3 FY26 remaining performance obligations, up 325% year over year. | 中 | SI012 |
| CO085 | OpenAI said Microsoft's 2023 extension was a multi-year, multi-billion investment. | 中 | SO005 |
| CO086 | OpenAI said its Microsoft partnership extension preserved the company's nonprofit-governed capped-profit structure. | 中 | SO005 |
| CO087 | OpenAI said the nonprofit is now the OpenAI Foundation and the for-profit is OpenAI Group PBC. | 中 | SO002 |
| CO088 | Menlo estimated that enterprises spent $37 billion on generative AI in 2025. | 中 | SP016 |
| CO089 | Menlo estimated that the application layer captured $19 billion of enterprise generative AI spend in 2025. | 中 | SP016 |
| CO090 | Frontier-model private-market pricing remained euphoric in spring 2026 because Anthropic moved from a $380 billion February round to reported $900 billion fundraising talks by late April. | 中 | SV004, SO044 |
| CM001 | OpenAI markets business products as tools to create, code, and innovate with tools and APIs. | 中 | SM001 |
| CM002 | OpenAI positions ChatGPT Business and Enterprise as products intended to empower an entire workforce. | 中 | SM001 |
| CM003 | OpenAI says its business offering includes specialized AI agents for teams, including Codex for software development and workspace agents for recurring workflows. | 中 | SM001 |
| CM004 | OpenAI describes one ChatGPT pricing plan as a development-focused plan with pay-as-you-go pricing. | 中 | SM002 |
| CM005 | The same plan has no fixed seat fee and charges based on usage. | 中 | SM002 |
| CM006 | OpenAI says it does not train its models on business data by default. | 中 | SM004, SM002 |
| CM007 | OpenAI says customers control retention duration for ChatGPT Enterprise, ChatGPT for Healthcare, and ChatGPT Edu. | 中 | SM004 |
| CM008 | OpenAI API pricing lists GPT-5.5 input at $5.00 per 1M tokens and output at $30.00 per 1M tokens. | 中 | SM003 |
| CM009 | OpenAI says ChatGPT Edu is built for universities to deploy AI to students, faculty, researchers, and campus operations. | 中 | SM005 |
| CM010 | OpenAI describes ChatGPT Edu as an affordable offering for universities. | 中 | SM005 |
| CM011 | Microsoft reports that 82% of leaders say 2025 is a pivotal year to rethink core aspects of strategy and operations. | 中 | SM006 |
| CM012 | Microsoft reports that 82% of leaders expect to use digital labor to expand their workforce in the next 12 to 18 months. | 中 | SM006 |
| CM013 | Microsoft reports that 46% of leaders say their organization is already using agents to fully automate workstreams or business processes. | 中 | SM006 |
| CM014 | Microsoft identifies customer service, marketing, and product development as the top AI investment priorities. | 中 | SM006 |
| CM015 | Microsoft reports that 71% of workers at Frontier Firms say their company is thriving versus 37% globally. | 中 | SM006 |
| CM016 | Menlo Ventures reports that enterprise generative AI spending reached $13.8 billion in 2024, up more than 6x from $2.3 billion in 2023. | 中 | SM007 |
| CM017 | Menlo Ventures reports that 60% of enterprise generative AI investment currently comes from innovation budgets. | 中 | SM007 |
| CM018 | Menlo Ventures reports that 40% of enterprise generative AI spending comes from permanent budgets. | 中 | SM007 |
| CM019 | Menlo Ventures reports that 58% of permanent-budget generative AI spending is redirected from existing allocations. | 中 | SM007 |
| CM020 | Menlo Ventures reports that enterprises spent $4.6 billion on generative AI applications in 2024, nearly 8x the $600 million reported the prior year. | 中 | SM007 |
| CM021 | Menlo Ventures reports that code copilots had 51% enterprise adoption in 2024. | 中 | SM007 |
| CM022 | Menlo Ventures reports that support chatbots had 31% enterprise adoption in 2024. | 中 | SM007 |
| CM023 | Menlo Ventures reports that enterprise search and retrieval had 28% enterprise adoption in 2024. | 中 | SM007 |
| CM024 | Menlo Ventures reports that data extraction and transformation had 27% enterprise adoption in 2024. | 中 | SM007 |
| CM025 | Menlo Ventures reports that meeting summarization had 24% enterprise adoption in 2024. | 中 | SM007 |
| CM026 | Menlo Ventures reports that implementation costs contributed to 26% of failed pilots. | 中 | SM007 |
| CM027 | Menlo Ventures reports that data privacy hurdles contributed to 21% of failed pilots. | 中 | SM007 |
| CM028 | Menlo Ventures reports that disappointing ROI contributed to 18% of failed pilots. | 中 | SM007 |
| CM029 | Menlo Ventures reports that hallucinations contributed to 15% of failed pilots. | 中 | SM007 |
| CM030 | Menlo Ventures reports that IT accounts for 22% of enterprise generative AI spending. | 中 | SM007 |
| CM031 | Menlo Ventures reports that Product and Engineering account for 19% of enterprise generative AI spending. | 中 | SM007 |
| CM032 | Menlo Ventures reports that Data Science accounts for 8% of enterprise generative AI spending. | 中 | SM007 |
| CM033 | Menlo Ventures reports that Support accounts for 9% of enterprise generative AI spending. | 中 | SM007 |
| CM034 | Menlo Ventures reports that Sales accounts for 8% of enterprise generative AI spending and Marketing 7%. | 中 | SM007 |
| CM035 | Menlo Ventures reports that HR and Finance each account for 7% of enterprise generative AI spending, while Legal accounts for 3%. | 中 | SM007 |
| CM036 | Menlo Ventures reports that healthcare represented $500 million of enterprise generative AI spend in 2024. | 中 | SM007 |
| CM037 | Menlo Ventures reports that legal represented $350 million of enterprise generative AI spend in 2024. | 中 | SM007 |
| CM038 | Menlo Ventures reports that financial services represented $100 million of enterprise generative AI spend in 2024. | 中 | SM007 |
| CM039 | Menlo Ventures reports that organizations typically deploy three or more foundation models in their AI stacks. | 中 | SM007 |
| CM040 | Menlo Ventures reports that OpenAI’s enterprise model share fell from 50% in 2023 to 34% in 2024. | 中 | SM007 |
| CM041 | Menlo Ventures reports that Anthropic’s enterprise model share rose from 12% in 2023 to 24% in 2024. | 中 | SM007 |
| CM042 | Menlo Ventures reports that security and safety are cited by 46% of organizations when moving to a new LLM provider. | 中 | SM007 |
| CM043 | Menlo Ventures reports that RAG has 51% adoption in production architectures, up from 31% a year earlier. | 中 | SM007 |
| CM044 | Menlo Ventures reports that agentic architectures already power 12% of implementations. | 中 | SM007 |
| CM045 | Deloitte reports that more than two-thirds of respondents say 30% or fewer of their experiments will be fully scaled in the next three to six months. | 中 | SM008 |
| CM046 | Deloitte reports that nearly three-quarters of respondents say their most advanced generative AI initiative is meeting or exceeding ROI expectations. | 中 | SM008 |
| CM047 | Deloitte reports that regulatory compliance rose from 28% in wave 1 to 38% in wave 4 as the top barrier to developing and deploying generative AI tools. | 中 | SM008 |
| CM048 | Deloitte reports that 69% of respondents say fully implementing a governance strategy will take over a year. | 中 | SM008 |
| CM049 | Gartner forecasts worldwide generative AI spending will reach $644 billion in 2025. | 中 | SM009, SM010 |
| CM050 | VentureBeat reports that Gartner’s $644 billion 2025 forecast implies 76.4% year-over-year growth versus 2024. | 中 | SM010 |
| CM051 | McKinsey estimates generative AI could add $2.6 trillion to $4.4 trillion annually across 63 use cases. | 中 | SM011 |
| CM052 | McKinsey estimates that about 75% of generative AI value falls across customer operations, marketing and sales, software engineering, and R&D. | 中 | SM011 |
| CM053 | McKinsey estimates banking could capture $200 billion to $340 billion of annual value from generative AI. | 中 | SM011 |
| CM054 | McKinsey estimates retail and consumer packaged goods could capture $400 billion to $660 billion of annual value from generative AI. | 中 | SM011 |
| CM055 | PwC reports that AI-skilled workers earned an average 56% wage premium in 2024, up from 25% a year earlier. | 中 | SM012 |
| CM056 | PwC reports that job availability grew 38% in roles more exposed to AI between 2019 and 2024. | 中 | SM012 |
| CM057 | PwC reports that industries most exposed to AI saw 27% growth in revenue per employee versus 9% in the least exposed industries. | 中 | SM012 |
| CM058 | PwC reports that required skills are changing 66% faster in jobs most exposed to AI. | 中 | SM012 |
| CM059 | OECD.AI frames managing the risks and benefits of generative AI as a dedicated policy topic. | 中 | SM013 |
| CM060 | OECD.AI states that AI has risks and all actors must be accountable. | 中 | SM014 |
| CM061 | The European Commission describes the AI Act as the first-ever comprehensive legal framework on AI worldwide. | 中 | SM015 |
| CM062 | NIST maintains an AI Risk Management Framework as an official U.S. risk-management reference for AI deployments. | 中 | SM016 |
| CM063 | The public IDC spending-guide URL reviewed for this chapter returned an unavailable article response on the access date. | 中 | SM017 |
| CP001 | OpenAI markets ChatGPT Business/Codex as usage-priced with no fixed seat fee on the reviewed pricing page. | 中 | SM002 |
| CP002 | The reviewed OpenAI business pricing page advertises essential admin controls, SAML SSO, and MFA for business use. | 中 | SM002 |
| CP003 | OpenAI says it does not train models on business data by default. | 中 | SM004 |
| CP004 | OpenAI says business customers own and control their business data and can control retention windows for several managed plans. | 中 | SM004 |
| CP005 | OpenAI lists GPT-5.5 API pricing at $5.00 input, $0.50 cached input, and $30.00 output per 1M tokens. | 中 | SM003 |
| CP006 | OpenAI lists GPT-5.4 API pricing at $2.50 input, $0.25 cached input, and $15.00 output per 1M tokens. | 中 | SM003 |
| CP007 | Anthropic documents Claude availability through Amazon Bedrock and Google Cloud Vertex AI, indicating multi-cloud distribution beyond Anthropic direct channels. | 中 | SP003 |
| CP009 | TechCrunch reported that Anthropic held 32% enterprise LLM usage share versus OpenAI’s 25% and 42% coding share versus OpenAI’s 21% in mid-2025 Menlo data. | 中 | SP017 |
| CP010 | Google Workspace Enterprise publicly advertises DLP, context-aware access, enterprise data regions, endpoint management, and AI classification for Google Drive. | 中 | SP004 |
| CP011 | Google Cloud says it supports compliance through certifications, documentation, third-party audits, and an AI trust paper. | 中 | SP005 |
| CP012 | Google Cloud’s agent platform pricing page positions Gemini enterprise AI on the same infrastructure as Google and explicitly markets it as enterprise-ready AI. | 中 | SP006 |
| CP013 | Menlo estimated Google’s share of enterprise LLM spend rose from 7% in 2023 to 21% in 2025. | 中 | SP016 |
| CP014 | Microsoft’s business Copilot page centers AI chat, agents for work, building your own agents, and Copilot inside Microsoft apps. | 中 | SP007 |
| CP015 | Microsoft distributes OpenAI models through Azure OpenAI in Foundry Models. | 中 | SP008 |
| CP016 | Microsoft Learn publishes a dedicated data, privacy, and security page for Azure Direct Models in Microsoft Foundry. | 中 | SP009 |
| CP017 | Menlo’s 2025 horizontal AI section names Microsoft Copilot among the copilots leading a $7.2 billion spend category. | 中 | SP016 |
| CP018 | AWS lists Amazon Q Business Lite at $3 per user per month and Amazon Q Business Pro at $20 per user per month. | 中 | SP011 |
| CP019 | AWS says Bedrock supports Anthropic, Meta, Mistral, and Amazon models and offers batch inference at 50% below on-demand pricing. | 中 | SP010 |
| CP020 | AWS says Bedrock Guardrails can block up to 88% of harmful content and provide 99% accurate validation explanations. | 中 | SP012 |
| CP021 | Meta says Llama 4 models are optimized for easy deployment, cost efficiency, and performance that scales to billions of users. | 中 | SP013 |
| CP022 | Meta says Llama 4 Scout is natively multimodal and offers a 10M context window. | 中 | SP013 |
| CP023 | Menlo reported that open-source enterprise LLM share fell from 19% to 11% while Llama remained the most widely adopted open-weight model in the enterprise. | 中 | SP016 |
| CP024 | Mistral markets Studio as an enterprise platform with privacy, security, and full ownership of customer data. | 中 | SP014 |
| CP025 | Mistral says API rate limits apply at the organization level across all workspaces. | 中 | SP015 |
| CP026 | CNBC reported that Mistral was valued at $14 billion when ASML took a major stake on 2025-09-09. | 中 | SP019 |
| CP027 | Menlo reported that 76% of enterprise AI use cases are purchased rather than built internally in 2025. | 中 | SP016 |
| CP028 | Menlo estimated enterprise AI spend reached $37 billion in 2025, with $19 billion in the application layer. | 中 | SP016 |
| CP029 | Menlo estimated horizontal AI spend at $8.4 billion in 2025, with copilots taking 86% share. | 中 | SP016 |
| CP030 | Menlo estimated Anthropic at 40%, OpenAI at 27%, and Google at 21% of enterprise LLM spend in 2025, with the remaining 12% spread across others. | 中 | SP016 |
| CP031 | Menlo estimated Anthropic at 54% of coding-model share versus OpenAI at 21% in 2025. | 中 | SP016 |
| CP032 | Menlo reported that only 16% of enterprise deployments qualify as true agents in 2025. | 中 | SP016 |
| CP033 | Public cloud distribution patterns suggest enterprise model usage is increasingly multi-homed because Anthropic is available through AWS and Google while OpenAI is distributed separately through Azure. | 中 | SP003, SP008, SP010 |
| CP034 | Google, Microsoft, and Amazon each hold structural distribution power over OpenAI because they can sell AI inside broader suite, cloud, or procurement relationships. | 中 | SP004, SP006, SP007, SP010, SP016 |
| CP035 | OpenAI still combines unusually broad product reach across ChatGPT and APIs, but independent 2025 market-share evidence shows that advantage is no longer translating into uncontested enterprise leadership. | 中 | SM002, SM003, SP016, SP017 |
| CP036 | Public pricing and routing evidence indicates commoditization risk is rising for frontier-model APIs because buyers can compare token economics, use Bedrock-style multi-model access, and shift workloads when performance gaps narrow. | 中 | SM003, SP010, SP011, SP016 |
| CP037 | Open-weight and sovereignty-oriented substitutes reduce buyer lock-in even though closed models still dominate enterprise share today. | 中 | SP013, SP014, SP016 |
| CP038 | OpenAI’s help-center materials show ChatGPT Business pricing and seat structure changed on 2026-04-02, including a $5 per-seat reduction and a new Codex-only seat type. | 中 | SP001, SP002 |
| CI001 | OpenAI’s Business page presents ChatGPT Business and Enterprise as workforce-oriented plans. | 中 | SM001 |
| CI002 | OpenAI’s business pricing page describes Codex as a pay-as-you-go plan for development-focused teams. | 中 | SM002 |
| CI003 | OpenAI’s business pricing page says Codex has no fixed seat fee. | 中 | SM002 |
| CI004 | OpenAI Help says ChatGPT Business billing includes standard ChatGPT seats and usage-based Codex seats. | 中 | SP002 |
| CI005 | OpenAI Help says subscription-based ChatGPT Business seats were reduced by USD $5 per month on April 2, 2026. | 中 | SP001 |
| CI006 | OpenAI Help says ChatGPT Business introduced a Codex-only seat with flexible pricing. | 中 | SP001 |
| CI007 | OpenAI said on August 29, 2025 that ChatGPT Team was renamed ChatGPT Business and that features and pricing did not change. | 中 | SI002 |
| CI008 | OpenAI launched ChatGPT Enterprise as a contact-sales product with enterprise-grade security and privacy. | 中 | SI001 |
| CI009 | OpenAI says it does not train its models on business data by default. | 中 | SM004 |
| CI010 | OpenAI says enterprise customers can control data-retention settings for ChatGPT Enterprise and ChatGPT Edu. | 中 | SM004 |
| CI011 | OpenAI Help says ChatGPT Plus costs $20 per month. | 中 | SI006 |
| CI012 | OpenAI said ChatGPT Pro costs $200 per month and explicitly linked the plan to higher compute usage. | 中 | SI003 |
| CI013 | OpenAI’s API pricing page lists GPT-5.5 input pricing at $5.00 per 1 million tokens. | 中 | SM003 |
| CI014 | OpenAI’s API pricing page lists GPT-5.5 output pricing at $30.00 per 1 million tokens. | 中 | SM003 |
| CI015 | OpenAI’s API pricing page lists GPT-5.4 input pricing at $2.50 per 1 million tokens. | 中 | SM003 |
| CI016 | OpenAI’s API pricing page lists GPT-5.4 output pricing at $15.00 per 1 million tokens. | 中 | SM003 |
| CI017 | OpenAI said in December 2025 that ChatGPT served more than 800 million users every week. | 中 | SI004 |
| CI018 | OpenAI’s Business page uses both try-now and contact-sales calls to action for business adoption. | 中 | SM001 |
| CI019 | OpenAI said Enterprise customers in 2024 already included Block, Canva, Carlyle, The Estée Lauder Companies, PwC, and Zapier. | 中 | SI002 |
| CI020 | TechCrunch reported that the majority of OpenAI’s revenue still came from consumer subscriptions in December 2025. | 中 | SI013 |
| CI021 | TechCrunch reported that OpenAI enterprise message volume grew 8x since November 2024. | 中 | SI013 |
| CI022 | TechCrunch reported that organizations using OpenAI’s API consumed 320 times more reasoning tokens than a year earlier. | 中 | SI013 |
| CI023 | TechCrunch reported that custom GPT use jumped 19x and accounted for 20% of enterprise messages. | 中 | SI013 |
| CI024 | TechCrunch reported that enterprise workers using OpenAI products said they saved roughly 40 to 60 minutes per day. | 中 | SI013 |
| CI025 | OpenAI said Stargate intends to invest $500 billion over four years building AI infrastructure for OpenAI in the United States. | 中 | SI005 |
| CI026 | OpenAI said Stargate would begin deploying $100 billion immediately. | 中 | SI005 |
| CI027 | CoreWeave said its OpenAI infrastructure agreement had a contract value of up to $11.9 billion. | 中 | SI008 |
| CI028 | CoreWeave said OpenAI would invest $350 million of CoreWeave stock as part of the infrastructure deal. | 中 | SI008 |
| CI029 | Sam Altman said the CoreWeave deal complements OpenAI’s commercial deals with Microsoft and Oracle and its joint venture with SoftBank on Stargate. | 中 | SI008 |
| CI030 | Oracle reported March 2025 remaining performance obligations of $130 billion, up 62% year over year. | 中 | SI010 |
| CI031 | Oracle said in March 2025 that it had signed cloud agreements with OpenAI, xAI, Meta, NVIDIA, and AMD. | 中 | SI010 |
| CI032 | Oracle said in March 2025 that it expected to sign its first Stargate contract in the near term. | 中 | SI010 |
| CI033 | Oracle said in March 2025 that it was on schedule to double its data-center capacity during calendar 2025. | 中 | SI010 |
| CI034 | Oracle said GPU consumption for AI training grew 244% over the prior 12 months. | 中 | SI010 |
| CI035 | Oracle said in February 2026 that it expected to raise $45 billion to $50 billion of gross cash proceeds during 2026 to expand OCI. | 中 | SI011 |
| CI036 | Oracle said the 2026 financing plan was intended to build capacity for contracted demand from customers including OpenAI. | 中 | SI011 |
| CI037 | Oracle reported March 2026 remaining performance obligations of $553 billion, up 325% year over year. | 中 | SI012 |
| CI038 | Oracle said in March 2026 that most of the increase in remaining performance obligations related to large-scale AI contracts. | 中 | SI012 |
| CI039 | Oracle said most equipment required for those large-scale AI contracts was funded upfront via customer prepayments or customer-supplied GPUs. | 中 | SI012 |
| CI040 | Oracle said it had already raised $30 billion after announcing an up-to-$50 billion 2026 financing program. | 中 | SI012 |
| CI041 | CoreWeave filed an S-1 registration statement with the SEC on March 3, 2025. | 中 | SI009 |
| CI042 | CNBC reported that OpenAI capped revenue-share payments to Microsoft in a 2026 partnership reset. | 中 | SI007 |
| CI043 | Forbes listed OpenAI’s 2025 revenue at $3.7 billion. | 中 | SO017 |
| CI044 | Microsoft Azure markets Azure OpenAI in Foundry Models as a separate channel for OpenAI model distribution. | 中 | SP008 |
| CE001 | OpenAI presents its API platform as the surface for building AI products. | 中 | SE001 |
| CE002 | OpenAI says its frontier API models deliver advanced intelligence and multimodal capabilities. | 中 | SE001 |
| CE003 | OpenAI lists GPT-5.5 at $5.00 input and $30.00 output per 1M tokens. | 中 | SE001, SM003 |
| CE004 | OpenAI lists GPT-5.5 with 1.05M context length and 128K max output tokens. | 中 | SE001 |
| CE005 | OpenAI lists GPT-5.4 at $2.50 input and $15.00 output per 1M tokens. | 中 | SE001, SM003 |
| CE006 | OpenAI says ChatGPT Business and Enterprise provide unlimited chats plus access to advanced models, tools, and capabilities. | 中 | SM001, SE010 |
| CE007 | OpenAI says its business products include specialized agents such as Codex and workspace agents. | 中 | SM001 |
| CE008 | As of April 2, 2026, ChatGPT Business supports both standard ChatGPT seats and usage-based Codex seats. | 中 | SE011, SP001 |
| CE009 | OpenAI says the ChatGPT Business subscription seat price was reduced by $5 per month and the Codex rate card was aligned with token-based pricing in April 2026. | 中 | SP001 |
| CE010 | OpenAI says organizational data across its business products and API platform remains confidential, secure, and customer-owned. | 中 | SE003 |
| CE011 | OpenAI says it does not train models on organization data by default for its business products and API platform. | 中 | SE003 |
| CE012 | OpenAI says users can choose whether their data is used for training and model improvement. | 中 | SE002 |
| CE013 | OpenAI says Advanced Account Security adds stronger protections against unauthorized access. | 中 | SE002 |
| CE014 | OpenAI’s safety page frames its safety process as teach, test, and share. | 中 | SE004 |
| CE015 | OpenAI describes the Responses API as its core API primitive for building agentic applications. | 中 | SE005, SE013 |
| CE016 | OpenAI added support for remote MCP servers to the Responses API in May 2025. | 中 | SE005, SE022 |
| CE017 | OpenAI added image generation, Code Interpreter, and file search as built-in Responses API tools in May 2025. | 中 | SE005 |
| CE018 | OpenAI launched Codex as a cloud-based software engineering agent powered by codex-1. | 中 | SE006 |
| CE019 | OpenAI says Codex can work on many tasks in parallel. | 中 | SE006 |
| CE020 | Codex launched for ChatGPT Pro, Business, and Enterprise users, then expanded to Plus users on June 3, 2025 with optional internet access during task execution. | 中 | SE006 |
| CE021 | OpenAI launched deep research on February 2, 2025. | 中 | SE007, SE023 |
| CE022 | OpenAI’s February 10, 2026 deep research update added MCP or app connections, trusted-site restriction, real-time progress tracking, and interrupt/refine controls. | 中 | SE007 |
| CE023 | OpenAI launched GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano in the API on April 14, 2025. | 中 | SE008 |
| CE024 | OpenAI says GPT-4.1 models outperform GPT-4o and GPT-4o mini on coding and instruction following and have larger context windows. | 中 | SE008 |
| CE025 | OpenAI says the Sora web and app experiences were discontinued on April 26, 2026. | 中 | SE012 |
| CE026 | OpenAI says the Sora API will be discontinued on September 24, 2026. | 中 | SE012 |
| CE027 | On May 4, 2026, OpenAI’s status page showed an active incident in which some ChatGPT workspace-connector write actions were automatically disabled. | 中 | SE009 |
| CE028 | OpenAI’s status page reported aggregate API uptime of 99.99% for the February-to-May 2026 period. | 中 | SE009 |
| CE029 | OpenAI’s status page reported aggregate uptime of 99.83% for ChatGPT, 99.98% for Codex, and 99.93% for FedRAMP over the February-to-May 2026 period. | 中 | SE009 |
| CE030 | ChatGPT’s pricing page publicly lists Free, Go, Plus, Pro, Business, and Enterprise plans and surfaces features including Agent, Atlas, Deep Research, Voice, and Codex. | 中 | SE010 |
| CE031 | OpenAI’s customer-stories hub says 1M businesses use OpenAI. | 中 | SE018 |
| CE032 | Microsoft says retailer Worten saves 11,000 hours annually with an Azure OpenAI deployment. | 中 | SE020 |
| CE033 | NIST published a pre-deployment evaluation of OpenAI’s o1 model in December 2024. | 中 | SE019 |
| CE034 | OpenAI maintains an official Python library for the OpenAI API. | 中 | SE016 |
| CE035 | OpenAI maintains an official JavaScript / TypeScript library for the OpenAI API. | 中 | SE017 |
| CE036 | TechCrunch reported on March 11, 2025 that OpenAI launched tools to help businesses build AI agents. | 中 | SE022 |
| CE037 | TechCrunch reported on February 2, 2025 that OpenAI unveiled a ChatGPT agent for deep research. | 中 | SE023 |
| CE038 | The Verge reported in December 2024 that OpenAI had closed Sora applications shortly after launch. | 中 | SE024 |
| CE039 | SecurityWeek reported in November 2025 that researchers hacked ChatGPT memories and web-search features. | 中 | SE025 |
| CE040 | OpenAI’s GPT-4o System Card was submitted to arXiv on October 25, 2024. | 中 | SE026 |
| CE041 | OpenAI’s o1 System Card was revised on April 30, 2026. | 中 | SE027 |
| CE042 | OpenAI publishes a dedicated Responses Overview reference in its developer docs. | 中 | SE013 |
| CE043 | OpenAI publishes an all-models catalog in its API docs. | 中 | SE014 |
| CE044 | OpenAI publishes a general API Overview reference in its developer docs. | 中 | SE015 |
| CE045 | Microsoft published a February 2026 customer story describing Discovery Bank using Azure OpenAI in personal banking. | 中 | SE021 |
| CU001 | OpenAI’s customer-stories hub says 1M businesses use OpenAI. | 中 | SE018 |
| CU002 | OpenAI announced that more than 1 million business customers around the world were directly using OpenAI on 2025-11-05. | 中 | SU002 |
| CU003 | OpenAI’s 2025 enterprise report says more than 1 million business customers use OpenAI’s tools. | 中 | SU003 |
| CU004 | OpenAI’s 2025 enterprise report says it serves more than 7 million ChatGPT workplace seats. | 中 | SU003 |
| CU005 | OpenAI’s 2025 enterprise report says ChatGPT Enterprise seats increased approximately 9x year over year. | 中 | SU003 |
| CU006 | OpenAI’s 2025 enterprise report says weekly Enterprise messages grew approximately 8x since November 2024. | 中 | SU003 |
| CU007 | OpenAI’s 2025 enterprise report says weekly users of Custom GPTs and Projects increased approximately 19x year to date. | 中 | SU003 |
| CU008 | OpenAI’s 2025 enterprise report says approximately 20% of all Enterprise messages were processed via a Custom GPT or Project. | 中 | SU003 |
| CU009 | OpenAI’s 2025 enterprise report says more than 9,000 organizations have processed over 10 billion tokens on the API. | 中 | SU003 |
| CU010 | OpenAI’s 2025 enterprise report says nearly 200 organizations have exceeded 1 trillion tokens on the API. | 中 | SU003 |
| CU011 | OpenAI’s 2025 enterprise report says average reasoning token consumption per organization increased approximately 320x over the prior 12 months. | 中 | SU003 |
| CU012 | OpenAI’s 2025 enterprise report says 75% of surveyed workers reported AI improved either the speed or quality of their output. | 中 | SU003 |
| CU013 | OpenAI’s 2025 enterprise report says ChatGPT Enterprise users attribute 40–60 minutes of time saved per active day to AI. | 中 | SU003 |
| CU014 | OpenAI’s 2025 enterprise report says 75% of workers report being able to complete tasks they previously could not perform. | 中 | SU003 |
| CU015 | OpenAI’s 2025 enterprise report says professional services, finance, and technology are the largest ChatGPT Enterprise sectors by current scale. | 中 | SU003 |
| CU016 | OpenAI’s 2025 enterprise report says customer support and content generation represented approximately 20% of API activity. | 中 | SU003 |
| CU017 | OpenAI’s 2025 enterprise report says non-technology firm API use grew 5x year over year. | 中 | SU003 |
| CU018 | OpenAI says business customers own and control their business data. | 中 | SM004 |
| CU019 | OpenAI says it does not train models on business data by default. | 中 | SM004 |
| CU020 | OpenAI says customers control how long data is retained for ChatGPT Enterprise, ChatGPT for Healthcare, and ChatGPT Edu. | 中 | SM004 |
| CU021 | VentureBeat reported that OpenAI’s business user base surged 50% since February 2025 to 3 million paying enterprise customers on 2025-06-04. | 中 | SU004 |
| CU022 | TechCrunch reported that Ramp’s AI Index showed 32.4% of U.S. businesses on Ramp were paying for OpenAI subscriptions in April 2025. | 中 | SU001 |
| CU023 | TechCrunch reported that Ramp’s OpenAI adoption figure was 18.9% in January 2025 and 28% in March 2025. | 中 | SU001 |
| CU024 | TechCrunch reported that Ramp’s April 2025 business-subscription figures were 8% for Anthropic and 0.1% for Google AI. | 中 | SU001 |
| CU025 | ChartMogul says some AI contracts have three-month opt-out provisions where 70–80% of customers opt out. | 中 | SU005 |
| CU026 | ChartMogul argues that experimental AI revenue, short opt-out contracts, and usage-heavy pricing can make ARR look more durable than it is. | 中 | SU005 |
| CU027 | High Alpha says expansion becomes the dominant growth engine beyond roughly $20 million in ARR. | 中 | SU006 |
| CU028 | High Alpha says hybrid, consumption, and outcome-based pricing can pair growth with better retention. | 中 | SU006 |
| CU029 | Morgan Stanley says AI @ Morgan Stanley Debrief generates client-meeting notes, surfaces action items, drafts an email, and saves a note into Salesforce with client consent. | 中 | SU007 |
| CU030 | Morgan Stanley says 98% of Financial Advisor teams have adopted AI @ Morgan Stanley Assistant. | 高 | SU007, SU008 |
| CU031 | One Morgan Stanley advisor said AI @ Morgan Stanley Debrief saves about half an hour per meeting. | 中 | SU007 |
| CU032 | OpenAI says Morgan Stanley’s AI solutions give financial advisors faster insights, more informed decisions, and efficient summarization tools. | 中 | SU008 |
| CU033 | Lowe’s says Mylow was developed with OpenAI and provides real-time home-improvement answers, project guidance, and product discovery. | 高 | SU009, SU011 |
| CU034 | Lowe’s says Mylow is available on desktop and mobile web for MyLowe’s Rewards members. | 中 | SU009 |
| CU035 | Lowe’s says Mylow Companion launched to associates across more than 1,700 stores. | 中 | SU010 |
| CU036 | Lowe’s says Mylow Companion gives associates fast access to product details, project advice, and inventory information on sales-floor devices. | 中 | SU010 |
| CU037 | OpenAI says Lowe’s brought Mylow Companion to all retail associates using the same AI foundation as the customer-facing Mylow advisor. | 中 | SU012 |
| CU038 | OpenAI’s BBVA case study says employees saved 3 hours per employee per week. | 中 | SU013 |
| CU039 | OpenAI’s BBVA case study says weekly active usage reached 83%. | 中 | SU013 |
| CU040 | OpenAI’s BBVA case study says employees created more than 20,000 custom GPTs and used around 4,000 frequently. | 中 | SU013 |
| CU041 | BBVA said it formed a strategic alliance with OpenAI to redefine banking with artificial intelligence. | 中 | SU014 |
| CU042 | Klarna said its OpenAI-powered assistant handled 2.3 million conversations, equal to two-thirds of Klarna’s customer service chats, in its first month. | 中 | SU016 |
| CU043 | Klarna said its AI assistant was doing the equivalent work of 700 full-time agents. | 高 | SU015, SU016 |
| CU044 | Klarna said its AI assistant led to a 25% drop in repeat inquiries. | 中 | SU016 |
| CU045 | Klarna said average errand-resolution time fell to less than 2 minutes from 11 minutes previously. | 中 | SU016 |
| CU046 | Intercom says thousands of customers using Fin see an average conversation resolution rate of 41%, with some reaching 50%. | 中 | SU018 |
| CU047 | OpenAI says Intercom built a scalable AI platform that ships new capabilities in days rather than quarters. | 中 | SU017 |
| CU048 | OpenAI’s 2025 enterprise report says Intercom’s Fin resolves millions of customer queries each month. | 中 | SU003 |
| CU049 | OpenAI’s 2025 enterprise report says Fin Voice latency decreased 48% since March. | 中 | SU003 |
| CU050 | OpenAI’s 2025 enterprise report says Fin Voice resolves 53% of calls end-to-end on average. | 中 | SU003 |
| CU051 | OpenAI’s 2025 enterprise report says calls that still need human agents are resolved 40% faster after Fin Voice completes the initial steps. | 中 | SU003 |
| CU052 | OpenAI’s Mercado Libre case says 17,000 developers can build on Verdi across more than 30,000 microservices. | 中 | SU019 |
| CU053 | Accenture said it would equip tens of thousands of professionals with ChatGPT Enterprise. | 中 | SU020 |
| CU054 | Accenture said its OpenAI flagship AI program targets workflows in financial services, healthcare, the public sector, and retail. | 中 | SU020 |
| CU055 | Microsoft’s customer story says Worten saves 11,000 hours annually with Azure OpenAI. | 中 | SE020 |
| CU056 | Microsoft’s customer story describes Discovery Bank’s deployment as redefining personal banking with Azure OpenAI. | 中 | SE021 |
| CR001 | OpenAI said on 2025-06-05 that it was responding to The New York Times’ data demands in order to protect user privacy. | 中 | SO024 |
| CR002 | OpenAI said the earlier legal order to retain consumer ChatGPT and API content indefinitely ended on 2025-09-26. | 中 | SO024 |
| CR003 | OpenAI said that after the order ended, deleted ChatGPT conversations, Temporary Chats, and API data returned to standard automatic deletion practices. | 中 | SO024 |
| CR004 | OpenAI’s New York Times litigation page said that, as the case progressed, the court rejected several of the Times’ claims. | 中 | SO018 |
| CR005 | CourtListener lists The New York Times Company v. Microsoft Corporation under docket 1:23-cv-11195. | 中 | SR001 |
| CR006 | CourtListener lists Musk v. Altman under docket 4:24-cv-04722. | 中 | SO021 |
| CR007 | CourtListener shows Elon Musk filed a complaint on 2024-08-05 against Altman, Brockman, and multiple OpenAI entities. | 中 | SO021 |
| CR008 | OpenAI’s March 2024 review materials said the 2023 board crisis was precipitated by a breakdown in trust between the prior board and Sam Altman. | 中 | SO010 |
| CR009 | OpenAI’s March 2024 review materials said the prior board action did not arise from product safety or security, the pace of development, OpenAI finances, or statements to investors, customers, or partners. | 中 | SO010 |
| CR010 | OpenAI’s March 2024 board announcement said Sue Desmond-Hellmann, Nicole Seligman, and Fidji Simo joined the board and Sam Altman rejoined it. | 中 | SO011 |
| CR011 | OpenAI’s structure page says the OpenAI Foundation governs and controls OpenAI Group PBC. | 中 | SO002 |
| CR012 | The European Data Protection Board published a Report of the work undertaken by the ChatGPT Taskforce on 2024-05-24. | 中 | SR009 |
| CR013 | The European Commission said in April 2025 that the AI Continent Action Plan includes an AI Act Service Desk to support implementation of the AI Act across the EU. | 中 | SR008 |
| CR014 | OpenAI’s Data Processing Addendum says it is effective January 1, 2026 and supplements the agreement governing use of the services. | 中 | SR002 |
| CR015 | OpenAI’s Services Agreement says it applies to APIs, ChatGPT Enterprise, ChatGPT Business, ChatGPT for Clinicians, and other business or developer services, rather than consumer use. | 中 | SR003 |
| CR016 | OpenAI’s status history shows recurring incidents across ChatGPT, the API, login, file uploads, and model-specific services across 2025 and 2026. | 中 | SO012 |
| CR017 | On 2026-05-04, OpenAI’s status page showed an incident in which some ChatGPT workspace-connector write actions were automatically disabled. | 中 | SO012 |
| CR018 | OpenAI’s status page reported aggregate API uptime of 99.99% for the February-to-May 2026 period. | 中 | SO012 |
| CR019 | OpenAI’s status page reported aggregate uptime of 99.83% for ChatGPT, 99.98% for Codex, and 99.93% for FedRAMP over the February-to-May 2026 period. | 中 | SO012 |
| CR020 | SecurityWeek reported in late 2025 that researchers hacked ChatGPT memories and web-search features. | 中 | SE025 |
| CR021 | OECD.AI recorded that OpenAI issued an urgent security update for Mac apps after a supply-chain attack on 2026-04-11. | 中 | SR007 |
| CR022 | OpenAI published an incident-response post titled Our response to the Axios developer tool compromise. | 中 | SR004 |
| CR023 | OpenAI’s Help Center publishes a process for reporting security vulnerabilities. | 中 | SR005 |
| CR024 | OpenAI operates a public bug bounty program through Bugcrowd. | 中 | SR006 |
| CR025 | OpenAI’s security and privacy page says its platforms are built with security and privacy in mind and tested by a team of security experts. | 中 | SE002 |
| CR026 | OpenAI and Microsoft announced a multiyear partnership extension on 2023-01-23. | 中 | SO005 |
| CR027 | OpenAI’s January 2023 partnership materials said Azure remained OpenAI’s exclusive cloud provider across research, API, and product workloads. | 中 | SO005 |
| CR028 | OpenAI announced that Stargate aims to invest $500 billion over four years in AI infrastructure. | 中 | SI005 |
| CR029 | OpenAI said Stargate would begin deploying $100 billion immediately. | 中 | SI005 |
| CR030 | CoreWeave said its OpenAI infrastructure agreement had a contract value of up to $11.9 billion. | 中 | SI008 |
| CR031 | CoreWeave said OpenAI would invest $350 million of CoreWeave stock as part of the infrastructure deal. | 中 | SI008, SI009 |
| CR032 | Oracle announced a 2026 financing plan to raise $45 billion to $50 billion of gross cash proceeds to build additional OCI capacity for customers including OpenAI. | 中 | SI011 |
| CR033 | Accenture said OpenAI would be one of Accenture’s primary AI partners for its next generation of AI-powered services. | 中 | SU020 |
| CR034 | ChartMogul says some AI contracts have three-month opt-out provisions where 70% to 80% of customers opt out. | 中 | SU005 |
| CR035 | ChartMogul argues that experimental AI revenue, short opt-out contracts, and usage-heavy pricing can make ARR look more durable than it is. | 中 | SU005 |
| CR036 | OpenAI’s 2025 enterprise report says average reasoning token consumption per organization increased approximately 320x over the prior 12 months. | 中 | SU003 |
| CR037 | OpenAI announced Fidji Simo as CEO of Applications in 2025 and said she would report directly to Sam Altman. | 中 | SO023 |
| CR038 | OpenAI’s structure page says the OpenAI Foundation appoints all members of the OpenAI Group board and can replace directors. | 中 | SO002 |
| CR039 | OpenAI’s current board includes independent directors added after the 2023 crisis, but the structure still keeps mission governance tightly bound to the Foundation and a small leadership group. | 中 | SO010, SO011, SO002 |
| CR040 | OpenAI says business customers own and control their business data. | 中 | SM004 |
| CR041 | OpenAI says it does not train models on business data by default. | 中 | SM004 |
| CR042 | OpenAI says customers control how long data is retained for ChatGPT Enterprise, ChatGPT for Healthcare, and ChatGPT Edu. | 中 | SM004 |
| CR043 | Oracle said in March 2025 that it had signed cloud agreements with OpenAI and expected to sign its first Stargate contract in the near term. | 中 | SI010 |
| CR044 | Oracle said in March 2025 that it was on schedule to double its data-center capacity during calendar 2025. | 中 | SI010 |
| CR045 | OpenAI’s public pages provide privacy controls, security reporting routes, and governance updates, but they do not publish the full contract, SLA, or committed-spend schedule an investor would need for underwriting. | 中 | SR002, SR003, SM004, SE002, SR005, SR006, SI011 |
| CV001 | CNBC reported that OpenAI closed a $122 billion funding round at an $852 billion valuation on 2026-03-31. | 中 | SO014 |
| CV002 | Reuters framed OpenAI's immediate post-round challenge as finding focus at an $852 billion valuation. | 中 | SV001 |
| CV003 | CNBC reported in April 2026 that OpenAI capped revenue-share payments to Microsoft. | 中 | SI007 |
| CV004 | MarketScreener reported that OpenAI topped $25 billion in annualized revenue at the end of February 2026. | 中 | SV003 |
| CV005 | The same MarketScreener/Reuters report said OpenAI generated $21.4 billion in annualized revenue at the end of 2025. | 中 | SV003 |
| CV006 | MarketScreener reported that OpenAI went from effectively zero revenue in late 2022 to more than $20 billion in annualized revenue in 2025. | 中 | SV003 |
| CV007 | MarketScreener reported that OpenAI is targeting roughly $600 billion in total compute spending through 2030. | 中 | SV003 |
| CV008 | Anthropic said it raised $30 billion in Series G funding at a $380 billion post-money valuation in February 2026. | 中 | SV004 |
| CV009 | CNBC reported that Mistral was valued at $14 billion in September 2025. | 中 | SP019 |
| CV010 | CompaniesMarketCap listed Microsoft's market capitalization at $3.076 trillion as of May 2026. | 中 | SV005 |
| CV011 | CompaniesMarketCap listed Microsoft's trailing-twelve-month revenue at $305.45 billion. | 中 | SV006 |
| CV012 | Microsoft screens at roughly 10.1x market cap to trailing revenue using the May 2026 CompaniesMarketCap figures. | 中 | SV005, SV006 |
| CV013 | CompaniesMarketCap listed Oracle's market capitalization at $494.19 billion as of May 2026. | 中 | SV007 |
| CV014 | CompaniesMarketCap listed Oracle's trailing-twelve-month revenue at $64.07 billion. | 中 | SV008 |
| CV015 | Oracle screens at roughly 7.7x market cap to trailing revenue using the May 2026 CompaniesMarketCap figures. | 中 | SV007, SV008 |
| CV016 | CompaniesMarketCap listed Palantir's market capitalization at $345.35 billion as of May 2026. | 中 | SV009 |
| CV017 | CompaniesMarketCap listed Palantir's trailing-twelve-month revenue at $4.47 billion. | 中 | SV010 |
| CV018 | Palantir screens at roughly 77.3x market cap to trailing revenue using the May 2026 CompaniesMarketCap figures. | 中 | SV009, SV010 |
| CV019 | OpenAI's $852 billion financing mark implies roughly 34.1x annualized revenue on the reported $25 billion end-February 2026 run-rate. | 中 | SO014, SV003 |
| CV020 | OpenAI's $852 billion financing mark implies roughly 39.8x annualized revenue on the reported $21.4 billion end-2025 run-rate. | 中 | SO014, SV003 |
| CV021 | Anthropic's $380 billion valuation and roughly $9 billion annualized revenue imply an approximate 42.2x run-rate multiple. | 中 | SV004, SV003 |
| CV022 | At a 25x revenue multiple, OpenAI would need roughly $34.1 billion of revenue support to justify an $852 billion valuation. | 中 | SO014, SV003 |
| CV023 | At an 18x revenue multiple, OpenAI would need roughly $47.3 billion of revenue support to justify an $852 billion valuation. | 中 | SO014, SV003 |
| CV024 | At a 12x revenue multiple, OpenAI would need roughly $71.0 billion of revenue support to justify an $852 billion valuation. | 中 | SO014, SV003 |
| CV025 | OpenAI said more than 1 million business customers were directly using OpenAI products in November 2025. | 中 | SU002 |
| CV026 | OpenAI said ChatGPT served more than 800 million users every week in December 2025. | 中 | SI004 |
| CV027 | OpenAI said Stargate intends to invest $500 billion over four years and begin deploying $100 billion immediately. | 中 | SI005 |
| CV028 | Oracle said it expected to raise $45 billion to $50 billion of gross cash proceeds during calendar 2026 to expand OCI capacity for customers including OpenAI. | 中 | SI011 |
| CV029 | CoreWeave said its OpenAI infrastructure deal had contract value up to $11.9 billion and included a $350 million OpenAI equity investment. | 中 | SI008 |
| CV030 | Microsoft investor relations publicly provides SEC filings and financial statements. | 中 | SV011 |
| CV031 | Oracle investor relations publicly provides annual, quarterly, proxy, and Section 16 filings. | 中 | SV012 |
| CV032 | Alphabet investor relations publicly provides a route to SEC filings for investors. | 中 | SV013 |
| CV033 | OpenAI does not provide public-company-style filing disclosure comparable to Microsoft, Oracle, or Alphabet because it remains a foundation-controlled private company. | 中 | SO002, SV011, SV012, SV013 |
| CV034 | Reuters said the OpenAI-versus-Anthropic revenue race has implications for eventual IPOs. | 中 | SV002 |
| CV035 | CNBC tied OpenAI's March 2026 round to rising anticipation for an eventual IPO. | 中 | SO014 |
| CV036 | The current public record supports a research-more recommendation rather than a buy call at $852 billion because demand is strong but economics and financing terms are still under-disclosed. | 中 | SO014, SV003, SO002, SV011, SV012 |
| CV037 | OpenAI's current implied run-rate multiple sits well above Microsoft and Oracle but below Palantir's public AI-premium multiple. | 中 | SO014, SV003, SV005, SV006, SV007, SV008, SV009, SV010 |
| CV038 | OpenAI's current implied run-rate multiple is closer to Anthropic's private-market benchmark than to Microsoft or Oracle public-comp benchmarks. | 中 | SO014, SV003, SV004, SV005, SV006, SV007, SV008 |
| CV039 | If public markets ultimately value OpenAI closer to Microsoft- or Oracle-like platform multiples than to frontier or AI-premium scarcity multiples, the current entry price embeds substantial downside. | 中 | SO014, SV003, SV005, SV006, SV007, SV008, SV009, SV010 |
| CV040 | The clearest public downside transmission path runs through compute intensity and partner economics because the Microsoft reset, Stargate ambition, Oracle financing plan, and CoreWeave contract all directly affect what portion of revenue OpenAI can ultimately keep. | 中 | SI007, SI005, SI008, SI011 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | OpenAI | About | OpenAI | OpenAI is an AI research and deployment company and says its mission is to ensure AGI benefits all of humanity. |
| SO002 | OpenAI | Our structure | OpenAI | The OpenAI Foundation continues to control OpenAI Group and holds a 26% equity stake, while Microsoft holds roughly 27%. |
| SO003 | OpenAI | OpenAI Charter | OpenAI's mission is to ensure AGI benefits all of humanity and its primary fiduciary duty is to humanity. |
| SO004 | OpenAI Foundation | OpenAI Foundation | The OpenAI Foundation says it made an initial $50 million commitment in 2025 to nonprofits and mission-focused organizations. |
| SO005 | OpenAI | OpenAI and Microsoft extend partnership | The post describes a multi-year, multi-billion dollar Microsoft investment and says Azure remains the exclusive cloud provider. |
| SO006 | Microsoft | Microsoft and OpenAI extend partnership | Microsoft announced the third phase of its long-term partnership with OpenAI through a multiyear, multibillion dollar investment. |
| SO007 | OpenAI | Introducing ChatGPT | OpenAI introduced ChatGPT as a research preview on November 30, 2022. |
| SO008 | OpenAI | GPT-4 | OpenAI | The GPT-4 research page is dated March 14, 2023 and marks the model as the latest milestone in OpenAI's scaling effort. |
| SO009 | OpenAI | OpenAI leadership team update | Greg Brockman became President and Mira Murati became CTO in the May 5, 2022 leadership update. |
| SO010 | OpenAI | Review completed & Altman, Brockman to continue to lead OpenAI | WilmerHale found the prior board's action arose from a breakdown in trust, not product safety, finances, or statements to investors. |
| SO011 | OpenAI | OpenAI announces new members to board of directors | Sue Desmond-Hellmann, Nicole Seligman, and Fidji Simo joined the board and Sam Altman rejoined it. |
| SO012 | OpenAI | OpenAI Status history | The status history lists repeated ChatGPT, API, login, and model-specific incidents across 2025 and 2026. |
| SO013 | TechCrunch | OpenAI raises $6.6B and is now valued at $157B | TechCrunch reported a $6.6 billion round at a $157 billion post-money valuation and said total raised reached $17.9 billion. |
| SO014 | CNBC | OpenAI closes funding round at an $852 billion valuation | CNBC reported OpenAI closed a $122 billion round at an $852 billion valuation. |
| SO015 | Bloomberg | OpenAI Valued at $852 Billion After Backing From Amazon, Nvidia, SoftBank | Bloomberg reported Amazon, Nvidia, and SoftBank were the largest participants in the $122 billion round. |
| SO016 | Axios | OpenAI reorg sets up $500 billion for-profit AI behemoth | Axios described the 2025 reorganization as a controversial split with the nonprofit controlling OpenAI Group PBC. |
| SO017 | Forbes | OpenAI | Company Overview & News | Forbes listed OpenAI at 6,400 employees, headquartered in San Francisco, with 2025 revenue of $3.7 billion. |
| SO018 | OpenAI | Reporting the facts about the New York Times' lawsuit | OpenAI says it learned of the New York Times lawsuit on December 27, 2023 and continues to dispute the claims. |
| SO019 | JD Supra / Nelson Mullins | From Copyright Case to AI Data Crisis: How The New York Times v. OpenAI Reshapes Companies' Data Governance and eDiscovery Strategy | The analysis says a May 13, 2025 order required OpenAI to retain ChatGPT conversation logs affecting over 400 million users worldwide. |
| SO020 | Loeb & Loeb | In Re: OpenAI Inc., Copyright Infringement Litigation | Loeb summarizes OpenAI copyright disputes and notes links between Raw Story/AlterNet claims and New York Times litigation issues. |
| SO021 | CourtListener | Musk v. Altman, 4:24-cv-04722 | CourtListener shows Elon Musk filed a complaint on August 5, 2024 against Altman, Brockman, and multiple OpenAI entities. |
| SO022 | OpenAI | Adebayo Ogunlesi joins OpenAI's Board of Directors | OpenAI announced Adebayo Ogunlesi joined the board on January 14, 2025. |
| SO023 | OpenAI | OpenAI Expands Leadership with Fidji Simo | Fidji Simo was named CEO of Applications, reporting directly to Sam Altman. |
| SO024 | OpenAI | How we're responding to The New York Times' data demands in order to protect user privacy | OpenAI said it would securely store limited April-September 2025 data under legal hold while continuing to challenge the order. |
| SO025 | OpenAI | Built to benefit everyone | OpenAI says the Foundation holds equity currently valued at about $130 billion and that the for-profit business was established in 2019. |
| SO026 | OpenAI | ChatGPT for enterprise | ChatGPT for enterprise. |
| SO027 | OpenAI | ChatGPT Education | ChatGPT Education. |
| SO028 | Microsoft WorkLab | Work Trend Index: Microsoft’s latest research on the ways we work | 31,000 people. 31 countries. Trillions of productivity signals. |
| SO029 | Deloitte | State of Generative AI in the Enterprise | Deloitte Global | Every organisation is setting its own course to AI scaling. |
| SO030 | Stanford HAI | AI Index | Stanford HAI | The mission of the AI Index is to provide unbiased, rigorously vetted, and globally sourced data. |
| SO031 | Stack Overflow | 2024 Stack Overflow Developer Survey | In May 2024, over 65,000 developers responded to our annual survey. |
| SO032 | Anthropic | The Anthropic Economic Index | Understanding AI’s effects on the economy. |
| SO033 | Anthropic | Plans & Pricing | |
| SO034 | Anthropic Privacy Center | Home | Anthropic Privacy Center | Commercial Customers — API, Console, Team & Enterprise plans. |
| SO035 | Anthropic | Responsible Disclosure Policy | The security of our systems and user data is Anthropic’s top priority. |
| SO036 | Anthropic | Anthropic Trust Center | Anthropic Trust Center. |
| SO037 | Gemini Developer API pricing | Gemini Developer API pricing. | |
| SO038 | Mistral Docs | Models Overview | Models — Overview. |
| SO039 | DatacenterDynamics | OpenAI announces “The Stargate Project:” $500bn over four years on AI infrastructure | DatacenterDynamics tracked Stargate as a $500 billion AI infrastructure buildout around OpenAI demand. |
| SO040 | TechCrunch | OpenAI's GPT-5 is here | TechCrunch | OpenAI's GPT-5 is here. |
| SO041 | Forbes | Moving Beyond ChatGPT: OpenAI's New Revenue Model | Moving Beyond ChatGPT: OpenAI's New Revenue Model. |
| SO042 | OpenAI | BBVA puts AI in the hands of every team with OpenAI | OpenAI | In just 5 months, BBVA has widely adopted ChatGPT Enterprise. |
| SO043 | OpenAI | Stargate advances with 4.5 GW partnership with Oracle | OpenAI | Stargate advances with 4.5 GW partnership with Oracle. |
| SO044 | CNBC | Anthropic in talks with investors to raise funds at $900 billion valuation, higher than OpenAI | Anthropic in talks with investors to raise funds at $900 billion valuation, higher than OpenAI. |
| SO045 | SEC | EDGAR Entity Landing Page - Microsoft | EDGAR Entity Landing Page. |
| SO046 | SEC | EDGAR Entity Landing Page - Oracle | EDGAR Entity Landing Page. |
| SM001 | OpenAI | AI Platforms to Accelerate your Business | OpenAI | Create, code, and innovate with OpenAI's tools and APIs. |
| SM002 | OpenAI | ChatGPT Pricing | OpenAI | A plan for development-focused teams with pay-as-you-go pricing. |
| SM003 | OpenAI | OpenAI API Pricing | OpenAI | Standard Batch -50% Data residency +10%. |
| SM004 | OpenAI | Enterprise privacy at OpenAI | We do not train our models on your data by default. |
| SM005 | OpenAI | Introducing ChatGPT Edu | ChatGPT built for universities to responsibly deploy AI to students, faculty, researchers, and campus operations. |
| SM006 | Microsoft | The 2025 Annual Work Trend Index: The Frontier Firm is born | 82% of leaders say this is a pivotal year to rethink core aspects of strategy and operations. |
| SM007 | Menlo Ventures | 2024: The State of Generative AI in the Enterprise | AI spending surged to $13.8 billion this year, more than 6x the $2.3 billion spent in 2023. |
| SM008 | Deloitte | State of Generative AI Q4 – Press Release | Deloitte US | Regulatory compliance has emerged as the top barrier holding organizations back. |
| SM009 | Gartner | Gartner Forecasts Worldwide GenAI Spending to Reach $644 Billion in 2025 | Worldwide GenAI Spending to Reach $644 Billion in 2025. |
| SM010 | VentureBeat | Gartner forecasts gen AI spending to hit $644B in 2025: What it means for enterprise IT leaders | Global gen AI spending will hit $644 billion in 2025, a 76.4% year-over-year increase. |
| SM011 | McKinsey | Economic potential of generative AI | Generative AI could add the equivalent of $2.6 trillion to $4.4 trillion annually. |
| SM012 | PwC | PwC 2025 Global AI Jobs Barometer | AI-skilled workers see average 56% wage premium in 2024. |
| SM013 | OECD.AI | The OECD Artificial Intelligence Policy Observatory | Managing the risks and benefits of generative AI. |
| SM014 | OECD.AI | Risk & Accountability Overview | AI has risks and all actors must be accountable. |
| SM015 | European Commission | AI Act | Shaping Europe’s digital future | The AI Act is the first-ever comprehensive legal framework on AI worldwide. |
| SM016 | NIST | AI Risk Management Framework | AI Risk Management Framework. |
| SM017 | IDC | IDC Spending Guide summary (article unavailable) | Article Not Found. |
| SP001 | OpenAI Help Center | What is ChatGPT Business? | We're introducing a new seat type: a Codex seat that offers codex-only access, based on flexible pricing. |
| SP002 | OpenAI Help Center | Managing billing and seats in ChatGPT Business | As of April 2, 2026, we've changed the pricing structure for ChatGPT Business and have reduced the price of standard ChatGPT seats by USD $5 per month. |
| SP003 | Anthropic | Models overview - Claude API Docs | Models overview ... Amazon Bedrock ... Google Cloud's Vertex AI. |
| SP004 | Google Workspace | Compare Flexible Pricing Plan Options | Enterprise data regions ... Enterprise endpoint management ... AI Classification for Google Drive. |
| SP005 | Google Cloud | Cloud compliance and regulations resources | Google Cloud’s industry-leading certifications, documentation, and third-party audits to help support your compliance. |
| SP006 | Google Cloud | Agent Platform Pricing | Get enterprise-ready AI. Build on the same infrastructure as Google. |
| SP007 | Microsoft | Microsoft 365 Copilot for Business: Enterprise AI Solutions | AI chat ... Agents for work ... Build your own agents ... Copilot in your apps. |
| SP008 | Microsoft Azure | Azure OpenAI in Foundry Models | Azure OpenAI in Foundry Models. |
| SP009 | Microsoft Learn | Data, privacy, and security for Azure Direct Models in Microsoft Foundry | Data, privacy, and security for Azure Direct Models in Microsoft Foundry. |
| SP010 | AWS | Amazon Bedrock Pricing | Amazon Bedrock offers select foundation models from leading AI providers like Anthropic, Meta, Mistral AI, and Amazon for batch inference at a 50% lower price compared to on-demand inference pricing. |
| SP011 | AWS | AI Assistant - Amazon Q Pricing | The Amazon Q Business Lite subscription of $3 per user/month ... The Amazon Q Business Pro subscription of $20 per user/month. |
| SP012 | AWS | Amazon Bedrock Guardrails | Block up to 88% of harmful content and deliver auditable, mathematically verifiable explanations for validation decisions with 99% accuracy. |
| SP013 | Llama | Unmatched Performance and Efficiency | Llama 4 | Our models are optimized for easy deployment, cost efficiency, and performance that scales to billions of users. |
| SP014 | Mistral AI | Mistral AI Studio - your AI production platform | Create AI use cases ... all with enterprise privacy, security, and full ownership of your data. |
| SP015 | Mistral Docs | Rate limits and usage tiers | Rate limits are set at the Organization level, meaning they apply across all Workspaces within your organization. |
| SP016 | Menlo Ventures | 2025: The State of Generative AI in the Enterprise | We estimate Anthropic now earns 40% of enterprise LLM spend ... OpenAI ... 27% ... Google ... 21%. |
| SP017 | TechCrunch | Enterprises prefer Anthropic's AI models over anyone else's, including OpenAI's | Anthropic now holds 32% of the enterprise large language model market share by usage ... OpenAI ... 25%. |
| SP018 | CNBC | Amazon-backed AI firm Anthropic valued at $61.5 billion after latest round | Amazon-backed AI firm Anthropic valued at $61.5 billion after latest round. |
| SP019 | CNBC | AI firm Mistral valued at $14 billion as chip giant ASML takes major stake | AI firm Mistral valued at $14 billion as chip giant ASML takes major stake. |
| SI001 | OpenAI | Introducing ChatGPT Enterprise | We’re launching ChatGPT Enterprise, which offers enterprise-grade security and privacy. |
| SI002 | OpenAI | Introducing ChatGPT Team | August 29, 2025 update: ChatGPT Team is now called ChatGPT Business. |
| SI003 | OpenAI | Introducing ChatGPT Pro | Today, we’re adding ChatGPT Pro, a $200 monthly plan. |
| SI004 | OpenAI | The state of enterprise AI | ChatGPT now serves more than 800 million users every week. |
| SI005 | OpenAI | Announcing The Stargate Project | The Stargate Project ... intends to invest $500 billion over the next four years ... We will begin deploying $100 billion immediately. |
| SI006 | OpenAI Help Center | What is ChatGPT Plus? | ChatGPT Plus is a subscription plan ... for $20/month. |
| SI007 | CNBC | OpenAI shakes up partnership with Microsoft, capping revenue share payments | CNBC reported that OpenAI capped revenue-share payments to Microsoft in a 2026 partnership reset. |
| SI008 | CoreWeave | CoreWeave Announces Agreement with OpenAI to Deliver AI Infrastructure | The contract value for this strategic deal is up to $11.9 billion. As part of this deal, OpenAI will become an investor in CoreWeave through the issuance of $350.0 million of CoreWeave stock. |
| SI009 | SEC / CoreWeave | Form S-1 for CoreWeave, Inc. | As filed with the U.S. Securities and Exchange Commission on March 3, 2025. |
| SI010 | Oracle | Oracle Announces Fiscal 2025 Third Quarter Financial Results | Q3 Remaining Performance Obligations $130 billion ... We have now signed cloud agreements with ... OpenAI. |
| SI011 | Oracle | Oracle Announces Equity and Debt Financing Plan for Calendar Year 2026 | Oracle expects to raise $45 to $50 billion of gross cash proceeds during the 2026 calendar year. |
| SI012 | Oracle | Oracle Announces Fiscal Year 2026 Third Quarter Financial Results | Most of the increase in RPO in Q3 related to large scale AI contracts ... funded upfront via customer prepayments or the customer buys the GPUs and supplies them to Oracle. |
| SI013 | TechCrunch | OpenAI boasts enterprise win days after internal “code red” on Google threat | The majority of OpenAI’s revenue still comes from consumer subscriptions ... OpenAI also has committed $1.4 trillion to infrastructure commitments over the next few years. |
| SE001 | OpenAI | API Platform | OpenAI | Build leading AI products on OpenAI’s platform. |
| SE002 | OpenAI | Security and privacy at OpenAI | OpenAI | You decide whether your data is used for training and model improvement. |
| SE003 | OpenAI | Business data privacy, security, and compliance | OpenAI | We don’t train our models on your organization’s data by default. |
| SE004 | OpenAI | Safety & responsibility | OpenAI | Teach, Test, Share. |
| SE005 | OpenAI | New tools and features in the Responses API | OpenAI | Today, we’re adding new built-in tools to the Responses API—our core API primitive for building agentic applications. |
| SE006 | OpenAI | Introducing Codex | OpenAI | A cloud-based software engineering agent that can work on many tasks in parallel, powered by codex-1. |
| SE007 | OpenAI | Introducing deep research | OpenAI | February 10, 2026 update: You can now connect deep research to any MCP or app and restrict web searches to trusted sites. |
| SE008 | OpenAI | Introducing GPT-4.1 in the API | OpenAI | A new series of GPT models featuring major improvements on coding, instruction following, and long context. |
| SE009 | OpenAI | OpenAI Status | We have identified an issue where write actions for some ChatGPT workspace connectors were automatically disabled. |
| SE010 | ChatGPT | ChatGPT Plans | Free, Go, Plus, Pro, Business, and Enterprise | ChatGPT Plans | Free, Go, Plus, Pro, Business, and Enterprise. |
| SE011 | OpenAI Help Center | ChatGPT Business: General FAQ | OpenAI Help Center | Starting April 2, 2026, ChatGPT Business supports two seat types: standard ChatGPT seats and Codex seats. |
| SE012 | OpenAI Help Center | What to know about the Sora discontinuation | OpenAI Help Center | The Sora web and app experiences were discontinued on April 26, 2026. The Sora API will be discontinued on September 24, 2026. |
| SE013 | OpenAI Developers | Responses Overview | OpenAI API Reference | Responses Overview | OpenAI API Reference. |
| SE014 | OpenAI Developers | All models | OpenAI API | All models | OpenAI API. |
| SE015 | OpenAI Developers | API Overview | OpenAI API Reference | API Overview | OpenAI API Reference. |
| SE016 | GitHub | GitHub - openai/openai-python: The official Python library for the OpenAI API | The official Python library for the OpenAI API. |
| SE017 | GitHub | GitHub - openai/openai-node: Official JavaScript / TypeScript library for the OpenAI API | Official JavaScript / TypeScript library for the OpenAI API. |
| SE018 | OpenAI | Customer stories | OpenAI | 1M businesses use OpenAI. |
| SE019 | NIST | Pre-Deployment Evaluation of OpenAI's o1 Model | NIST | Pre-Deployment Evaluation of OpenAI's o1 Model. |
| SE020 | Microsoft Customer Stories | Worten saves 11,000 hours annually with Microsoft Azure OpenAI, revolutionizing information search in stores | Worten saves 11,000 hours annually with Microsoft Azure OpenAI. |
| SE021 | Microsoft Customer Stories | Redefining personal banking with Discovery Bank and Azure OpenAI | Redefining personal banking with Discovery Bank and Azure OpenAI. |
| SE022 | TechCrunch | OpenAI launches new tools to help businesses build AI agents | TechCrunch | OpenAI released new tools designed to help developers and enterprise customers build AI agents. |
| SE023 | TechCrunch | OpenAI unveils a new ChatGPT agent for deep research | TechCrunch | OpenAI unveils a new ChatGPT agent for deep research. |
| SE024 | The Verge | OpenAI closed Sora applications shortly after launch. | The Verge | OpenAI closed Sora applications shortly after launch. |
| SE025 | SecurityWeek | Researchers Hack ChatGPT Memories and Web Search Features - SecurityWeek | Researchers Hack ChatGPT Memories and Web Search Features. |
| SE026 | arXiv | [2410.21276] GPT-4o System Card | Title: GPT-4o System Card. |
| SE027 | arXiv | [2412.16720] OpenAI o1 System Card | Title: OpenAI o1 System Card. |
| SU001 | TechCrunch | OpenAI's enterprise adoption appears to be accelerating, at the expense of rivals | TechCrunch | OpenAI’s enterprise adoption appears to be accelerating, at the expense of rivals. |
| SU002 | OpenAI | 1 million business customers: the fastest-growing business platform in history | OpenAI | Today, we’re announcing that more than 1 million business customers around the world are directly using OpenAI. |
| SU003 | OpenAI | The state of enterprise AI | 2025 Report | ChatGPT message volume grew 8x and API reasoning token consumption per organization increased 320x year-over-year. |
| SU004 | VentureBeat | OpenAI hits 3M business users and launches workplace tools to take on Microsoft | VentureBeat | OpenAI announced Wednesday that its business user base has surged 50% since February, reaching 3 million paying enterprise customers. |
| SU005 | ChartMogul | The SaaS Retention Report: The AI churn wave | ChartMogul | Contracts with a three month opt-out where 70-80% opt-out. |
| SU006 | High Alpha | 2025 SaaS Benchmarks Report by High Alpha | Beyond roughly $20 million in ARR, expansion becomes the dominant growth engine. |
| SU007 | Morgan Stanley | Launch of AI @ Morgan Stanley Debrief | Morgan Stanley | To date, 98% of Financial Advisor teams have adopted the Assistant. |
| SU008 | OpenAI | Morgan Stanley uses AI evals to shape the future of financial services | OpenAI | 98% adoption, increased engagement, and new services potential. |
| SU009 | Lowe’s | LOWE’S LAUNCHES FIRST AI-POWERED HOME IMPROVEMENT VIRTUAL ADVISOR | Developed in collaboration with OpenAI and leveraging Lowe’s expert advice, Mylow delivers the expertise of a trusted Lowe’s associate anytime and anywhere. |
| SU010 | Lowe’s | LOWE’S DEPLOYS FIRST AT-SCALE AI ASSISTANT FOR RETAIL ASSOCIATES | The launch to all associates across Lowe’s more than 1,700 stores marks the first time a retailer has successfully implemented this kind of technology at scale. |
| SU011 | Digital Commerce 360 | Lowe’s introduces its Mylow virtual adviser, built with OpenAI | Mylow is available to shoppers when they log in and offers AI-generated recommendations about processes, purchases and more. |
| SU012 | OpenAI | Lowe’s puts project expertise into every hand | OpenAI | With OpenAI, Lowe’s brings their Mylow Companion app to all retail associates, applying the same AI foundation behind their customer-facing Mylow virtual advisor. |
| SU013 | OpenAI | How BBVA is scaling AI from pilot to practice across the org | OpenAI | 83% weekly active usage. |
| SU014 | BBVA | BBVA and OpenAI Seal a Strategic Alliance to Redefine Banking with Artificial Intelligence | BBVA and OpenAI Seal a Strategic Alliance to Redefine Banking with Artificial Intelligence. |
| SU015 | OpenAI | Klarna’s AI assistant does the work of 700 full-time agents | OpenAI | Klarna is using AI to revolutionize personal shopping, customer service, and employee productivity. |
| SU016 | PR Newswire / Klarna | Klarna AI assistant handles two-thirds of customer service chats in its first month | The AI assistant has had 2.3 million conversations, two-thirds of Klarna’s customer service chats. |
| SU017 | OpenAI | Intercom’s three lessons for creating a sustainable AI advantage | OpenAI | Intercom created a scalable AI platform that ships new capabilities in days, not quarters. |
| SU018 | Intercom | Fin, the AI Agent for Customer Service, Keeps Getting Better | Thousands of Intercom customers are achieving incredible results with Fin, with an average conversation resolution rate of 41%. |
| SU019 | OpenAI | Introducing Verdi, an AI dev platform powered by GPT-4o | OpenAI | Mercado Libre introduces Verdi, an AI developer platform powered by GPT-4o. |
| SU020 | Accenture | OpenAI and Accenture Accelerate Enterprise Reinvention with Advanced AI | Accenture will equip tens of thousands of its professionals with ChatGPT Enterprise. |
| SR001 | CourtListener | The New York Times Company v. Microsoft Corporation, 1:23-cv-11195 – CourtListener.com | The New York Times Company v. Microsoft Corporation, 1:23-cv-11195. |
| SR002 | OpenAI | OpenAI Data Processing Addendum | OpenAI | Effective: January 1, 2026. |
| SR003 | OpenAI | Services agreement | OpenAI | This OpenAI Services Agreement only applies to use of OpenAI's APIs, ChatGPT Enterprise, ChatGPT Business, ChatGPT for Clinicians, and other services for customers who are businesses and developers. |
| SR004 | OpenAI | Our response to the Axios developer tool compromise | OpenAI | Our response to the Axios developer tool compromise. |
| SR005 | OpenAI Help Center | How to Report Security Vulnerabilities to OpenAI | How to Report Security Vulnerabilities to OpenAI. |
| SR006 | Bugcrowd | OpenAI bug bounty program | Bugcrowd | OpenAI. |
| SR007 | OECD.AI | OpenAI Issues Urgent Security Update for Mac Apps After Supply Chain Attack - OECD.AI | OpenAI Issues Urgent Security Update for Mac Apps After Supply Chain Attack. |
| SR008 | European Commission | European approach to artificial intelligence | Shaping Europe’s digital future | The plan includes actions to facilitate the implementation of the AI Act. |
| SR009 | European Data Protection Board | Report of the work undertaken by the ChatGPT Taskforce | European Data Protection Board | Report of the work undertaken by the ChatGPT Taskforce. |
| SV001 | Reuters | Artificial Intelligencer: OpenAI’s $852 billion problem: finding focus | Reuters framed the post-raise challenge as turning extraordinary scale into a focused business. |
| SV002 | Reuters | Anthropic may have closed the revenue gap on OpenAI. Here’s what it means for their IPOs | Reuters compared OpenAI and Anthropic through revenue momentum and IPO framing. |
| SV003 | MarketScreener | OpenAI tops $25 billion in annualized revenue, The Information reports | OpenAI topped $25 billion in annualized revenue as of the end of last month. |
| SV004 | Anthropic | Anthropic raises $30 billion in Series G funding at $380 billion post-money valuation | We have raised $30 billion in Series G funding ... valuing Anthropic at $380 billion post-money. |
| SV005 | CompaniesMarketCap | Microsoft (MSFT) - Market capitalization | As of May 2026 Microsoft has a market cap of $3.076 Trillion USD. |
| SV006 | CompaniesMarketCap | Microsoft (MSFT) - Revenue | Revenue in 2025 (TTM): $305.45 Billion USD. |
| SV007 | CompaniesMarketCap | Oracle (ORCL) - Market capitalization | As of May 2026 Oracle has a market cap of $494.19 Billion USD. |
| SV008 | CompaniesMarketCap | Oracle (ORCL) - Revenue | Revenue in 2026 (TTM): $64.07 Billion USD. |
| SV009 | CompaniesMarketCap | Palantir (PLTR) - Market capitalization | As of May 2026 Palantir has a market cap of $345.35 Billion USD. |
| SV010 | CompaniesMarketCap | Palantir (PLTR) - Revenue | Revenue in 2025 (TTM): $4.47 Billion USD. |
| SV011 | Microsoft | Microsoft Investor Relations - SEC Filings | Microsoft Investor Relations - SEC Filings. |
| SV012 | Oracle | Oracle - Investor Relations - SEC Filings | Oracle - Investor Relations - SEC Filings. |
| SV013 | Alphabet | Alphabet Investor Relations - Investors | Alphabet Investor Relations - Investors. |