初创公司尽调
尽调报告 consumer / education Private (Series C) 2026-05-23

Mercor

AI 专家市场,具备基准测试软件上行空间 — 收入质量和信任改善前保持观察

Mercor 确实是前沿 AI 工作流里的真实 franchise,benchmark 和软件上行也可信;但 $10B 估值已经计入更干净的经济性、更好的多元化和更强信任度,公开证据还没证明这些,结论应是观察。

封面要素

最新估值 01
10000 $M [CO011]
已披露累计融资 02
~$483.6M [CO012]
Sep 2025 总额收入运行率 03
~$450M [CI011]
2026 公司声称年化收入 04
$1B+ [CI015]
承包商名单规模 05
30,000+ [CO023]
每日承包商付款 06
>$2M/day [CI016]

公司概况

Mercor 是一家高速增长的私人 AI 人力与评估公司,由 Brendan Foody、Adarsh Hiremath 和 Surya Midha 于 January 2023 创立。 公司从 AI 辅助招聘起步,后来扩展到为前沿模型训练、评估和企业智能体项目提供领域专家、基准测试和工作流基础设施。 公开证据显示 Mercor 的客户包括 OpenAI、Anthropic 和 Meta;产品页面和文档则表明,APEX 基准、Enterprise AI、评估和 RL Studio 是公司摆脱纯市场交易量、向上游产品层爬升的最清晰信号。

官网
mercor.com
成立时间
2023-01-01
创始人
Brendan Foody, Adarsh Hiremath, Surya Midha
创立地点
San Francisco Bay Area, California, USA
总部
San Francisco, California, USA
产品
Mercor 把专家市场、AI 面试与匹配、承包商运营、APEX 等基准产品,以及面向 AI 实验室和企业智能体团队的新兴工作流软件组合在一起。
客户
前沿 AI 实验室和企业:它们需要领域专家、基准设计、评估工作流和人类反馈基础设施,用于模型训练和部署。
商业模式
Mercor 的收入似乎来自客户为专家工作支付的款项、匹配或寻源经济,以及日益产品化的基准和工作流服务;公开来源显示,表观收入按承包商付款前总额口径报告。
阶段
Private (Series C, October 2025)
融资情况
种子轮:$3.6M(2023);Series A:$30M,估值 $250M;Series B:$100M,估值 $2B;Series C:$350M,估值 $10B。 已披露新股融资总额约 $483.6M。
[CO010, CO011, CO012, CE021, CV040]

执行摘要

主要优势

  • 前沿 AI 客户相关性真实存在:公开报道点名 OpenAI、Anthropic 和 Meta,Mercor 称自己服务前五大 AI 实验室以及 Magnificent Seven 中的六家。
  • Mercor 增长异常快:2025 年 2 月估值 $2B,2025 年 10 月升至 $10B,期间公开来源称收入大幅提速。
  • APEX、Enterprise AI、测评和 RL Studio 共同证明,Mercor 正试图从劳动力聚合上移到 benchmark 与工作流基础设施。
  • 公司的承包商运营和付款栈看起来足够成熟,可以支撑全球专家工作流规模化。

主要风险

  • 收入似乎集中在少数 AI 实验室,公开留存数据也未披露。
  • Mercor 最显眼的收入数字似乎是扣除承包商付款前的总额,净收入、抽成率、利润率和现金指标仍未披露。
  • 2026 年数据泄露、Meta 工作暂停以及集体诉讼后果带来信任阴影,直接影响估值倍数。
  • Mercor 运营复杂的全球承包商体系,除执行负担外,还会带来劳动、隐私和合规风险。
  • 在 $10B 估值下,Mercor 定价远高于公开劳动力平台可比公司,必须获得实质性软件化重估才有吸引力。

未决问题

  • 总收入到净收入的桥、抽成率、毛利率以及任何经常性软件收入拆分未公开。
  • 前 10 大客户集中度、泄露后暂停账户细节以及产品层面留存未公开。
  • 数据泄露后信任修复和控制改进的独立证据仍然有限。
  • 承包商司法辖区组合、争议率,以及劳动或隐私风险的任何法律准备金未公开。
  • $10B 估值下的股权结构表条款、优先权堆叠和老股定价机制未公开。

目录

Chapter 01

01公司概况

1.1 身份定位与商业模式

Mercor 现在展示给市场的,不再像一家泛招聘初创公司,而更像为前沿 AI 系统供给专家的市场。 官网将公司定位为给 AI 经济组织人类智能的平台;专家页则揭示运营实况:Mercor 招募医生、律师、工程师、金融专业人士等专家,让他们以远程合同工形式推进 AI 系统。 产品承诺是速度和匹配度。Mercor 借助 AI 面试、匹配和工作流自动化判断谁适合哪个项目,再通过市场管理付款。 这个定位很关键,因为它同时解释了估值重估和投资者面对的风险画像。Mercor 不再只和招聘软件或人力派遣机构竞争。 它如今进入增长更快、争议也更大的后训练、评估和人在回路 AI 数据市场;这里客户预算大、切换成本低,数据权利问题也更尖锐。[CO001, CO002, CO003, CO015, CO016]

快照 KPI 表
指标数值 / 状态日期置信度缺口 / 备注
成立January 20232023-01官方介绍文章和 KTVU 报道支撑
最新阶段已盈利的 Series C 私营公司2026-05 抓取招聘页措辞;无审计报表
最新估值$10B2025-10Series C 公告与新闻报道相互印证
已披露新股融资总额$483.6M2025-10种子轮、Series A、Series B 和 Series C 合计
收入运行率~$450M 年化2025-09TechCrunch 报道的投资人沟通口径数字
2025 上半年利润$6M2025-H1TechCrunch 援引 Forbes 报道
承包商名册30,000+ 名专家2025-10官方 Series C 文章与 CNBC 相互印证
每日承包商付款>$1.5M / day2025-10官方 Series C 文章与 CNBC 相互印证

私营公司数据依赖公司公告和已报道的投资人材料;Mercor 不发布审计财务报表。

[CO004, CO010, CO011, CO012, CO021, CO022]
FO001: 公司快照逻辑

Mercor 把专家供给、AI 面试、匹配和项目执行转成 AI 实验室需要的模型改进产出。

[CO002, CO003, CO015, CO016, CO023, CO026]

1.2 创始人、领导层与治理

本次获取的公开来源一致将 Mercor 的创始三人组识别为 Brendan Foody、Adarsh Hiremath 和 Surya Midha,而不是二级摘要里有时反复出现的其他名字。 PR Newswire、KTVU 和 Forbes 都把这些创始人与 Bay Area 辩论网络、Harvard 和 Georgetown 辍学经历、Thiel Fellowship 联系起来。 创始人故事仍是 Mercor 品牌核心:一群极年轻的操盘者不到三年就从宿舍招聘软件转向 AI 实验室基础设施。 这个神话有利于融资,但也引出典型的关键人物和成熟度问题。Mercor 已开始在外围补强专业管理,引入前 Uber 高管 Sundeep Jain 担任首任总裁,并从 OpenAI 和 Scale 招揽资深人才。 不过,公司仍高度由创始人定义;除 Benchmark 和主要投资者选择性披露的董事会参与外,治理能见度有限。[CO005, CO006, CO007, CO031, CO032, CO033]

领导层与创始人表
人物职位背景创始人-市场匹配 / 覆盖面关键人物依赖
Brendan FoodyCEO 兼联合创始人Georgetown 辍学生;与联合创始人同为辩论队队友对外露面的经营者和主要战略叙事者关键
Adarsh HiremathCTO 兼联合创始人Harvard 辍学生;Forbes 报道的技术联合创始人主导核心技术与产品架构叙事
Surya Midha联合创始人;后任董事长Georgetown 辍学生;Forbes 资料称曾任 COO运营与治理连续性
Sundeep Jain总裁TechCrunch 称曾任 Uber 首席产品官在创始三人组之外补上资深高管厚度中等
Victor Lazarte董事会成员,BenchmarkPR Newswire 称在 Series A 时加入董事会投资人治理与融资支持中等

Mercor 治理公开披露很少;本表结合公司和媒体报道,并标注后续报道中出现差异的职位变化。

[CO005, CO006, CO007, CO031, CO032, CO033]
利益相关方 / 投资人地图
利益相关方角色控制权或经济重要性当前信号尽调问题
FelicisSeries B 和 Series C 领投方锚定最近两轮定价融资持续领投估值上调,直到 Series C优先股堆叠与按比例跟投细节
BenchmarkSeries A 支持方及董事席位早期机构治理影响力后续轮次仍持股董事会权利及任何否决条款
General Catalyst种子轮支持方和持续投资人跨轮次持续支持方从最早轮次支持公司预留资金策略与后续跟投能力
DST GlobalSeries B 参与方释放跨阶段成长投资兴趣加入 2025 年融资财团按投资人划分的持股集中度
Menlo VenturesSeries B 参与方补充 AI 市场网络与信号估值快速跃升后仍留在财团老股出售或流动性预期
Robinhood VenturesSeries C 新投资人拓展后期零售 / 消费网络以 $10B 标记进入资本之外的战略价值

投资人角色根据已披露轮次参与情况推断;董事会和优先权细节未公开。

[CO013, CO014]

1.3 融资、规模与运营版图

即便按 2025 AI 标准看,Mercor 的融资曲线也异常陡峭。 公司从 2023 的 $3.6 million 种子轮,走到 2024 估值 $250 million 的 Series A,再到 February 2025 的 $2 billion Series B 和 October 2025 的 $10 billion Series C。 由此形成的已披露新股融资约 $483.6 million。公开报道的业务牵引力也跟着资本一起上升。 CNBC 报道称,到 February 2025 公司已处理 300,000 名候选人、完成超过 100,000 场面试;TechCrunch 则在大致同一时间报道,Mercor 已评估 468,000 名申请人,年经常性收入(ARR)运行率为 $75 million。 到 September 2025,TechCrunch 称 Mercor 接近 $450 million 收入运行率,并在上半年实现 $6 million 利润。 本次获取的招聘数据还显示,公司一边声称已盈利,一边在工程、运营、财务和企业岗位上激进招聘,并把办公室扩展到 San Francisco、New York 和 London。[CO008, CO009, CO010, CO011, CO012, CO017]

运营足迹与人才网络表
维度证据日期重要性
主要办公室旧金山、纽约和伦敦2026-05 抓取显示 Mercor 已从单一湾区办公室扩张出去
早期用户基础种子轮融资前,在 25 个国家拥有 100,000 用户2023显示早期就聚合跨境劳动力
最大人才来源印度2025-02凸显地域和劳动力集中度
需求最高的专家群体软件工程、医学、法律和银行业2025-02释放向高技能领域专家供给转移的信号
观察到的开放职位582026-05 抓取表明在声称盈利的同时,公司仍在持续内部建设

Mercor 不发布审计员工数;运营足迹信号来自招聘信息、官方发布材料和管理层访谈。

[CO006, CO018, CO029, CO030]
FO002: 可投性快照

Mercor 同时具备出色增长证据,以及未解决的法律、安全和集中度风险。

[CO010, CO011, CO021, CO022, CO027, CO028]

1.4 里程碑与负面事件

Mercor 的加速伴随着可见摩擦。积极一面是,Meta 投资 Scale AI 后,想要中立供应商的大模型实验室开始不安,AI 数据供应链发生结构性转移,公司从中受益。 这给 Mercor 留出空间,把自己定位成面向后训练工作的高端专家市场。 但 Mercor 也遇到同一类风险:当公司大规模处理敏感工作流和承包商知识时,泄露、滥用和信任问题会随之而来。 Scale AI 于 September 2025 起诉 Mercor 和前 Scale 员工 Eugene Ling,指控其围绕客户战略文件侵占商业秘密。 法院记录显示,该案后来在 January 2026 被不可再诉地驳回,移除了一项诉讼阴影,但没有抹去围绕专家知识外泄的底层担忧。 对当前运营更严重的是,March 2026 与 LiteLLM 恶意软件相关的泄露事件触发客户审查;TechCrunch 报道称 Meta 暂停了合同,OpenAI 则调查自身暴露面。 这些事件合在一起,让信息安全成熟度和企业信任成为眼前尽调重点,而不是遥远的扩张问题。[CO023, CO024, CO026, CO027, CO028, CO034]

里程碑表
日期事件类型金额 / 状态参与方含义
2023-01Mercor 在宿舍中创立创立公司成立Foody、Hiremath、Midha起源故事绑定大学辍学创始人神话
2023种子轮融资融资$3.6MGeneral Catalyst 和天使投资人为最初的自动化招聘平台提供资金
2024-09Series A 融资融资$30M,估值 $250MBenchmark 领投财团搭起首个机构化董事会结构
2025-02Series B 融资融资$100M,估值 $2BFelicis 领投财团将市场注意力从招聘转向 AI 实验室需求
2025-03KTVU 关于超高速增长的访谈规模化$100M 收入运行率;极高盈利Brendan Foody公开把 Mercor 定位为增长最快的公司之一
2025-06Scale AI 中立性冲击合作CNBC 称 OpenAI 和 Google 从 Scale 撤回Meta 和 Scale AI为 Mercor 打开需求错位窗口
2025-09Scale AI 提起商业秘密诉讼反向已在 N.D. Cal. 提交诉状Scale AI 诉 Mercor.io抬高法律与信息治理风险
2025-09面向投资人的营销指标见报规模化~$450M 运行率;$6M 上半年利润Mercor 和潜在投资人建立 2025 年末经营杠杆叙事
2025-10Series C 融资融资$350M,估值 $10BFelicis、Benchmark、GC、Robinhood锁定 AI 服务领域最快的估值跃升之一
2026-04数据泄露后果公开浮现反向Meta 暂停据报发生;客户审查进行中Mercor、Meta、OpenAI安全成熟度升为董事会层面议题

这条时间线混合公司公告、访谈、市场报道和法律里程碑;日期取自已抓取来源中的事件日期。

[CO004, CO008, CO009, CO010, CO011, CO021]
Chapter 02

02市场分析

2.1 市场边界与待完成任务

Mercor 被描述成「HR 科技」或「数据标注」时最容易被误读。这两个标签都不完整。 公司当前市场位于高端劳动力市场、AI 后训练服务和评估基础设施的交汇处。 公司公开页面和 TechCrunch 报道显示,Mercor 向前沿 AI 实验室提供领域专家——医生、律师、工程师、银行家、顾问——处理排名、评估、表单、报告和基准任务等高度依赖判断的工作。 这意味着 Mercor 不再销售与申请人追踪系统、传统人力派遣公司或商品化众包市场相同的产品。 它的核心待完成任务,是足够快地把稀缺专家知识送进前沿模型改进回路,让实验室和企业愿意为此付费。 这个更窄的定义很重要:它大幅压低了相较自上而下 AI 支出标题的真实可服务市场,也更清楚地指出哪些竞争对手和风险真正重要。[CM001, CM002, CM003, CM004, CM005, CM025]

市场定义表
细分 / 类别纳入支出排除支出买方 / 付款方与 Mercor 的关系
专家型后训练服务领域专家 RLHF、评测、排序、红队测试、基准测试设计商品化微任务标注前沿实验室和企业 AI 团队当前核心市场
评测环境和基准测试任务设计、测试框架、隐藏测试集、工作流模拟与模型训练无关的通用软件 QA研究运营和模型评测负责人重要产品邻接
AI 人才市场层专家招募、筛选、匹配和薪酬管理传统长期岗位中介AI 实验室和企业Mercor 核心运营模式
招聘软件和人力配置 SaaS简历筛选和面试自动化后训练项目交付HR 和人才团队历史切入口,现在是次要业务
广义标注平台大规模图像 / 文本 / 音频标注高技能专业判断工作模型构建者和数据运营团队相邻但技能要求更低的细分市场

市场边界围绕投入前沿模型改进的高技能人类输入,而不是整个劳动力或 HR 科技市场。

[CM001, CM002, CM003, CM004, CM005, CM025]
FM001: 市场规模测算视角

Mercor 的实际市场,从广义 AI 支出收窄到规模小得多的专家后训练和评估细分市场。

[CM011, CM025, CM026, CM032, CM033, CM034]

2.2 机会规模测算

公开市场规模来源只能给出外边界。MarketsandMarkets 预计,到 2027 数据标注市场将达到 $3.6 billion,到 2029 AI 训练数据集市场将达到 $9.58 billion,但这两个类别都远大于 Mercor 的高端专家利基。 Stanford HAI 的 2025 AI Index 经 IBM 摘要后显示了更宽的需求背景:2024 企业 AI 投资达到 $252.3 billion,美国私人 AI 投资达到 $109.1 billion,新获融资的生成式 AI 初创公司数量接近三倍。 这些数字支持「买方预算正在快速形成」这一判断,但没有揭示实验室具体在专家承包商、基准创建或带有丰富工作流的评估环境上花多少钱。 实际结论是:Mercor 的总可用市场(TAM)小于泛 AI 投资或标注市场标题暗示的规模,但仍在更大的前沿模型支出和商业化浪潮中增长。 自上而下支出与自下而上可服务预算之间的错配,是这家公司最核心的尽调难题之一,也应让投资者对任何单张巨大 TAM 幻灯片保持怀疑。[CM006, CM007, CM008, CM009, CM010, CM011]

TAM / SAM / SOM 或规模测算视角表
视角发布方 / 年份地域数值增长方法论 / 局限
数据标注与标签市场MarketsandMarkets / 2023全球2027 年 $3.6B33.2% CAGR包含低技能工作的宽口径类别
AI 训练数据集市场MarketsandMarkets / 2024全球2029 年 $9.58B27.7% CAGR包含比 Mercor 更宽的软件和服务
企业 AI 投资池Stanford HAI 经 IBM / 2024全球$252.3B私募投资同比增长 44.5%口径太宽,不能作为 Mercor TAM
美国私募 AI 投资Stanford HAI 经 IBM / 2024美国$109.1Bn/a资本信号,不是专家支出信号
生成式 AI 初创公司成立Stanford HAI 经 IBM / 2024全球接近翻三倍n/a需求侧初创公司创建代理指标
Mercor 可服务市场本报告 / 2026全球小于标注 TAMn/a限定在高技能后训练和评测预算;公开数据不足以精确测算 SAM

公开第三方市场报告描述的是宽口径类别;本报告把它们作为外沿,并对 Mercor 更窄的可服务市场保留定性判断。

[CM006, CM007, CM008, CM009, CM010, CM011]
市场规模和采用尽调缺口表
缺口当前公开信息为什么不够对投资判断的影响具体尽调路径
专家后训练可服务市场(SAM)只有标注和数据集 TAM 的外框估算公开数据没有单独切出高端专家任务若机械套用自上而下 TAM,可能高估上行空间向买方索取按工作流和领域拆分的支出
买方集中度已公开具名实验室和企业意向没有客户级支出分布不能只凭客户标识推断收入耐久性索取客户集中度和续约数据
预算归属推断由研究和产品负责人掌握没有按职能拆分的公开采购图谱难以建模销售打法和周期长度访谈买方并收集组织架构图
经常性支出与项目支出一次性基准测试和评测需求可见没有公开复购率限制留存和 NRR 分析询问项目频率和队列留存
净支出与转付公开表述多用总市场和总收入口径劳务转付比例未知,会扭曲市场规模测算可能把总流水误读成净收入机会核对供应商抽佣率和承包者付款占比

本表明确划出公开市场数据的止点,以及一手尽调必须接上的起点。

[CM011, CM026, CM032, CM033, CM034]

2.3 买方、用户、付款方与采用路径

Mercor 类服务的买方地图异常不对称。买方通常是研究运营负责人、模型评估负责人或企业 AI 产品负责人。 用户可能是后训练团队或基准测试团队。付款方来自模型开发预算或企业转型预算。 供给侧专业人士既是劳动力投入,也是产品本身,因为公司货币化的是专家判断,而不只是软件席位。 这种结构解释了市场为什么能跑得很快:当模型质量瓶颈变得紧迫时,单个实验室可以迅速给出有意义的预算。 它也解释了市场为什么可能同样快速反转。少数前沿实验室权重过大;既有雇主可能不愿分享让专家有价值的工作流;企业买方在扩大项目之前仍会强制进行安全、IP 和信任审查。 因此,Mercor 受益于预算紧迫性,但在受监管领域仍会遇到采购拖累和拉长的价值验证周期。 这让客户背书质量和信任姿态成为异常重要的竞争武器。[CM012, CM013, CM014, CM015, CM016, CM026]

细分市场 / 买方地图
细分市场买方用户付款方工作流 / 预算所有者采用触发因素
前沿 AI 实验室研究或数据运营负责人后训练和评测团队模型开发预算需要专家判断或基准测试数据模型质量瓶颈
企业 AI 构建者产品 / AI 平台负责人内部 AI 团队企业转型预算需要领域数据,但不暴露完整语料部署到受监管工作流
基准测试创建者研究负责人评测工程师研发预算需要经济上真实的测试环境智能体可靠性担忧
专业专家独立承包商人类训练师 / 评测员Mercor 向供给侧付款远程变现专业能力高时薪和灵活工作
既有雇主 / 数据所有者法务或运营负责人n/an/a决定共享数据,还是抵制去中介化担心价值链自动化

Mercor 位于买方预算和专家劳动力供给之间;买方、用户、付款方和现状阻力方并不总是同一方。

[CM012, CM013, CM014, CM015, CM016, CM031]
FM002: 买方 / 客群地图

买方需求从模型质量瓶颈出发,经由专家供给、评估设计,最终落到部署信任要求。

[CM012, CM015, CM017, CM019, CM022, CM023]
FM003: 采用漏斗或价值链图

采用路径先识别模型瓶颈,再进入付费专家工作流,最后变成受信任门槛约束的重复部署。

[CM012, CM013, CM014, CM015, CM023, CM024]

2.4 增长驱动与采用约束

最强的需求驱动来自技术。OpenAI 的 InstructGPT 和 Anthropic 的 Constitutional AI 都表明,即便部分反馈回路已经自动化,对齐和模型质量提升仍然高度依赖反馈。 Mercor、Labelbox、Appen、iMerit 和 CloudFactory 如今都在营销以专家为中心的 RLHF、评估和对齐服务,说明市场已经从低技能标注转向更高判断力的工作。 Mercor 还受益于一次性行业事件:Meta 投资 Scale AI 引发中立性担忧,推动一些实验室寻找替代方案。 但同一市场也存在重要摩擦。劳动权利审视、商业秘密敏感性和低切换成本,让买方保持谨慎,也让定价更有竞争性。 Snorkel 等自动化导向替代品也可能吃掉人类数据支出的低价值层。 结果是,这个市场增长快、紧迫性真实,但并不是只靠自上而下 TAM 就能保证持久支出或轻松续约经济。 实际操作中,买方仍需要信任、安全和可衡量的模型收益,才会大规模铺开。[CM017, CM018, CM019, CM020, CM021, CM022]

增长驱动因素与约束表
驱动因素 / 约束方向时点含义尽调要求
前沿模型需要人类反馈和评测正向当前支撑专家判断工作的持续需求按实验室和领域量化支出
从大宗标注转向专家正向当前相比低技能市场,更利于 Mercor 的高端定位按项目衡量专家占比
Meta 交易后 Scale 中立性受冲击正向2025-2026推动大型实验室更换供应商验证该需求能否延续
智能体 AI 推高基准测试复杂度正向当前提高对工作流更丰富的评测环境需求衡量评测产品复购
商业秘密和知识产权顾虑负向当前限制买方愿意共享的工作流数据量审查客户合同和红线条款
数据工作面临劳工权益审查负向当前可能推高合规成本和品牌风险评估承包者治理和地域组合
供应商切换成本低负向当前价格压力持续偏高衡量合同期限和续约
自动化和合成数据替代负向中期低价值任务可能比高价值任务更快被压缩跟踪软件在哪些环节替代人工

本表把增长驱动和采用约束拆开;多项正向和负向因素可能同时成立。

[CM017, CM018, CM019, CM020, CM021, CM022]
Chapter 03

03竞争格局

3.1 格局结构与直接同业

Mercor 身处一个拥挤但仍在快速洗牌的人类数据市场。直接同业包括 Scale AI、Surge AI、Labelbox、Appen、iMerit、CloudFactory、Invisible、Toloka,以及 Snorkel 等越来越多的自动化导向替代品。 这些供应商并不都以同一种方式解决同一个问题。Scale 和 Appen 从规模与广泛服务宽度切入市场。 Labelbox 从工作流软件加专家网络切入。Mercor 则从高端专家市场切入,并试图爬升到基准和评估软件。 这个差异很重要,因为买方不是在一组完全相同的供应商之间选择。 他们选择的是中立性、专家深度、工作流控制、企业治理和交付速度的组合。 因此,Mercor 不必在每个维度击败每个竞争对手,但它必须在高价值人类判断最重要、买方最不愿接受泛化众包的地方守住清晰优势。[CP001, CP002, CP003, CP004, CP005, CP006]

竞争对手概况表
竞争对手类别规模 / 融资目标客群差异化局限
Mercor专家市场 + 后训练服务2025 年估值 $10B前沿实验室和专家主导的企业项目高端专家寻源速度仍在补软件 / 工作流锁定效应
Scale AI人类数据基础设施老牌厂商Meta 交易后隐含估值约 $29B大型实验室和企业品牌规模、广泛企业覆盖Meta 入股后中立性受质疑
Surge AI高端 RLHF 服务大型非上市竞争对手前沿实验室高技能 RLHF 定位公开产品细节较少
Labelbox数据工厂软件 + 专家网络VC 支持的软件平台模型构建者和企业 AI 团队工作流控制 + 专家供给平台原生程度不如 Mercor
Appen上市人类数据老牌厂商上市公司业绩和投资者关系信息企业 AI 和广泛训练数据规模与治理广度传统众包业务敞口
iMerit托管服务专家非上市服务商高风险领域和模型评测领域专家深度公开品牌存在感较低
Invisible TechnologiesAI 运营平台非上市邻近竞争对手需要智能体 + 人的企业模块化运营栈定位更宽,专业化较弱

规模或融资只反映已抓取来源中的公开数字;多家非上市竞争对手不披露当前资本或收入。

[CP001, CP002, CP003, CP004, CP005, CP006]
FP001: 竞争定位图

相比软件优先和规模优先的对手,Mercor 位于专家深度高、工作流控制中等的象限。

[CP008, CP009, CP010, CP011, CP023, CP025]

3.2 能力宽度与产品包装

这个品类的能力重叠正在加深。Mercor、Scale、Labelbox 和 Appen 现在都在营销 RLHF、评估、对齐或基准式服务的某种组合。 这种收敛削弱了简单基于功能的差异化,并把竞争推向执行力。 Mercor 仍然突出,因为它明确营销领域专家劳动力和高技能类别;软件中心化的竞争对手则强调工作流工具、数据运营和工厂式控制。 产品包装也强化了这个分野。直接同业普遍缺少公开标价,意味着买方通常会谈定制化企业合同,并按速度、信任、中立性和质量评估供应商,而不是看表面价目表。 实际上,竞争焦点不是谁的网站公布了价格,而是谁的专家供给、产品控制和治理组合最适合某个具体项目。 这往往有利于窄用例里的专业厂商,以及大规模企业铺开中的更宽老牌厂商,尤其当采购团队需要客户背书、安全控制和已验证的集成深度时。[CP008, CP009, CP010, CP011, CP012, CP016]

功能 / 能力矩阵
采购标准MercorScale AILabelboxAppenSnorkelInvisible
专家领域市场中等中等中等中等
工作流软件 / 数据工厂中等中等
基准测试 / 评测环境中等中等中等中等
公开治理 / 报告
Meta-Scale 交易后的中立性叙事更强叙事更弱中性中性中性中性
自动化优先的替代风险

单元格是基于证据的序位判断,综合已抓取的官方页面和新闻,而非供应商自行撰写的对比页。

[CP008, CP009, CP010, CP011, CP012, CP019]
定价 / 包装对比
供应商可观察定价模式公开透明度包含能力含义
Mercor按小时专家工作,加匹配 / 寻访收费逻辑专家寻源、项目交付、新兴评测软件灵活,但更难对标
Scale AI定制企业定价数据服务、RLHF、企业 AI 系统销售主导的老牌厂商打法
Surge AI定制企业定价高端 RLHF 和数据工作高端竞争对手,但没有公开价格锚
Labelbox定制企业定价低到中工作流软件、数据工厂、专家网络软件主导的扩张路径
Appen定制企业定价广泛人类数据生命周期、前沿对齐、智能体 AI老牌厂商广度可能挤压专业化空间
Toloka / 平台型众包供应商平台式任务经济模型相对同业为中众包任务和训练数据凸显专家市场深度与大宗吞吐的差异

该品类基本没有公开标价;定价对比因此聚焦包装和透明度,而非费率表精度。

[CP016, CP017, CP018]

3.3 切换成本、多栖与分发

这个品类看起来在结构上就是多栖的。专家很可能同时在多个市场工作,买方也可以测试多家供应商,因为大多数供给仍按项目或定制方案销售,而不是深度嵌入的记录系统软件。 这削弱了纯市场护城河。同时,市场又会受到阶段性分发冲击塑形。 Meta 投资 Scale AI 引发中立性担忧、CNBC 报道 OpenAI 正在收尾与 Scale 的合作时,Mercor 显然从中受益。 但如果 Mercor 不能把这次需求冲击转化成重复工作流或更硬的产品锁定,这些收益可能只是暂时的。 Appen 展示了相反模式:它是上市老牌厂商,具备治理宽度和更长的企业记录。 Mercor 的挑战,是一边保持速度和中立性的叙事,一边建立足够多的工作流所有权,让买方不会在市场条件变化时简单轮换到下一个可接受供应商。[CP013, CP014, CP015, CP019, CP020, CP021]

切换成本和多平台使用表
维度当前证据重要性尽调要求
买方切换成本看起来低到中实验室可并行测试多家供应商索取合同期限和排他条款
专家切换成本专家大概率可跨市场接单衡量专家复用率和排他性
工作流锁定效应软件型竞争对手更强可能把价值捕获从劳动力匹配转走审查 API、数据和评测产品留存
装机基础优势Appen 强于 Mercor有利于企业采购比较客户年限和交叉销售
中立性溢价Scale / Meta 冲击后的临时溢价市场重置后可能消退验证买方迁移是否延续到 2026 年

本表概括当前市场为何结构性多平台使用,以及哪些证据能证明锁定效应会随时间增强。

[CP013, CP014, CP015, CP024, CP025, CP027]

3.4 护城河耐久性与竞争风险

Mercor 最强的竞争论点,是它能快速找到高端专家,并比更慢或更偏软件的竞争对手更快把他们转化成有经济价值的后训练工作流。 但单靠这一点还不是持久护城河。Labelbox 和 Snorkel 等软件优先厂商正在争夺通常产生锁定效应的工作流层。 Appen 和 iMerit 等托管服务老牌厂商可以销售治理、宽度和已建立的买方关系。 与此同时,如果专家多栖,或匹配变得更容易自动化,市场型优势往往会被侵蚀。 因此,Mercor 进军基准和评估环境具有战略重要性:这是在劳动力市场之上创造产品级粘性的尝试。 投资者应把公司视为在部分紧迫买方场景中拥有当下竞争优势,但它仍在市场成熟、当下中立性顺风消退之前,争取建立更持久的胜出权。 在经常性产品绑定得到证明之前,公司应被视为强劲但仍处于过渡期的竞争者,而不是已经设防的品类赢家。[CP023, CP024, CP025, CP026, CP027, CP028]

护城河耐久性 / 竞争风险清单
护城河主张威胁严重性缓释 / 应对剩余风险
高端专家供给专家可在多家供应商同时接单沉淀可复用工作流和供给端忠诚度仍高
中立性叙事Mercor 自身也可能集中在少数实验室客户上分散客户结构
寻源速度软件型竞争对手可把寻源嵌入更强工作流产品切入评测软件和基准测试
基准测试产品老牌厂商可复制或收购类似评测资产用经济上真实的任务拉开差异
年轻公司敏捷性上市老牌厂商可凭治理赢得企业信任让管理层和控制体系专业化
专家市场经济模型自动化替代会压缩低价值任务层聚焦专家判断,而非机械标注
Scale 动荡带来的顺风Scale 可能逐步修复客户信任中立优势消退前锁定买方
品牌动能Mercor 仍小于 Scale,可能也小于 Surge抓住当前增长窗口

风险评级反映竞争耐久性,不包括法律或安全风险;后者在报告后文覆盖。

[CP019, CP020, CP021, CP022, CP023, CP024]
FP002: 护城河 / 就绪度 KPI

Mercor 在专家供给速度和当前中立性叙事上得分较高,但公开治理证据和软件锁定效应偏弱。

[CP013, CP019, CP020, CP023, CP024, CP025]
Chapter 04

04财务情况

4.1 收入模式与收入质量

Mercor 的公开财务故事很亮眼,但结构上很容易被误读。 这不是一个有干净席位订阅收入的经典 SaaS 模型。公开报道和 Mercor 自有文档描述的是市场加服务模式:客户付钱给 Mercor,让它为 AI 模型训练、评估和相关工作流设计寻找、筛选并协调领域专家;Mercor 再通过自己的付款栈向这些专家付款。 这个差异很重要,因为 Mercor 增长最快的表观数字似乎包含承包商付款前的客户总账单额。 TechCrunch 明确报道,Mercor 统计的是专家拿到自己份额之前客户支付的总额。 这种会计选择在该品类中也许常见,但意味着投资者必须先拆分吞吐量与净经济性,才可以把表观收入运行率当作软件收入处理。 公开的劳动者侧定价信号和已报道小时费率机制显示公司具备真实货币化能力,但确切抽佣率和实际折扣仍未披露。[CI001, CI002, CI003, CI004, CI018, CI019]

收入流表
收入流机制单位 / 分母当前价值 / 状态质量 / 利润率判断尽调要求
专家市场业务客户为模型训练和评测购买领域专家服务小时数或项目范围当前核心收入流需求大概率健康,但劳务转付占比高按工作流拆分总账单额、专家付款和 Mercor 抽佣率
匹配 / 寻访经济性Mercor 在专家工作之上叠加寻访费或匹配费率小时价差 / 安置费TechCrunch 公开报道显示变现不止纯工资代发处理提供标准合同模板和实际费率表
基准测试和评测服务Mercor 销售基准测试设计、评测环境和相关数据工作项目或计划费用战略层持续扩张如果可复用资产能附加,毛利率可能好于纯人力派遣披露基准测试绑定到市场工作流的比例
企业 AI 工作流设计Mercor 正向企业智能体和工作流产品扩展项目费或软件赋能服务初现如果劳动密度更低,收入结构可能改善展示经常性收入以及软件 / 服务拆分
承包者总吞吐额客户向 Mercor 支付未扣专家付款的总额总账单额支撑表观收入规模若误当作净收入,会放大规模观感勾稽账单额、付款、净收入和递延收入
付款运营Mercor 通过 Stripe 管理每周付款,有时也用 Wise支付通道 / 交易流运营骨干必不可少,但可能产生费用且合规负担重量化支付成本、失败率和付款浮存敞口

本表区分两件事:公开来源显示 Mercor 卖什么,以及投资测算方还缺哪些变量,才能准确建模总额到净额的经济性。

[CI001, CI002, CI003, CI004, CI018, CI019]
定价 / 变现表
信号价格 / 单位标价与实际含义未知项
寻访 / 匹配经济性按小时寻访费加匹配费率TechCrunch 报道的实际经济性,不是标价Mercor 可能赚取价差,而不是收固定 SaaS 席位费按客户、领域或合同类型拆分的精确抽成阶梯
顶尖专家上行空间最高 $200 / 小时第三方报道的实际样本高端领域可支撑高价值项目该类费率出现多频繁,以及能留下多少毛利率
金融 / IR 专家列表$80-$160 / 小时观察到的市场信号显示高技能知识工作定位客户计费费率是否显著高于工作者费率
股票研究专家列表$120 / 小时观察到的市场信号支撑 AI 买方在金融领域的需求判断其代表性价格还是促销价格
付款入驻首笔 Stripe 付款保留 7 天;需绑定银行账户并启用 SMS 2FA运营政策给付款运营增加摩擦和支持工作量实际付款失败率和每名承包者支持成本

公开定价证据稀少,且多面向工作者或来自管理层转述;客户实际价格仍是尽调事项。

[CI003, CI004, CI020, CI021, CI022, CI038]
FI001: 收入模型桥

Mercor 似乎把集中的 AI 实验室需求转成总开票额、专家付款,以及一层更小但未公开拆分的净收入。

[CI001, CI002, CI003, CI004, CI016, CI018]

4.2 单位经济与运营杠杆

Mercor 的增长弧线异常陡峭。公开来源显示,公司从 February 2025 的 $75 million 收入运行率和已报道盈利,走到 September 2025 约 $450 million 年化收入,再到 Mercor 自己声称在 2026 早些时候已跨过 $1 billion 年化收入。 KTVU 另引述 Brendan Foody 称,公司到 March 2025 已跨过 $100 million 收入运行率,并且极其盈利。 这些数据点拼出一个可信的运营杠杆故事,尤其 TechCrunch 还报道公司在 2025 上半年实现 $6 million 利润。 但同一组证据也指向一个重度转付的劳动力基础,而不是纯软件式利润率结构。 每日承包商付款从 late 2025 超过 $1.5 million,上升到 Mercor 2026 帖子中的超过 $2 million;内部帖子还把支付、合同系统和控制升级描述为关键扩张工作。 结果是,收入端证明令人鼓舞,但公开毛利率、抽佣率或现金转化细节仍不足,无法把动能转成可完整承销的模型。[CI009, CI010, CI011, CI012, CI013, CI014]

单位经济性表
指标公开数值 / 状态置信度重要性尽调请求
Feb 2025 ARR / 运行率$75M+Series B 后最早的公开规模锚点提供月度经常性与非经常性收入桥接表
Sep 2025 年化运行率~$450M显示一年内收入端极快加速提供月度收入序列和队列拆解
Early-2026 年化运行率$1B+(公司声称)暗示 Series C 后继续超高速增长将 2026 年运行率方法与经审计或经复核数字对齐
H1 2025 利润$6M(第三方报道)承包者密集模型少见的运营杠杆信号提供含毛利、运营费用和现金流的损益表
每日承包者付款Oct 2025 为 $1.5M+;2026 帖文称 $2M+转付人工支出是现金转化和控制的核心提供付款规模、费用负担、拒付和准备金政策
抽成率 / 毛利率未公开披露投资测算收入质量的核心变量披露按项目拆分的客户费率、工作者费率、支付成本和毛利率

本表混合已佐证数字与明确缺失变量,方便投资测算方看清公开证据止步在哪里。

[CI009, CI010, CI011, CI012, CI013, CI015]
FI002: 单位经济性桥

公开证据显示 Mercor 增长和付款规模都异常高,但从总流水走到净利润率,中间步骤仍缺失。

[CI009, CI010, CI011, CI012, CI015, CI016]
FI004: 财务估算区间

公开财务信号给出了增长和付款规模的点估计,真正拉开区间的仍是缺失的现金和利润率数据。

[CI009, CI010, CI011, CI015, CI016, CI017]

4.3 资本充足性与融资依赖

表面看,Mercor 并不缺资本。公司披露从种子轮、Series A、Series B 到 Series C 共约 $483.6 million 新股融资,并在 October 2025 以 $10 billion 估值完成 $350 million Series C。 管理层还给出异常具体的资金用途:扩大人才网络、改善匹配、加快交付。 对一家既要防守市场流动性、又要建设更产品化评估能力的公司来说,这些优先级合理。 问题在于,已披露融资历史不等于已披露流动性。Mercor 在本章审阅的来源中没有公布手头现金、月度烧钱速度、现金跑道月数或任何债务安排。 这个缺口很重要,因为公司如今对客户集中、法律和泄露补救成本,以及高速增长要求的内控投资,都有真实下行情景敏感性。 因此,融资时间线是公司能拿到资本的强证据,但仍不足以断定压力情景下现金跑道安全。[CI005, CI006, CI007, CI008, CI024, CI025]

资本充足性表
字段公开数值 / 状态来源 / 时点含义缺口或下一步请求
已披露新股融资种子轮、Series A、B、C 合计 $483.6M2023-2025 公开融资公告Mercor 已筹得足够股权资金,可支持激进内部建设需要股权结构表、优先股堆叠和任何老股交易
最新定价轮$350M Series C,估值 $10BOct 2025后期股权融资显著减轻短期融资压力需要扣除费用后的现金到账额,以及任何投资人权利
资金用途人才网络、匹配、更快交付、更广能力建设Series B 和 Series C 帖文资金似乎同时投向供给层和产品层索取交割后实际预算分配
账上现金未公开披露n/a尽管融资轮规模很大,仍无法测试现金跑道索取最新资产负债表和流动性安排
月度烧钱速度未公开披露n/a无法把估值和融资额转换为现金跑道索取按月现金消耗和计划招聘支出
债务 / 项目融资义务已获取来源未披露公开债务或授信额度n/a表面是正面信号,但不足以排除义务索取债务明细、供应商融资和或有负债
下一轮融资触发因素可能更多取决于增长、信任和客户集中度冲击,而不是名义现金短缺由快速扩张和违规风险推断大额私募融资不消除应急规划需求建立客户复核拖慢增长或压低毛利率的下行情景模型

Mercor 的公开融资故事很强,但公开流动性数据仍太稀少,不能只靠公告测算现金跑道。

[CI005, CI006, CI007, CI008, CI024, CI025]
FI003: 资本强度 / 现金流图

Mercor 已披露的股权融资较充足,但未来资金需求仍取决于信任、集中度,以及缺失的现金转化数据。

[CI003, CI004, CI005, CI006, CI007, CI008]

4.4 负面信号与承销阻碍

Mercor 财务章节最大的未解问题,不是公司能不能融资,而是投资者能不能把公开增长叙事转化成可信的收入质量判断。 TechCrunch 称,一小部分大品牌贡献了过大的收入份额;Scale AI 诉讼则描述了价值数百万美元的单一客户机会。 这种组合意味着大客户上行空间,也意味着集中度暴露。April 2026 泄露事件又增加一层财务风险,因为 TechCrunch 报道 Meta 暂停合同,其他模型厂商正在审查关系,同时五名承包商就所谓数据暴露提起诉讼。 品类波动性也真实存在:CNBC 报道,Scale AI 在 Meta 交易后试图赢回客户时裁员 14%。 Appen 这类上市老牌厂商的披露展示了 Mercor 尚未提供的东西:定期业绩、更清晰的服务分部,以及更透明的报告。 在 Mercor 披露现金、烧钱速度、抽佣率、利润率和集中度指标之前,承销应把公司视为财务表现异常突出、但仍部分不透明。[CI017, CI018, CI026, CI027, CI028, CI029]

公开财务缺口表
缺失指标公开来源给出的替代信息为何不足对投资测算的影响具体尽调路径
现金余额和短期流动性大额融资轮和盈利轶事融资规模不等于剩余现金无法验证现金跑道索取最新现金余额、受限现金和预测表
月度烧钱速度和经营现金流只有利润和运行率片段会计利润看不出现金转化下行情景规划仍只能推测索取月度 P&L 和现金流量表
按工作流拆分的净抽成率总收入和付款数字需要把吞吐规模和净经济性分开估值倍数选择可能被扭曲索取各项目的总账单额、专家付款和 Mercor 净收入
按领域和客户类型拆分的毛利率只有专家小时费率样本和付款规模看不到扣除人工和支付成本后的贡献毛利难以清晰对比 Mercor 与软件或服务同业索取按用例拆分的毛利率桥接表
客户集中度和合同期限只定性提到部分品牌集中度没有分母或续约数据收入质量和耐久性仍未定索取前 10 大客户收入占比、合同期限和队列留存
违规 / 合规下行成本据报道 Meta 工作暂停,承包者提起诉讼未披露准备金或补救成本可能显著改变烧钱速度和融资需求索取事件成本估算、法律准备金和补救预算

尽管增长叙事异常强劲,上述项目仍是阻碍清晰投资测算模型的最小缺失输入。

[CI017, CI018, CI026, CI027, CI030, CI033]
Chapter 05

05产品与技术

5.1 产品形态与客户工作流

Mercor 的产品比「AI 招聘初创公司」这个标签更宽。 当前公开产品形态结合了专家市场、AI 面试与匹配系统、基准产品,以及新兴的企业工作流设计供给。 Mercor 的研究页面强调基准、评估环境和大规模人类数据集;专家页则展示底层运营基座:一个全球分布的专业人士池,可以被面试、匹配、管理并为改善 AI 系统的工作获得报酬。 公司的 Enterprise AI 帖子把产品边界又往前推了一步,主张企业智能体的瓶颈不只是模型智能,也包括缺少有证据支撑的工作流设计。 这种叙事把 Mercor 从人力中介转成工作流捕捉、专家判断和评估基础设施提供商。 客户买到的不是单一独立功能,而是一套协调高技能人类反馈、基准复用,并最终支持经常性企业工作流系统的操作系统。[CE001, CE002, CE003, CE009, CE016, CE020]

产品模块 / 资产矩阵
模块 / 资产主要用户状态 / 成熟度差异化尽调缺口
专家市场需要领域专家的 AI 实验室和企业核心 / 已上线高技能领域专家库规模大需要按领域拆分的活跃专家数和复用率
AI 面试官(Monty)候选人和内部运营团队规模化 / 已上线带岗位语境的大规模自动化面试需要客观面试质量和假阴性指标
匹配与录用邀约引擎Mercor 运营和招聘团队核心 / 已上线把档案、面试、评估和可用时间匹配到岗位需要按细分拆分的准确率或转化率指标
APEX 基准测试系列AI 实验室和模型评测团队已上线 / 扩展中覆盖专业、消费者和 SWE 工作,贴近真实经济活动的基准测试需要从基准测试曝光转化为付费工作流的附加率
企业 AI / 智能体设计企业转型团队初现 / 商业化早期从劳动力供给延伸到工作流固化和智能体部署需要具名客户部署和复用指标
合同 / 付款基础设施Mercor 财务和运营关键内部平台复杂计费和全球承包者付款管理是交付核心需要失败率、恢复时间和控制证据
信任与合规层客户、承包者、Mercor 风控团队已运行但披露不完整背景调查、LLM 使用规则、工时追踪和数据政策均有明示需要公开认证、信任中心深度和事件指标

本矩阵把内部交付系统视为产品关键项,因为 Mercor 卖的是协同专家工作流,而不只是静态软件席位。

[CE001, CE003, CE005, CE006, CE009, CE010]
工作流 / 用例表
用户任务当前工作流Mercor 方案可衡量收益限制
招募或筛选专家人才手动搜索、筛选、面试并验证专家Mercor 市场加 AI 面试官和匹配更快、大规模完成专家入驻和筛选精确转化率未公开
训练或评测前沿模型收集偏好数据、基准测试和领域判断Mercor 专家加基准测试资产和评测环境比通用标注更高技能的反馈需要证明产品附加能重复发生
评测智能体表现内部拼装定制测试任务和评分规则APEX、APEX-Agents 和 APEX-SWE 提供可复用评测资产提升模型和任务之间的可比性公开基准测试不等于付费客户采用
落地企业智能体靠人工摸索工作流、提示词和工具调用Mercor Enterprise AI 提出工作流发现和固化方案可减少用例摸索公开文档仍偏高层
运营全球承包者项目向客户开票、追踪工时,并跨司法辖区向工作者付款合同、付款文档和工时追踪工具支撑交付支持大规模劳动力协同公开 SLA、错误率和审计数据仍稀少
守住客户信任审核工作者,并约束不安全的模型评测行为背景调查、LLM 规则和数据使用政策让控制有正式规则提高敏感工作流的基础信任水位缺少公开认证深度或完整信任中心细节

收益只按已获取的工作流证据描述;没有支撑的性能主张仍明确收束为尽调缺口。

[CE003, CE020, CE022, CE025, CE027, CE028]
FE002: 客户工作流 / 运营流程

Mercor 把专家供给和内部编排转成面向客户的基准测试、评估和企业智能体产出。

[CE001, CE003, CE005, CE007, CE009, CE010]

5.2 架构与运营模型

Mercor 最能说明产品的证据来自工程帖子,而不是官网。 Monty 面试官文章描述了一个实时运营系统:每天运行约 10,000 场对话,每个会话隔离在自己的 Modal 容器里,并从暖池启动,把启动时间压到 200 milliseconds 以下。 Contracts-service 重写展示了另一条同样重要的架构真相:内部交付系统和面向用户的 AI 一样重要。 Mercor 公开描述称,公司在一周内重写一个瓶颈服务,将能力提升超过 10,000x、可靠性提升超过 75x,因为原先关于合同量和延迟的假设已经被增长打穿。 这让架构更像运营系统,而不是纯模型中心系统。 Mercor 依赖内部编排、计费、付款和工人管理系统,把软件、人力劳动和客户信任绑在一起。 实际上,它的产品栈是人类数据操作系统,而不只是匹配算法。[CE005, CE006, CE022, CE023, CE024, CE025]

技术 / 运营架构表
层 / 组件角色依赖风险
档案和面试摄入将简历、面试和公开档案转为可搜索的候选人信号Mercor 数据管线和 AI 解析数据质量和隐私敏感性
AI 面试运行时为候选人运行实时面试会话Modal 容器、预热池和房间配置状态冷启动、会话可靠性和基础设施依赖
匹配与合同编排将人才路由到岗位,并管理录用邀约或合同状态包括 Contracts 在内的核心内部服务扩展故障可能卡住交付和付款
基准测试与评测资产管线构建 APEX 数据集、排行榜、评分规则和评测环境Mercor 研究团队加专家贡献者基准测试污染、外部采用率低,或刷新周期成本高
企业智能体工作流层将真实工作映射为智能体任务、提示词、工具调用和评测循环Mercor Enterprise AI 和客户工作流发现公开文档尚未显示深度集成或 API 证据
运营与付款层追踪工时、向客户开票,并向全球承包者付款Stripe、Wise、Insightful/Workpuls、财务控制付款失败、欺诈和司法辖区合规
信任与政策控制背景调查、LLM 使用规则和数据治理约束行为Mercor 文档和内部审核流程认证或执行成熟度的公开证据稀少

Mercor 卖的是结果导向的专家工作流,因此架构既包括软件系统,也包括人工运营控制层。

[CE005, CE006, CE022, CE023, CE024, CE025]
FE001: 产品架构图

Mercor 的产品栈把专家供给、评估资产、内部编排和信任控制叠成一个交付系统。

[CE003, CE005, CE006, CE009, CE010, CE011]
FE003: 关键依赖图

Mercor 同时依赖外部模型进步、专家供给、基准可信度、支付通道和信任姿态。

[CE001, CE003, CE009, CE010, CE011, CE018]

5.3 基准、评估资产与差异化

Mercor 最清晰的产品差异化尝试是 APEX 家族。APEX、APEX-Agents 和 APEX-SWE 不是泛泛的博客营销,而是可复用基准资产,意在让 Mercor 从劳动力市场上升为评估基础设施。 产品逻辑在方法论里很明显。Mercor 称 APEX-Agents 参考了数百名专业人士的调查,APEX-SWE 则在集成和可观测性任务中使用人类撰写的评分标准,而不是单元测试式玩具问题。 扩展版 APEX 帖子还显示,公司愿意刷新方法论:把留出集翻倍,并发布更多关于置信区间和任务时长的细节。 这具有战略意义,因为 Appen、Scale、Toloka 和 iMerit 的竞争对手网站显示,市场已经在专家 RLHF、智能体工作流和评估服务上收敛。 因此,Mercor 需要基准真实性和工作流深度,而不只是思想领导力;这些资产是它超越原始劳动力聚合、形成产品粘性的最佳公开论据。[CE009, CE010, CE011, CE012, CE013, CE014]

路线图 / 发布 / 开发阶段表
日期 / 阶段功能或里程碑状态影响来源
2025-02Series B 轮运营叙事上线公开把产品建设与增长、高级运营招聘挂钩Series B 轮文章
2025-10Series C 轮聚焦匹配和更快交付上线显示公司继续投入工作流速度,而不是单纯堆人Series C 轮文章和 CNBC
2026-03APEX-SWE 发布已发布Mercor 将研究产品化,沉淀为软件工程评估资产APEX-SWE 文章
2026APEX-Agents 扩展已发布把基准范围从静态任务扩到长周期智能体工作APEX-Agents 文章
2026扩展 APEX 留出集已发布说明 Mercor 在迭代评估方法,而不是把一次性基准冻结下来Expanded APEX 文章
2026企业 AI 工作流产品初现推动 Mercor 面向企业沉淀工作流固化能力Enterprise AI 文章
2025-2026内部可靠性重写与 Monty 扩容运营中高速增长下,后端和面试运行时仍在演进Monty 与 Contracts 工程文章

公开页面给出事件日期时按事件日期;否则按获取页面可见的年份或阶段填写。

[CE007, CE010, CE016, CE018, CE020, CE022]
FE004: 产品成熟度 / 能力图

Mercor 在基准真实性和运营交汇处最强;公开集成证明和信任深度仍薄的地方最弱。

[CE001, CE009, CE010, CE020, CE022, CE025]

5.4 信任、控制与技术风险

Mercor 比许多年轻 AI 初创公司拥有更多公开流程文档,但信任画像仍然复杂。 文档索引暴露了一大组面向承包商的指南,覆盖数据使用、LLM 限制、付款、工时追踪和背景调查。 这些政策很重要,因为 Mercor 正在中介敏感工作流,并且据 TechCrunch 报道,处理客户视为商业秘密的数据集和流程。 同一批文档也揭示,产品体验多大程度取决于纪律化运营,而不是模型魔法。 工人要接受背景调查、被追踪,并在外部 LLM 使用上受到限制。 但 April 2026 泄露事件说明,文档不等于成熟信任姿态。TechCrunch 报道称,被盗材料包括源代码、API 密钥、候选人数据和雇主数据,正是会损害客户信心和产品可信度的类型。 在 Mercor 提供更深的公开认证和事件响应证据之前,信任仍是公司最直接的技术风险。[CE017, CE018, CE019, CE020, CE021, CE027]

信任 / 质量 / 合规表
控制 / 质量杠杆状态范围证据缺口
背景调查已有公开文档身份、教育、就业和执照政策页说明流程没有公开审计或完成率统计
LLM 使用限制公开文档可见防止承包方把评估判断外包出去LLM 使用政策没有公开的执行指标或升级处理数据
数据使用披露公开文档可见覆盖简历、面试媒体、公域资料、付款信息数据与 AI 政策没有按工作流披露数据留存期限细节
工时追踪运营文档可见通过 Insightful/Workpuls 追踪项目任务用时操作指南没有公开的准确率或争议率指标
付款运营运营文档可见以 Stripe 为主;有时使用 Wise;需完成银行账户绑定付款指南没有公开的付款失败或欺诈指标
安全成熟度重大顾虑据报道,泄露事件暴露了源代码、API 密钥和敏感客户数据TechCrunch 对泄露事件的报道公开 Trust Center 内容过薄,难以抵消事件风险

相比许多年轻创业公司,Mercor 的公开政策更有深度,但认证深度和事件响应成熟度的公开证据仍有限。

[CE027, CE028, CE029, CE030, CE031, CE032]
Chapter 06

06客户情况

6.1 客户细分与市场覆盖

Mercor 的客户基础锚定在前沿 AI 实验室——这些组织构建并打磨大语言模型和多模态 AI 系统,需要大规模人类反馈数据。 TechCrunch 在 October 2025 将 Mercor 描述为 AI 实验室构建训练数据集的首选供应商,Bloomberg 在 April 2026 的详细人物报道中强化了这一描述。 公开规模信号在吞吐量上强于活跃劳动力规模:CNBC 称 Mercor 到 February 2025 已处理 300,000 名候选人;Mercor 后来称,到 October 2025 名单中已有超过 30,000 名专家。 除前沿实验室外,Mercor 还推出了两个面向客户的新细分:Enterprise AI 产品,面向寻求 AI 辅助招聘和劳动力管理的大型组织;Research 门户,面向寻找领域专家评估者的学术和政府 AI 项目。 Experts 产品线代表市场内的高端层,让 AI 实验室可以接触到具备资质的领域专家,用于需要高级推理或主题知识的任务。 工人侧地理覆盖为全球,但公司没有披露按收入划分的客户地域构成。 公开证据勾勒出的客户细分图景是:少数高价值 AI 实验室关系贡献绝大多数收入,企业和研究细分的多元化仍处早期。[CU003, CU005, CU009, CU015, CU020, CU024]

客户细分表
客户分群代表性客户使用产品收入信号证据强度
前沿 AI 实验室OpenAI 生态、Anthropic 级实验室标注与 RLHF 评估$450M ARR 主要由此贡献高(多方来源)
企业 AI 团队大型科技公司Mercor Enterprise AI(企业 AI)在增长,但未披露中(公司博客)
研究机构学术与政府 AI 项目Mercor Research 门户未披露低(仅公司页面)
AI 创业公司早期模型公司核心标注市场早期增长队列中(Series A 轮报道)

细分基于产品线和媒体报道推断;公司未披露官方拆分。

[CU003, CU005, CU015, CU020, CU024]
FU001: 客户旅程图

从 AI 实验室认知 Mercor,到长期扩大合作的五阶段旅程。

[CU003, CU006, CU010]

6.2 采用轨迹与收入增长

Mercor 的收入增长轨迹是 AI 工具领域最醒目的案例之一。公司在 2025 从大约每月 $2 million 收入,增长到每日 $2 million——约 30 倍。 多个独立来源印证了这一增长:TechCrunch 报道 September 2025 年化收入运行率为 $450 million,Forbes 在 AI Cloud 100 报道中确认了这一数字,Bloomberg 在 April 2026 人物报道中提供了更多背景。 融资轨迹讲述了平行故事:February 2024 的 $34 million Series A、February 2025 以 $2 billion 估值完成的 $100 million Series B,以及 October 2025 以 $10 billion 估值完成的又一轮 $350 million Series C——约 20 个月估值增长 40x。 这种增长模式符合一家抓住 AI 实验室扩张模型训练运营需求的公司。CNBC 关于 Scale AI 失去 OpenAI 和 Google 客户的报道说明,Mercor 的部分增长可能来自客户从 Scale AI 关系迁移出来,或在原关系之外补充 Mercor。 从注册工人到活跃项目部署的采用漏斗没有公开细节,但 talent.docs.mercor.com 的文档显示,公司有结构化入职和项目里程碑系统,可支持新客户部署快速扩张。[CU001, CU002, CU006, CU007, CU008, CU016]

客户增长 / 采用轨迹表
时期收入指标估值关键客户事件来源
Q1 2024~$2M ARR(估计)$34M 融资(Series A 轮)Series A 轮完成TechCrunch 2024 年 2 月
Q1 2025~$75M ARR(据报道)$2B 估值Series B 轮完成TechCrunch/CNBC 2025 年 2 月
Q3 2025据报道 $450M ARR$2B(Series C 前)Scale AI 提起诉讼TechCrunch/Forbes 2025 年 9 月
Q4 2025$600M+ ARR(估计)$10B 估值Series C 轮完成TechCrunch/Forbes 2025 年 10 月
Q2 2026$700M+ ARR(估计)维持 $10BBloomberg 特写Bloomberg 2026 年 4 月

Q1 2024 和 2026 年的 ARR 估计值由增长叙事外推;只有 Q3 2025 数字有直接报道。

[CU001, CU002, CU006, CU007, CU016]
FU002: 采用 / 部署漏斗

从可用工人池到活跃客户项目部署的估算漏斗。

[CU009, CU017, CU025]

6.3 具名客户证明与证据质量

Mercor 的客户证明质量受 AI 实验室供应商关系保密性质限制。没有前沿 AI 实验室公开确认 Mercor 是供应商。 证据基础来自记者描述、公司博客帖子,以及 Scale AI 诉讼给出的隐含信号——该诉讼指控 Mercor 侵占客户关系商业秘密,意味着 Mercor 正在积极竞争或赢得 Scale AI 认为属于自己的 AI 实验室业务。 TechCrunch 在 October 2025 的详细文章中描述 AI 实验室使用 Mercor 构建训练数据集;Mercor 自己的 Disrupt 2024 博客帖子展示了实时 AI 评估工作流,这些工作流预设了真实客户部署。 Forbes AI Cloud 100 入选则提供第三方分析师验证,说明投资者和行业观察者认为 Mercor 的客户基础可信。 本章的具名客户证明表梳理了所有可公开归因的客户引用;结果是稀疏但方向一致的图景。 从尽调角度看,缺少具名背书是重大缺口。任何投资决定都应要求至少前三大收入贡献客户出具客户推荐信或 LOI。[CU003, CU004, CU008, CU011, CU012, CU014]

具名客户证据表
客户 / 交易对手关系类型证据来源证据类型置信度
AI 实验室生态(OpenAI 级)主要标注客户TechCrunch 2025 年 10 月记者报道
Scale AI(间接证据)竞争对手纠纷显示双方客户基础重叠Axios/Bloomberg 2025 年 9 月法律文件背景
Mercor Enterprise 早期采用者企业试点客户Mercor 博客 2025 年 3 月公司公告
Forbes AI Cloud 100 投票者行业认可暗示客户验证更广Forbes 2025 年 9 月行业榜单

尚无 AI 实验室公开点名确认 Mercor 是供应商。证据来自记者描述的推断。

[CU003, CU004, CU008, CU011, CU014]
FU003: 客户证明矩阵

将客户分群映射到证据质量维度,检验 Mercor 牵引力证明。

[CU001, CU003, CU008, CU012]

6.4 留存、扩张与集中度风险

Mercor 的留存和扩张经济性几乎完全不透明。公司没有公开披露净收入留存(NRR)、总收入留存(GRR)、流失率或队列数据。 公司也没有发布客户满意度评分、续约率或多年合同细节。 唯一间接留存证据来自收入增长叙事:一年内收入增长 30x,意味着现有客户大幅扩张、新客户高速获取,或两者同时发生。 集中度风险是首要结构性担忧。收入似乎压倒性来自少数前沿 AI 实验室关系;如果其中一两家客户减少支出——就像 OpenAI 和 Google 对 Scale AI 所做的那样——对 Mercor 收入的冲击可能很严重。 Scale AI 诉讼又增加了一个维度:关于 Mercor 挖走客户关系的指控,提高了合同条款被争议或客户层面法律风险的可能性。 在工人供给侧,Rest of World 报道过 AI 数据标注员普遍面临的质量和留存挑战;Mercor 的结构化入职和基于里程碑的访问系统说明公司意识到了这些动态。 talent.docs.mercor.com 文档显示,Mercor 使用合同机制管理工人访问权限,这可能成为留住优质工人的工具。[CU021, CU022, CU023, CU011, CU017, CU025]

留存 / 重复使用 / 满意度表
指标披露值来源缺口 / 备注
净收入留存率(NRR)未披露N/A关键缺失指标
人才留存率未披露N/A从人才规模增长看,推断较高
客户续约率未披露N/A无公开数据
满意度得分(NPS)未披露N/A未发布客户调研数据
重复项目率暗示较高(增长叙事)TechCrunch/Forbes 2025 年 10 月仅为间接信号

Mercor 尚未公开披露任何留存、满意度或重复使用指标。增长轨迹暗示留存强劲,但未获确认。

[CU023, CU025, CU026]
扩张与集中度风险表
风险因素证据严重性缓释因素
客户集中度未披露拆分;AI 实验室主导收入企业客户多元化推进中
单一分群依赖~100% 收入来自 AI 训练市场已推出 Research 和 Enterprise 产品
平台切换先例Google 和 OpenAI 削减了对 Scale AI 的支出Mercor 品牌差异化
Scale AI 诉讼阴影商业秘密诉讼仍在进行法律抗辩;案件未决
人才供给约束Rest of World 报道质量挑战结构化入驻流程有文档记录

由于公司未披露客户拆分且聚焦单一垂直市场,集中度风险是首要结构性担忧。

[CU021, CU022, CU011, CU017, CU025]
FU004: 留存 / 复用队列

基于可用代理证据估算的工人队列按项目月留存,仅作示意;客户层面留存未公开披露。

[CU023, CU025, CU017]
Chapter 07

07风险

7.1 监管、法律与诉讼风险

Mercor 最大的监管暴露是工人错误分类。公司公开披露,到 October 2025 名单中已有超过 30,000 名专家,同年早些时候还处理了 300,000 名候选人。 California 的 AB 5 对工人分类施加 ABC 测试,California DIR 和 FTB 也都发布了该法律如何适用于零工经济平台的具体指引。 2024 federal DOL 独立承包商规则进一步收紧了联邦层面的经济现实测试,IRS 也就同一问题发布了平行指引。 California Supreme Court 在 2024 一起重大卡车运输案中围绕 AB 5 作出的裁决,显示司法机构仍愿意把零工工人保护扩展到最初范围之外。 如果 Mercor 的标注员在 California 被重新分类为雇员,公司将面临欠薪、福利、工资税和罚金方面的潜在责任,影响其专家网络中相当大一部分人。 公司没有披露准备金金额或法律暴露估计。诉讼方面,Scale AI 在 September 2025 提起商业秘密诉讼,指控 Mercor 侵占专有客户数据和定价信息。 法院记录后来显示,该案在 January 2026 被自愿且不可再诉地驳回,但 Bloomberg、TechCrunch 和 Axios 的报道仍突出底层商业秘密和客户获取控制风险。 EU AI Act 已于 2024 生效,也带来额外风险:用于就业和工人分配决策的 AI 系统被归为高风险,部署到 EU 市场前可能需要进行合规评估。[CR001, CR002, CR003, CR004, CR005, CR010]

监管 / 法律风险登记表
风险项司法辖区严重性概率关键证据
AB 5 工人误分类加州极高DIR、FTB、加州立法机构 AB 5 文本
联邦承包方重新分类美国联邦DOL 2024 年独立承包方规则
Scale AI 商业秘密诉讼美国联邦(NDCA)进行中CourtListener、PACER、Bloomberg
数据泄露集体诉讼(Gill)美国联邦(NDCA)进行中CourtListener、Claim Depot
EU AI Act 的就业 AI 适用范围欧盟EU AI Act 官方文本(CELEX)
GDPR 跨境传输欧盟 / 国际由欧盟标注员基础推断
IP 所有权纠纷多司法辖区数据 AI 使用政策;推断

仅纳入公开已知或推断风险。内部法律风险登记表不可供审阅。

[CR001, CR002, CR004, CR005, CR006, CR020]
FR001: 风险热力图

Mercor 主要风险类别的严重性与发生概率矩阵。

[CR002, CR005, CR006, CR007, CR012]

7.2 运营、安全与质量风险

March 2026,Mercor 确认遭遇网络攻击,部分用户个人数据暴露。 TechCrunch 报道了这起泄露;几天内,一起联邦集体诉讼就在 California Northern District 提起——Gill v. Mercorio Corporation——指控公司存在疏忽的数据安全实践。 Claim Depot 和 CourtListener 都确认该案仍在审理。TechCrunch 在后续报道中指出,泄露与 Scale AI 诉讼同时发生,形成叠加声誉风险。 Mercor 在 trust.mercor.com 维护 Trust Center,但没有披露任何 SOC 2 Type II 认证、ISO 27001 认证或 NIST CSF 符合性评估。 2024 NIST CSF 为处理敏感个人数据的组织建立最佳实践控制;Mercor 是否符合仍未知。 在运营质量侧,Mercor 自己的博客文章描述平台在 10x 流量峰值下几乎失效,暴露出基础设施脆弱性;公司没有以任何公开描述的方式说明已补救。 Rest of World 记录了整个行业 AI 标注工人面临的系统性质量挑战,说明供给侧质量风险并非 Mercor 独有,但考虑到输出质量是 Mercor 的核心价值主张,这一风险很重要。 目前没有公开错误率、SLA 违约率或质量审计结果。[CR006, CR007, CR008, CR009, CR013, CR014]

运营 / 质量 / 安全风险登记表
风险项严重性证据缓释证据缺口
2026 年 3 月网络攻击 / 数据泄露极高TechCrunch 2026 年 3 月已公开披露未披露 SOC 2 或 NIST 合规情况
标注质量波动Rest of World 报道基于里程碑的访问系统未发布错误率
平台扩容宕机风险Mercor 10 倍量级博客事后工程投入未披露容量 SLA
人才数据访问控制由泄露范围推断Trust Center 存在无访问控制文档
GDPR 数据驻留合规推断有欧盟人才基础数据 AI 使用政策无 GDPR DPA 文档
第三方基础设施依赖标准云架构(推断)Trust Center未披露 BCP 或 RTO

运营风险评估仅基于公开证据。内部审计和安全认证不可获得。

[CR007, CR008, CR009, CR013, CR014, CR018]
FR002: 风险传导图

有向无环图展示 Mercor 的一级风险如何级联成二级和三级后果。

[CR005, CR006, CR007, CR012, CR022]

7.3 伙伴、依赖与人员风险

客户集中是 Mercor 的结构性风险。多名记者的证据指向一个由前沿 AI 实验室关系主导的收入基础。 直接先例很刺眼:Scale AI 在短时间内失去 OpenAI 和 Google 客户,导致裁员 14%。Reuters 确认 OpenAI 在 June 2025 收尾与 Scale AI 的合作。 如果 Mercor 前一或前二大客户以类似速度减少合作,在没有快速替代的情况下,收入冲击可能是灾难性的。 Mercor 已尝试通过 Enterprise AI 和 Research 细分多元化,但没有披露这些细分的收入贡献。 人员侧,创始团队由年轻工程师组成;KTVU 和 Times of India 的媒体报道强调了他们的技术能力,但公开披露中也缺少经验丰富的运营高管。 Rest of World 和 Time magazine 都记录过与 Mercor 类似平台上的 AI 标注员工资和劳动权利担忧;这些担忧直接适用于 Mercor 的全球专家网络,公司称该网络到 October 2025 已超过 30,000 人。 人才门户文档显示,Mercor 已建立争议解决路径和合同框架,说明公司意识到这一暴露,但没有披露独立劳动审计。[CR012, CR013, CR015, CR017, CR019, CR024]

合作伙伴 / 依赖风险登记表
依赖项类型集中度风险缓释证据
前沿 AI 实验室客户收入集中极高Enterprise/Research 多元化TechCrunch、Forbes、Bloomberg
云基础设施提供商技术依赖未知;多云未确认由规模推断
独立承包方供给劳动力供给300k+ 人才池;地域分散多家媒体来源
Scale AI 竞争压力市场风险靠品牌和速度差异化诉讼和媒体报道背景
付款 / 薪资处理商金融依赖有多个选项由承包方模式推断

合作伙伴依赖由商业模式推断;具体供应商名称未公开披露。

[CR012, CR014, CR019, CR026]
人员 / 执行风险登记表
风险项严重程度证据缓释措施
创始人执行风险(团队年轻)KTVU、Times of India(早期报道)已有资深投资人支持
工作者劳动权益 / 工资投诉Rest of World、Time magazine 等来源结构化合同;争议解决门户
对创始人的关键人物依赖未公开任命 COO/CPO管理层厚度未披露
文化扩张风险从增长推断员工数快速扩大未披露
规模化后的标注员质量下滑Rest of World(2023)基于里程碑开放项目权限

人员风险可通过媒体报道部分观察;公开来源中的管理团队细节有限。

[CR013, CR017, CR025, CR029]
FR003: 依赖图

依赖图展示 Mercor 的关键运营和财务依赖及其相互连接。

[CR010, CR012, CR014, CR032]

7.4 缓释、终止标准与尽调要求

Mercor 公开可观察的缓释措施是部分性的,而且大多没有文档支撑。 trust.mercor.com 的 Trust Center 提供了基础安全姿态信号。人才门户合同和法律支持文档显示,Mercor 为工人关系搭建了法律框架。 公司在 March 2026 公开披露泄露事件,说明具备可运转的事件响应能力。 但是,公司没有发布 SOC 2 报告、保险披露、容量 SLA、质量审计结果或监管准备金金额。 从投资监控角度看,终止标准应包括:AB 5 不利裁决或 DOL 执法行动;法院作出不利认定,或出现新的商业秘密争议,显示公司不当使用竞争对手客户材料;12 个月内发生第二次重大安全泄露;或确认因单一客户流失而失去超过 50% 收入。 需要监控的早期预警信号包括 CourtListener 上泄露案件案卷更新、任何新的竞争对手诉讼、DOL 和 NLRB 执法追踪器、Mercor Trust Center 更新,以及公司提供的月度 ARR 桥接数据。 最关键的尽调要求仍是:量化 AB 5 暴露的法律备忘录、SOC 2 Type II 报告或等同材料、前十大客户收入拆分,以及网络和 E&O 风险保险覆盖确认。[CR008, CR015, CR020, CR022, CR024, CR025]

缓释措施和否决标准表
风险类别当前缓释措施否决标准监测信号
监管 / AB 5承包商自我认证;法律顾问(推断)不利的 AB 5 裁决或 DOL 执法行动DOL 执法跟踪;NLRB 案件备案
商业秘密诉讼正在进行法律抗辩限制客户拓展的初步禁令CourtListener 案卷更新;媒体报道
数据泄露 / 网络安全Trust Center;公开泄露披露12 个月内发生第二次重大泄露HaveIBeenPwned;监管文件
客户集中度企业 / 研究客户多元化单一客户流失导致 >50% 收入损失月度 ARR 桥接;客户 NRR
工作者质量 / 供给基于里程碑开放权限;入职文档客户 SLA 违约率超过阈值SLA 违约报告;客户满意度

否决标准是投资人组合监控的示例阈值;实际阈值应在投资监控框架中设定。

[CR005, CR007, CR012, CR022, CR025]
Chapter 08

08估值

8.1 投资逻辑与反向逻辑

Mercor 值得估值关注,因为公司已经凑齐了市场明确想要的三样东西:极快增长、接触前沿 AI 客户的入口,以及从纯招聘走向基准、评估和工作流工具的可信路径。 官方融资历史显示,Mercor 从 $250 million Series A 估值,跃升到 February 2025 的 $2 billion,再到 October 2025 的 $10 billion。 独立报道为这个叙事补上真实运营证明,包括到 February 2025 的 $75 million 收入运行率、到 September 2025 约 $450 million 年化收入,以及 OpenAI、Anthropic、Meta 等客户名称。 Mercor 的研究、APEX、Enterprise AI、评估和 RL Studio 页面提供的产品证据很重要,因为它们说明公司试图建设可重复的工作流和基准资产,而不是只做劳动力经纪。 反向逻辑是,Mercor 的定价已经像这场向上游迁移已经成功一样。公开来源称收入按承包商付款前总额口径报告,客户集中度仍高,泄露后续打断了客户信任,诉讼和劳动复杂性仍会增加运营摩擦。 简言之,Mercor 看起来具备战略重要性,但当前估值标记给执行失误留下的空间很小。[CV002, CV003, CV005, CV006, CV008, CV010]

投资逻辑 / 反向逻辑表
投资逻辑支柱支撑反向逻辑什么会改变判断
增长与客户触达2025 年初,Mercor 估值 $2B、收入运行率约 $75M;到 2025 年 9 月,估值升至 $10B、年化收入约 $450M,并引用 OpenAI、Anthropic、Meta 为客户据报道,该收入未扣除承包商付款,且似乎集中在少数实验室披露净收入、抽成率和头部客户集中度
产品上移可选性APEX、Enterprise AI、评估和 RL Studio 显示出向基准测试和工作流基础设施延伸的路径这些产品层的公开采用证据仍薄;它们可能只是销售辅助,而非持久收入流披露产品模块的附加率、重复使用和客户证明
市场顺风AI Index 和市场报告仍支持高质量人类数据工作流需求扩张市场扩大并不能阻止估值倍数压缩;如果 Mercor 更像服务而非软件,压缩仍会发生证明 Mercor 把增长转成更有粘性的经济模型,而不只是放量
竞争位置Scale AI 受扰后,Mercor 位置看起来不错,并拥有强前沿实验室叙事Appen 等同业也在推专家 RLHF、智能体评估和完整性产品;品类趋同会侵蚀差异化证明基准测试真实度和产品深度正在带来收入结构迁移
信任修复泄露补救和诉讼驳回降低了部分头条风险Meta 暂停、集体诉讼,以及任何第二次事件都会迅速重启下行风险提供泄露后控制改进和客户留存的独立证据

反向逻辑并非假设。它直接来自围绕会计、集中度、泄露后续影响和品类趋同的公开证据。

[CV008, CV010, CV011, CV012, CV014, CV015]
FV001: 建议逻辑

从增长证明到收入质量调整、风险检查,再到最终观察建议的决策链。

该流程反映分析师对本章最强估值驱动因素和阻碍因素的综合判断。它是决策框架,不是数学模型。

[CV006, CV008, CV021, CV030, CV034, CV038]

8.2 投资建议与估值立场

建议立场:观察,置信度中,风险高;按当前 $10 billion 标记看,估值偏高。 这个判断刻意保持价格敏感。今天要形成买入逻辑,需要比公开记录更强的证明:经审计或至少更干净的净收入披露,基准和工作流产品正在提升绑定率和利润率的证据,以及泄露后信任补救已经稳定关键客户的迹象。 用当前 $10 billion 对比 September 2025 的 $450 million 总额年化收入运行率,Mercor 看起来约为 22x 总额收入。 即便采用 Mercor 自称 2026 已跨过 $1 billion 年化收入运行率,估值标记仍约为 10x 总额收入;而这个指标未经审计,且包含转付承包商支出。 对一家有软件抱负的公司来说,这并非显然荒谬,但对于一家仍带有市场、合规和事件响应风险的业务而言,价格过高。 实际结论是观察而不是追高:若入场价格降到约 $6 billion 至 $7.5 billion,或出现硬证据证明 Mercor 的基准和工作流层正在改变业务结构,风险回报会更有吸引力。[CV006, CV007, CV008, CV032, CV033, CV039]

建议摘要表
维度评估置信度投资含义
建议观察 — 公司值得关注,但以 $10B 买入的证据 / 价格比不对继续监测;价格或证据改善前,不为入场拉高出价
估值立场$10B 偏高;按当前公开证据,$6B-$7.5B 更站得住当前估值已假设上移产品栈执行成功、经济模型更干净
风险评级高 — 集中度、泄露后续影响、法律 / 劳工复杂性和收入质量不透明未来若入场要谨慎控制仓位,并要求更严尽调
什么会抬升判断经审计 / 净收入桥接、头部客户分散,以及泄露后的信任修复证据即便价格不大幅重置,也可能把立场从观察推向买入
最现实路径等后续私募流动性,或在证据里程碑之后通过老股入场当前估值下,耐心比追势更优

摘要反映的是对证据敏感的观点,而非泛泛的公司质量评分。Mercor 战略上可以很强,但按当前披露仍然太贵。

[CV032, CV033, CV039, CV040, CV041, CV042]
FV002: 估值敏感性

隐含股权价值对不同收入基数和收入倍数的敏感性。数值单位为百万美元。

低位柱把下行锚定在市场型和服务型结果上;高位柱则显示,要让当前估值看起来正常,必须满足哪些条件。收入输入来自公开信号,不是经审计报表。

[CV006, CV007, CV008, CV024, CV025, CV026]

8.3 融资背景与可比估值

Mercor 如今尴尬地夹在两组可比公司之间。一边是 Appen、Upwork、Fiverr 等劳动力市场和数据服务平台,公开市场数据显示这些公司大约按 1x 收入交易。 这类业务被市场当作交易或服务引擎估值,软件稀缺性有限。另一边是 Palantir;它的交易倍数高得多,因为投资者把它视为软件控制平面,具备持久产品锁定、强毛利率和深度关键任务嵌入。 Mercor 当前估值标记显然假设它更接近第二组,而不是第一组,但公开证据还没到位。 支撑溢价的是前沿 AI 客户入口、极端增长,以及围绕 APEX、Enterprise AI、评估和 RL Studio 的可见产品努力。 限制溢价的是缺少公开净收入、抽佣率、利润率或留存数据,加上表观收入数字按承包商付款前总额口径列示。 因此,估值争论不是 Mercor 是不是一家好公司,而是投资者是否应在会计和客户组合证据追上之前,就承销一个软件式未来。[CV008, CV021, CV022, CV023, CV024, CV025]

可比估值表
可比对象状态收入指标倍数 / 估值相关性局限
Mercor私有公司(标的)2025 年 9 月年化总收入约 $450M;公司后来声称 2026 年年化收入达 $1B估值 $10B;按 2025 年 9 月总收入运行率约 22x,按后续公司说法约 10x设定投资人必须承销的入场点总收入与净收入经济模型仍未披露
Appen上市公司收入约 $0.23B约 1x 收入(市值 $0.23B)直接的人类数据与评估可比公司,显示公开市场如何给服务偏重平台定价更成熟、增长更慢的上市公司,客户结构不同
Upwork上市公司收入约 $0.79B约 1.4x 收入(市值 $1.08B)对交易偏重劳动力平台估值有参考意义的市场平台可比对象更宽泛的自由职业市场平台,不是前沿 AI 基础设施
Fiverr上市公司收入约 $0.42B约 0.9x 收入(市值 $0.39B)下行估值框架的另一组市场平台锚SMB 自由职业者定位不同于 Mercor 的专家 AI 细分市场
Palantir上市公司收入约 $5.22B约 63x 收入(市值 $328.14B)显示具备强锁定效应的软件控制平面业务可享受的上行空间产品化、披露和客户嵌入程度都远高于 Mercor
Scale AI私有公司抓取资料中没有干净公开的收入分母根据 Axios/CNBC,Meta 49% 交易隐含价值约 $29BAI 数据基础设施最接近的品类龙头,也提醒叙事可以长期昂贵收入和条款不透明,限制了干净的倍数比较
Turing / 类似人才数据同业私有公司抓取估值来源中可靠公开分母不足Turing 2025 年 3 月估值达到 $2.2B显示 Mercor 的 $10B 估值明显高于相邻人才数据同业公开经济数据和条款细节稀疏

上市公司数字来自本次抓取的 CompaniesMarketCap 快照。私有公司行是估值背景,不是干净的同口径倍数行。

[CV022, CV023, CV024, CV025, CV026, CV027]

8.4 Bull / Base / Bear 情景与回报分析

乐观情景不只是量更大。Mercor 还要把基准测试和工作流资产转成更黏的软件式支出,并把客户面从少数前沿实验室拓宽。按这条叙事,APEX 和 Enterprise AI 会成为真实预算项,RL Studio 和评估产品改善匹配和交付经济性,数据泄露事件在没有第二次事故后淡出,客户集中度也降到足以让客户把 Mercor 看成基础设施,而不是顺手采购的供应商。长期看,这能支撑 $12 billion 到 $18 billion 的结果。基准情景更混合:收入继续增长,Mercor 修复信任,产品资产帮助销售,但客户集中度仍然明显,从总额到净额的质量依旧不透明。那种情况下,估值大概率落在 $7 billion 到 $10 billion 区间,也就是说当前标记已经计入大部分上行。悲观情景则明确叠加已知风险:大客户暂停或流失、数据泄露或集体诉讼后果拖长、劳动力或承包商合规成本上升,或者基准测试和软件层无法在观点输出之外变现。届时 Mercor 会更接近服务和劳动力平台估值,落到低得多的 $2.5 billion 到 $5 billion 区间。[CV011, CV012, CV013, CV017, CV018, CV019]

乐观 / 基准 / 悲观情景表
情景收入 / 结构假设退出估值相对 $10B 的回报含义关键风险 / 支撑概率信号
乐观情景Mercor 证实净收入转化率高,基准 / 工作流产品形成粘性,集中度缓解,泄露后续影响完全正常化$12B-$18B1.2x-1.8x软件上移带来的上行抵消市场平台折价;没有重大法律或安全复发可能,但需要多项证据同时升级
基准情景增长依旧强劲,信任修复足以留住关键账户,但总收入到净收入的不透明和集中度只部分改善$7B-$10B0.7x-1.0x当前价格已经计入该结果的大部分按今天的证据最可能
悲观情景一个头部 AI 实验室关系走弱,泄露或集体诉讼后续影响延续,或劳工 / 法律成本上升,同时基准测试产品未能实质变现$2.5B-$5B0.25x-0.5x即使 AI 需求持续,估值仍向服务 / 劳动力平台倍数重置当前估值留下的安全边际有限,真实下行存在

情景区间是基于公开收入信号和上市公司估值锚的判断,不基于管理层指引或经审计模型。

[CV032, CV033, CV035, CV036, CV037, CV038]
FV003: 估值 / 回报区间

乐观、基准和悲观估值区间,对比当前 $10B 标记和更有吸引力的观察入场区间。

区间基于判断,反映当前估值已经嵌入了多少 Mercor 未来产品组合和信任修复。

[CV035, CV036, CV037, CV039, CV040, CV041]

8.5 估值风险与投资逻辑失效触发点

下行主要由四个风险主导。第一是集中度:公开报道反复显示,收入基础只锚在少数 AI 实验室上,因此一段关系暂停或流失,对估值的冲击会远大于多元化软件公司。第二是信任和安全:Mercor 的 2026 年数据泄露及后续诉讼之所以重要,不只是直接成本,而是它挑战了 Mercor 处理敏感企业流程和模型开发工作流所需的可信度。第三是法律和劳工风险。Mercor 自己的文档显示,它运营着一台全球分布的承包商机器,包含按司法辖区而异的付款限制、工时追踪监督,以及争议人工复核;这套运营复杂性是护城河的一部分,也同样压在风险堆栈里。第四是叙事风险。如果 APEX、Enterprise AI、RL Studio 和评估产品不能转化为软件式绑定,投资人最终可能把 Mercor 归类为高端劳动力与服务市场。即使营收增长仍然不错,倍数也会被压缩。第二次重大安全事件、泄露后明确的客户流失,或法律 / 承包商成本挤压利润率的证据,都会迅速击穿当前投资逻辑。[CV011, CV012, CV013, CV017, CV018, CV020]

投资逻辑破裂与否决触发表
触发因素阈值重要性行动含义
客户集中度向不利方向破裂2026 年后,头部客户明确发生重大暂停、未续约或收入份额流失Mercor 尚无足够强的公开多元化证据,难以吸收大型实验室流失从观察转为回避,或要求大幅更低价格
第二次重大安全事件再次发生严重泄露,或 2026 年事件确认造成更深损害会削弱支撑基准测试、工作流和企业扩张的信任逻辑在控制措施获得外部验证前,视为投资逻辑破裂
劳工 / 承包商成本冲击有证据显示承包商、隐私或分类成本实质改变单位经济模型会把 Mercor 推向更接近服务的经济模型,同时降低增长信心重估至上市劳动力平台倍数
软件附加未显现又一轮融资或增长台阶之后,仍无可信基准 / 工作流收入证据如果没有上移产品栈转化,$10B 估值更多靠规模和叙事,而非产品经济模型即使收入规模仍大,也维持观察立场
下一个定价事件披露仍不透明到下一轮融资或老股窗口时,仍没有净收入桥接、集中度披露或清晰的泄露后控制措施投资人仍需要凭信念承销过多不确定性放弃该轮,或要求更大折扣

否决触发因素用于投资人判断 Mercor 是否配得上软件式溢价,还是应更接近服务和市场平台可比公司的估值。

[CV011, CV012, CV013, CV017, CV018, CV030]
FV004: 投资 KPI

截至 2026 年 5 月,Mercor 估值驱动因素的投委会式快照。

KPI 标签是辅助决策的定性判断,并不代表 Mercor 缺乏优势;它们只是指出,相对于当前价格,公开记录在哪些地方仍显单薄。

[CV021, CV023, CV030, CV031, CV034, CV038]

8.6 退出准备度与最终尽调问题

以这个价格看,Mercor 还不是可以立即拍板的尽调案例。公司最终可能拿到更高倍数,但这个结果取决于公开记录尚未回答的事实。第一项阻断问题是收入质量桥接:把总账单、承包商支出、Mercor 抽成率、净收入,以及任何经常性软件式收入拆开。第二是头部客户视图:投资人需要知道收入基础有多集中、数据泄露影响了多少收入、较新的产品是否在拓宽账户组合。第三,Mercor 要用公开市场或后期私募投资人能纳入承销判断的方式证明信任修复,理想状态是拿出泄露后的控制改进和独立保证,而不只是叙事。第四,承包商和法律堆栈需要更清晰量化,因为一家跨多司法辖区、又严密监控工时的公司,可能比典型软件供应商更容易卷入劳工、隐私或分类争议。在这些问题回答前,最佳姿态是把 Mercor 留在观察名单,而不是支付一个已经预设乐观答案的价格。[CV017, CV018, CV030, CV031, CV039, CV040]

最终尽调问题表
主题缺失证据重要性负责人 / 尽调路径
净收入和抽成率总账单额、承包商付款、Mercor 抽成率,以及任何经常性软件收入拆分这是市场平台估值与软件估值之间最关键的一座桥财务数据室和财务主管审阅
Top-10 客户结构和留存收入集中度、续约节奏、泄露后暂停账户,以及按产品线拆分的扩张集中度是当前模型里的核心下行变量CFO 数据室加客户访谈
泄露后信任修复LiteLLM 事件后,控制改进、影响范围和客户安抚的独立证据没有信任修复,基准测试和企业增购更难承销安全尽调、事件报告和客户核查
承包商和法律敞口分司法辖区的承包商结构、政策执行数据、争议,以及任何准备金或外部律师备忘录Mercor 的护城河部分靠劳动力编排;这也会带来合规成本劳动和隐私律师参与法律尽调
股权结构表和优先权优先股堆叠、参与权,以及 $10B 估值下任何老股条款如果条款异常干净,偏高估值仍可能可投;反过来也一样公司律师审阅和融资文件分析

这些问题按对估值信心的直接影响排序。在当前价格下,没有一项是“之后再补”的锦上添花。

[CV031, CV039, CV040, CV041, CV042]

免责声明

本报告元信息摘要基于截至 2026 年 5 月 23 日的公开来源生成,不构成投资建议。Mercor 是一家私营公司,若干最关键的投资测算输入项——包括净收入、利润率、客户集中度和优先权条款——并未公开披露。任何投资决定都应依赖直接尽调和公司一手材料,而不是只依赖这份公开信息摘要。

证据索引

结论
编号陈述可信度来源
CO001 Mercor describes itself as organizing human intelligence to power the AI economy. SO001
CO002 Mercor connects experts to AI projects and pays them remotely on contract engagements. SO002, SO001
CO003 Mercor says its work sits at the intersection of labor markets and AI research. SO007, SO003
CO004 Mercor was founded in January 2023. SO005, SO021
CO005 Official and independent 2025 coverage identifies Mercor's cofounders as Brendan Foody, Adarsh Hiremath, and Surya Midha. SO008, SO021, SO022
CO006 Mercor's founders dropped out of Georgetown and Harvard in 2023 to build the company. SO008, SO021, SO025
CO007 PR Newswire said all three cofounders had received the Thiel Fellowship by the time of Mercor's Series A announcement. SO008
CO008 Mercor raised a $3.6 million seed round led by General Catalyst in 2023. SO005, SO021
CO009 Mercor announced a $30 million Series A at a $250 million valuation in 2024. SO008, SO011
CO010 Mercor announced a $100 million Series B at a $2 billion valuation in February 2025. SO009, SO010, SO011, SO006
CO011 Mercor announced a $350 million Series C at a $10 billion valuation in October 2025. SO007, SO013, SO014
CO012 Mercor's total disclosed primary capital across seed, Series A, Series B, and Series C is about $483.6 million. SO005, SO008, SO009, SO013
CO013 Bloomberg said Mercor's February 2025 round included Felicis, General Catalyst, DST Global, Benchmark, and Menlo Ventures. SO011, SO009
CO014 Mercor says its Series C was led by Felicis with Benchmark, General Catalyst, and Robinhood Ventures participating. SO007, SO013, SO014
CO015 Mercor's business model shifted from AI-driven recruiting toward supplying domain experts for AI model training and evaluation. SO009, SO013, SO015
CO016 TechCrunch reported Mercor generated revenue by charging hourly finders' fees and matching rates to clients. SO009, SO012, SO013
CO017 CNBC reported Mercor had processed 300,000 candidates and conducted more than 100,000 interviews by February 2025. SO010
CO018 TechCrunch reported Mercor had helped HR teams evaluate 468,000 applicants by February 2025. SO009
CO019 TechCrunch reported Mercor reached a $75 million annual recurring revenue run rate by February 2025. SO009
CO020 CNBC quoted Brendan Foody saying Mercor had grown more than 51% month over month over the prior six months as of February 2025. SO010, SO006
CO021 TechCrunch reported Mercor was approaching a $450 million annualized revenue run rate in September 2025. SO012
CO022 TechCrunch said Mercor generated $6 million in profit in the first half of 2025, citing Forbes. SO012
CO023 Mercor said in October 2025 that it paid more than $1.5 million per day to contractors and had more than 30,000 experts on its roster. SO007, SO013, SO014
CO024 TechCrunch reported Mercor's experts earned more than $85 per hour on average in October 2025. SO013
CO025 TechCrunch reported Mercor paid some industry experts as much as $200 per hour for AI training work. SO015
CO026 Mercor's research page says the company is used by the top five AI labs and six of the Magnificent Seven. SO003
CO027 TechCrunch reported Mercor supplied contractors to Amazon, Google, Meta, Microsoft, OpenAI, and Nvidia. SO012
CO028 TechCrunch Disrupt coverage named OpenAI, Anthropic, and Meta as Mercor customers. SO015
CO029 Mercor's careers page says the company is a profitable Series C company with offices in San Francisco, New York, and London. SO004
CO030 Mercor's careers page listed 58 open roles across enterprise, engineering, operations, finance, and marketing when fetched for this report. SO004
CO031 Mercor said its team included the former Head of Human Data Operations at OpenAI and the previous Head of Growth at Scale by February 2025. SO006
CO032 TechCrunch reported Mercor appointed former Uber chief product officer Sundeep Jain as its first president in 2025. SO012
CO033 Forbes' Adarsh Hiremath profile said Surya Midha transitioned from chief operating officer to chairman in October 2025. SO022
CO034 Scale AI sued Mercor.io Corporation and former Scale employee Eugene Ling on September 3, 2025 over alleged trade-secret misappropriation. SO016, SO018, SO019
CO035 PacerMonitor shows the Scale AI lawsuit was voluntarily dismissed with prejudice in early January 2026. SO020
CO036 TechCrunch reported Mercor disclosed a March 2026 data breach linked to credential-harvesting malware in the open-source tool LiteLLM. SO017
CO037 TechCrunch reported Meta paused its contracts with Mercor after the breach while OpenAI said it was investigating but had not paused work at the time. SO017
CO038 KTVU quoted Brendan Foody saying Mercor had crossed a $100 million revenue run rate by March 2025 and was extremely profitable. SO021
CM001 Mercor operates in a narrower market than generic staffing: high-skill human-in-the-loop AI training, evaluation, and benchmark work. SM001, SM002, SM004
CM002 Mercor's market includes expert post-training labor for doctors, lawyers, engineers, bankers, and consultants rather than commodity click-work. SM003, SM005
CM003 The relevant adjacent market includes benchmark and evaluation-environment creation for frontier models and agents. SM001, SM016, SM017
CM004 Generic ATS, HRIS, and employer recruiting software are substitutes only for Mercor's original recruiting product, not for its current AI-training specialization. SM002, SM004
CM005 Low-skill crowd labeling is an adjacent but different category from Mercor's premium expert marketplace. SM021, SM022, SM003
CM006 MarketsandMarkets projected the global data annotation and labeling market to reach $3.6 billion by 2027 at a 33.2% CAGR. SM008
CM007 The same MarketsandMarkets preview projected the AI training dataset market to reach $9.58 billion by 2029 at a 27.7% CAGR. SM008
CM008 IBM's summary of Stanford HAI said total corporate AI investment reached $252.3 billion in 2024. SM006, SM007
CM009 IBM's Stanford AI Index summary said U.S. private AI investment reached $109.1 billion in 2024. SM006, SM007
CM010 IBM's Stanford AI Index summary said the number of newly funded generative AI startups nearly tripled in 2024. SM007
CM011 A top-down TAM from broad AI investment materially overstates Mercor's serviceable market because most AI spending is not spent on expert contractors. SM006, SM008, SM004
CM012 The primary buyers in Mercor's market are frontier AI labs and enterprise AI teams commissioning post-training data, evaluations, or benchmark work. SM004, SM005, SM001
CM013 The user inside the buyer organization is typically a model-training, evals, or research operations team rather than an HR department. SM004, SM005, SM013
CM014 The payer for Mercor-style services is usually an AI lab or enterprise AI budget owner rather than the individual expert. SM002, SM015
CM015 Experts on the supply side are both labor inputs and repositories of domain knowledge, making the supply base strategically important. SM003, SM009, SM021
CM016 Incumbent employers whose workflows are being encoded into models act as a status-quo substitute and a blocking constituency in the market. SM005, SM021
CM017 OpenAI's InstructGPT work established that reinforcement learning from human feedback depends on human rankings and preference data. SM010
CM018 Anthropic's Constitutional AI paper reduced some human-label requirements but still framed alignment and evaluation as feedback-intensive. SM012
CM019 Mercor's research page positions benchmark creation and evaluation environments as a frontier need beyond simple annotation. SM001
CM020 Labelbox's expert-economy report argues that frontier models increasingly need PhDs, clinicians, and high-skill specialists rather than general annotators. SM009
CM021 Meta's investment in Scale AI destabilized vendor neutrality and reopened demand for alternative post-training vendors. SM004, SM005
CM022 Agentic AI increases demand for evaluation environments that test multi-step reasoning and real-world workflows. SM001, SM016, SM017
CM023 Data-rights and trade-secret sensitivity constrain how much real enterprise workflow data buyers are willing to share with AI labs. SM005, SM004
CM024 Labor-rights scrutiny in AI data work creates a compliance and brand constraint on scaling contractor-heavy models. SM021, SM022, SM023
CM025 Mercor's market lies between AI infrastructure and flexible labor marketplaces, which complicates direct comparable selection. SM001, SM004, SM019
CM026 Because the market is concentrated among a few frontier labs, adoption can accelerate quickly but also pause suddenly if one buyer changes strategy. SM004, SM005
CM027 NIST's AI Risk Management Framework supports demand for auditable evaluations and trustworthy post-training processes among enterprise buyers. SM024
CM028 OpenAI's 2024 custom-models update implies continued enterprise willingness to buy specialized training and tuning work around frontier models. SM011
CM029 Appen, Scale, Labelbox, iMerit, and Toloka all market adjacent services, confirming that the market boundary spans both services and platform tooling. SM013, SM014, SM015, SM020, SM025
CM030 Snorkel and automation-focused vendors show that rote labeling spend may shift toward software-assisted data generation over time. SM016, SM018
CM031 Invisible Technologies illustrates an adjacent market where enterprises buy modular combinations of data, agents, and humans-in-the-loop rather than a pure labor marketplace. SM019
CM032 Mercor's serviceable market is likely much closer to the high-skill post-training and evals niche than to the full AI-investment pool. SM006, SM008, SM009
CM033 A durable buyer budget requires proof that expert labor improves model quality or accelerates deployment enough to justify premium rates. SM010, SM012, SM017
CM034 Mercor's market benefits from AI adoption growth, but procurement must still clear security, IP, and trust objections in regulated industries. SM024, SM021, SM023
CM035 Public sources do not disclose how much frontier labs spend specifically on benchmark creation, expert evaluations, or contractor pass-through versus other AI infrastructure.
CM036 Public sources also do not disclose how much of Mercor's opportunity lies in recurring enterprise workflows versus one-off frontier-lab projects.
CP001 Mercor competes most directly with Scale AI, Surge AI, Labelbox, Appen, iMerit, and other human-data vendors serving frontier models. SP003, SP014, SP005, SP008, SP012
CP002 Scale AI remains the best-known incumbent in human-data infrastructure and RLHF among frontier-model buyers. SP003, SP004, SP021
CP003 Surge AI is a premium RLHF-focused competitor with a similarly high-skill positioning to Mercor. SP014, SP020
CP004 Labelbox competes as a full-stack data factory combining workflows, RLHF, and an expert network rather than only a labor marketplace. SP005, SP006, SP007
CP005 Appen competes from the opposite end of the market: a large public human-data vendor trying to move upmarket into frontier alignment and agentic AI services. SP008, SP010, SP011
CP006 iMerit and CloudFactory compete through managed-service human-in-the-loop delivery rather than Mercor's marketplace-led positioning. SP012, SP013
CP007 Snorkel acts as a substitute class by pushing programmatic and automation-heavy data generation instead of expert marketplace labor. SP015, SP016
CP008 Mercor's clearest differentiation is explicit concentration on domain experts such as doctors, lawyers, bankers, and engineers. SP002, SP025
CP009 Scale markets enterprise-grade AI systems and RLHF, but its brand is broader infrastructure rather than a pure expert-talent marketplace. SP003, SP004
CP010 Labelbox positions on software workflow control and data-factory automation more than on Mercor-style labor aggregation. SP005, SP006
CP011 Appen competes on scale, breadth, and public-company credibility rather than Mercor's frontier-startup speed narrative. SP008, SP009
CP012 Mercor, Scale, Labelbox, and Appen all now market evaluation or alignment services, showing convergence around post-training workflows. SP001, SP004, SP006, SP010, SP024
CP013 Mercor has less evidence of platform lock-in than software-first competitors because its core value still depends on ongoing labor-market coordination. SP002, SP005, SP018
CP014 SuperAnnotate and Toloka illustrate how buyers can still choose broad annotation platforms instead of premium expert marketplaces. SP017, SP019, SP023
CP015 Invisible illustrates an adjacent alternative where enterprises buy combined agents, data, and human operations rather than a specialist RLHF vendor. SP018
CP016 Mercor's pricing is not publicly listed; the most visible economic signal is expert hourly rates and client-specific matching fees. SP025, SP020
CP017 Scale, Labelbox, and Appen similarly avoid transparent public list pricing for most enterprise RLHF and evaluation work. SP004, SP006, SP010
CP018 A lack of public pricing across the category makes sales execution, neutrality, speed, and trust more important than headline list prices. SP004, SP006, SP010, SP025
CP019 Mercor benefited competitively when Meta's investment in Scale AI raised neutrality concerns among large labs. SP020, SP022, SP026
CP020 CNBC reported OpenAI had been winding down work with Scale AI and that Google was also reportedly cutting ties after the Meta deal. SP022
CP021 Meta's $14.3 billion investment gave Scale AI a roughly $29 billion implied valuation and kept it substantially larger than Mercor. SP021
CP022 Despite Mercor's growth, TechCrunch still described Surge and Scale AI as larger competitors by late 2025. SP025
CP023 The main moat candidate for Mercor is speed in sourcing premium experts and converting that supply into frontier-model improvement workflows. SP002, SP025
CP024 That moat is fragile because experts can multi-home across vendors and buyers can test several providers simultaneously. SP014, SP019, SP022
CP025 Software-centric competitors may develop stronger lock-in through integrated data, model-evaluation workflows, and analytics than Mercor can through matching alone. SP005, SP006, SP015, SP016
CP026 Mercor's benchmark and evaluation products are an attempt to move from marketplace coordination toward higher-sticky workflow ownership. SP001, SP020
CP027 Appen's investor materials show a public incumbent with broad lifecycle positioning, which can appeal to enterprise buyers who prefer scale and governance over startup speed. SP009
CP028 Snorkel's automation-centric workflow is an adverse signal for any vendor whose value depends on repeating human-labor tasks rather than capturing harder expert judgment. SP016
CP029 Mercor still has a cleaner neutrality narrative than Scale AI after the Meta deal, but that advantage could fade if Mercor itself becomes concentrated with a few labs. SP022, SP025
CP030 Public market data does not reveal realized win rates, pricing discounts, or retention differences across these vendors.
CP031 Mercor is strongest where buyers value expert judgment and faster supply mobilization more than deep platform workflow control. SP002, SP025, SP005
CP032 Mercor is weaker where buyers prioritize software governance, entrenched workflows, or broad installed bases over marketplace speed. SP005, SP009, SP018
CP033 The category remains structurally multi-homed because no single vendor appears to own both the labor supply and the full workflow stack. SP004, SP006, SP010, SP018
CP034 Mercor's category leadership case depends on moving up the stack before software-centric rivals commoditize matching and sourcing. SP001, SP016, SP018
CP035 The biggest competitive unknown is whether Mercor can turn benchmark and eval workflows into genuine product lock-in.
CI001 Mercor monetizes customer demand for expert work by matching specialists to AI-lab and enterprise projects, then administering payment through its platform. SI001, SI002, SI007
CI002 Mercor's experts page says professionals work remotely on contract opportunities and get paid weekly. SI002, SI007
CI003 Mercor's payments documentation says Stripe is the primary payment rail for ongoing work. SI007
CI004 Mercor's payments documentation says Wise is sometimes used for one-time or fallback payouts. SI007
CI005 Mercor announced a $30 million Series A at a $250 million valuation in 2024. SI008, SI011
CI006 Mercor announced a $100 million Series B at a $2 billion valuation in February 2025. SI004, SI009, SI010, SI011
CI007 Mercor announced a $350 million Series C at a $10 billion valuation in October 2025. SI005, SI013
CI008 Mercor's total disclosed primary capital across seed, Series A, Series B, and Series C is about $483.6 million. SI008, SI010, SI005
CI009 CNBC reported Mercor was profitable and running above a $75 million revenue run rate by February 2025. SI009
CI010 TechCrunch reported Mercor had reached a $75 million annual recurring revenue run rate by February 2025. SI010
CI011 TechCrunch reported Mercor was approaching a $450 million annualized revenue run rate in September 2025. SI012
CI012 TechCrunch reported Mercor told investors it was on track to hit $500 million ARR faster than Anysphere. SI012
CI013 TechCrunch reported Mercor generated $6 million of profit in the first half of 2025, citing Forbes. SI012
CI014 KTVU quoted Brendan Foody saying Mercor had crossed a $100 million revenue run rate by March 2025 and was extremely profitable. SI014
CI015 Mercor's 2026 payments-systems engineering post said the company crossed a $1 billion annualized revenue run rate earlier in 2026. SI006, SI015
CI016 The same 2026 post said Mercor was paying out more than $2 million each day to more than 30,000 weekly active contractors. SI006
CI017 By October 2025, Mercor said it had more than 30,000 contractors and was paying over $1.5 million per day to them. SI005, SI013
CI018 TechCrunch reported Mercor's headline revenue includes the full amount customers pay before contractors receive their portion. SI012
CI019 TechCrunch said Mercor management framed that gross presentation as common among peers such as Surge AI and Scale AI. SI012
CI020 TechCrunch reported Mercor earns money through an hourly finder's fee and matching rate layered onto expert work. SI012
CI021 TechCrunch reported some Mercor experts earned as much as $200 per hour for AI training work. SI016
CI022 Mercor's fetched homepage showed finance or investor-relations experts at $80-$160 per hour and equity research experts at $120 per hour. SI001
CI023 Mercor's careers page describes the company as a profitable Series C company and, when fetched, showed six finance roles alongside 32 engineering roles. SI003
CI024 Mercor said its Series C capital would expand the talent network, improve matching, and speed delivery. SI005, SI013
CI025 Mercor said its Series B capital would accelerate its ability to match billions of people with their calling. SI004
CI026 TechCrunch reported an outsized share of Mercor's revenue came from a subset of major brands including OpenAI, indicating concentration risk. SI012
CI027 TechCrunch's April 2026 breach coverage said paused Meta contracts and customer reviews could put meaningful revenue at risk. SI017
CI028 Scale AI's September 2025 lawsuit described one customer opportunity as a contract worth millions of dollars to Mercor. SI018
CI029 Court records show Scale AI filed suit against Mercor on September 3, 2025 and later voluntarily dismissed the case with prejudice by early January 2026. SI018, SI019, SI020
CI030 Appen's investor-relations page shows public incumbents in this category publish full-year and half-year results, unlike Mercor. SI023
CI031 Appen publicly describes itself as serving AI lifecycle work with a global crowd of over 1 million contributors and real-world model evaluation. SI023, SI024
CI032 Appen's model-evaluation page shows hallucination benchmarking, regulatory audits, and continuous monitoring are monetizable service lines in this market. SI024
CI033 California's AB 5 text underscores that contractor-heavy marketplaces still face worker-classification compliance risk. SI025
CI034 CNBC reported OpenAI and Google were pulling back from Scale AI after Meta's investment, creating a near-term demand-dislocation opportunity for alternatives. SI021, SI027
CI035 CNBC reported Scale AI later cut 14% of its workforce while trying to win back customers that had slowed work, highlighting category volatility. SI022
CI036 Mercor's payments-systems post says its infrastructure must invoice clients across multiple complex billing structures and contractual terms while paying contractors globally. SI006, SI007
CI037 That same engineering post says hypergrowth exposed gaps in data models and controls, forcing more investment in financial operations and correctness. SI006
CI038 Public AI-spending and annotation-market proxies show the broader category remains large, but those top-down figures do not reveal Mercor's take rate, burn, or cash conversion. SI028, SI029
CI039 Mercor's public blog now spans company, research, stories, and engineering categories, consistent with management publicly treating payments and controls as scaling priorities rather than back-office details. SI026, SI006
CE001 Mercor says it develops benchmarks, evaluation environments, and large-scale human datasets through a marketplace of top-tier experts. SE003, SE007
CE002 Mercor's research page says the company is used by the top five AI labs and six of the Magnificent Seven. SE003
CE003 Mercor's experts page presents the product as remote, high-paying expert work that advances AI systems. SE001, SE002
CE004 Mercor's careers page says every team works directly with frontier models. SE004
CE005 The careers page listed research-engineering roles focused on environments, data and post-training as well as benchmarking, evals, and failure analysis. SE004
CE006 The same careers page listed infrastructure, payments, security, application-security, automation, cloud-infrastructure, site-reliability, and agents roles. SE004
CE007 Mercor said its Series C capital would expand the talent network, improve matching, and speed delivery. SE006, SE033
CE008 Mercor's Series B post said the team included the former Head of Human Data Operations at OpenAI and the previous Head of Growth at Scale. SE005
CE009 Mercor's APEX family now spans APEX, APEX-Agents, APEX-SWE, and ACE. SE003, SE007
CE010 Mercor's APEX-SWE post says the benchmark was created with Cognition to test real software engineering work rather than narrow coding tasks. SE008, SE009
CE011 Mercor's APEX-SWE leaderboard says the benchmark contains 200 cases split between integration and observability tasks. SE009
CE012 The same leaderboard says each task has a human-authored rubric grading functional requirements, robustness, and code style. SE009
CE013 Mercor says it open-sourced 50 in-distribution APEX-SWE cases plus the evaluation harness. SE009
CE014 At release, Mercor reported GPT-5.3 Codex as the top APEX-SWE model at 41.5% Pass@1. SE008, SE009
CE015 Mercor's APEX-SWE post says developers spend only 16% of their time writing code and 84% on CI/CD, infrastructure, deployment, and debugging. SE008
CE016 Mercor's APEX-Agents post says the benchmark tests long-horizon, cross-application tasks in investment banking, consulting, and corporate law. SE010
CE017 Mercor said the APEX-Agents design began with surveys of hundreds of experts from firms including Goldman Sachs, McKinsey, and Cravath. SE010
CE018 Mercor's expanded APEX post says the heldout evaluation set doubled from 200 to 400 cases. SE011
CE019 The same post says APEX tasks take more than two and a half hours on average for seasoned professionals and contributors typically had over seven years of experience. SE011
CE020 Mercor's Enterprise AI post says many enterprise agent projects stall because teams guess the use case, hand-write prompts, and lack evidence of real workflow value. SE012
CE021 Mercor's RL-environment post argues that academic evals saturate and economically valuable work increasingly requires richer real-world environments and tools. SE015
CE022 Mercor's Monty engineering post says someone starts an interview every nine seconds, creating roughly 10,000 conversations a day lasting about 15 minutes each. SE013
CE023 The Monty post says each interview session runs in its own container on Modal. SE013
CE024 The same post says Mercor keeps about 30 compute-prebooted containers and about 10 fully initialized interview environments, allowing starts in well under 200 milliseconds. SE013
CE025 Mercor's Contracts-service post says a critical service rewrite made the system more than 10,000 times more capable and over 75 times more reliable. SE014
CE026 That post says the old Contracts system had been tuned for about 3,000 active contracts a month, 20 to 50 concurrent requests, and roughly 100-second completions. SE014
CE027 Mercor's data-and-AI policy says the platform collects résumés, interview audio and video, AI transcripts, public profile data, and payment or tax details. SE017
CE028 The same policy says Mercor uses that data for matching, interviews, payouts, compliance, and communication. SE017
CE029 Mercor's LLM Usage Policy prohibits contractors from using LLMs to assess model outputs or predict code behavior. SE018
CE030 Mercor's background-check policy says it verifies identity, education, employment history, and relevant licenses or certifications. SE019
CE031 Mercor's time-tracking guide says project workers use the Workpuls or Insightful desktop tool to record task time. SE020
CE032 Mercor's payments guide says Stripe is the primary payment rail and Wise is sometimes used for one-time or fallback payments. SE021
CE033 OpenAI's InstructGPT work is direct technical evidence that human feedback remains foundational to aligning capable models. SE022
CE034 OpenAI's custom-model and fine-tuning announcement shows enterprises are buying tailored model-training workflows rather than only raw API access. SE023
CE035 Anthropic's Constitutional AI paper shows AI-generated feedback can automate part of alignment, but still depends on carefully designed oversight and objectives. SE024
CE036 Appen's Frontier Alignment page says domain-expert RLHF now spans medicine, law, science, finance, preference ranking, and multi-turn evaluation. SE025
CE037 Appen's agentic-AI page markets golden trajectories, RL environment design, failure taxonomies, and SWE-driven deep evaluation workflows. SE026
CE038 Appen's model-evaluation page markets hallucination benchmarking, regulatory audits, continuous monitoring, and LLM-as-a-judge rubric design. SE027
CE039 iMerit's RLHF tooling overview says automation platforms exist to address human-labeling bottlenecks, reward-model complexity, and safety-compliance issues. SE028
CE040 Toloka markets context-rich simulated environments, RL gyms with MCP replicas, computer-use testbeds, and expert-captured workflows for AI agents. SE029
CE041 Scale AI's RLHF page shows incumbent competitors also sell expert human-feedback workflows, which limits differentiation from website copy alone. SE030
CE042 TechCrunch reported Mercor handles custom datasets and processes that AI model makers consider trade-secret-sensitive. SE031
CE043 TechCrunch's April 2026 breach story said attackers claimed access to source code, API keys, candidate data, and employer data from Mercor's systems. SE032
CE044 Taken together, Mercor's careers page, engineering posts, benchmark pages, and docs imply the product surface now spans marketplace matching, AI interviewing, benchmark creation, enterprise agent design, payouts, and trust or compliance operations. SE004, SE007, SE012, SE013, SE014, SE016, SE017, SE020, SE021
CU001 Mercor reported $450 million in annualized revenue in September 2025, up from $2 million daily earlier that year. SU003, SU011
CU002 Mercor closed a $350 million Series C at a $10 billion valuation in October 2025. SU004, SU007, SU010
CU003 Mercors primary customers are AI labs and technology companies that need large-scale training-data annotation and evaluation. SU005, SU012, SU017
CU004 Scale AI was Mercors direct predecessor in serving major AI labs including OpenAI and Google, and those customers subsequently reduced Scale AI work. SU008, SU009
CU005 Mercor launched an Enterprise AI product in early 2025 targeting large organizations that want AI-assisted hiring and workforce solutions. SU018
CU006 Mercor grew revenue roughly 30x in 2025, going from approximately $2 million per month to $2 million per day. SU016, SU003
CU007 The Series A in February 2024 raised $34 million and was used to deepen AI lab customer relationships. SU001, SU029
CU008 Bloomberg described Mercor as the default sourcing partner for AI labs building training datasets as of April 2026. SU012
CU009 Public sources show two separate scale signals: CNBC said Mercor had processed 300,000 candidates by February 2025, and Mercor later said it had more than 30,000 experts on its roster by October 2025. SU006, SU014
CU010 The talent portal at talent.docs.mercor.com documents project onboarding flows, suggesting structured customer-facing deployment processes. SU023, SU024
CU011 Scale AI sued Mercor in September 2025 alleging trade secret misappropriation; this signals direct competition for the same AI-lab customer base. SU027
CU012 Mercor appeared on Forbes AI Cloud 100 in 2025, reflecting recognition of its customer base quality among AI-sector analysts. SU011
CU013 Mercor offered AI researchers the ability to test models with domain-expert evaluators as part of its Experts product line. SU020, SU005
CU014 At TechCrunch Disrupt 2024 Mercor demonstrated live AI evaluation workflows, showcasing its customer-facing capabilities. SU017
CU015 Mercors early customers were startups and mid-size AI companies; the customer base has since expanded to include top-tier frontier AI labs. SU019, SU005
CU016 Mercors Series B in February 2025 valued the company at $2 billion, with investor confidence driven by AI-lab customer traction. SU002, SU006
CU017 Rest of World reported that data annotation workers often struggled to meet quality requirements, pointing to supply-side retention challenges. SU026
CU018 Mercors worker onboarding documentation indicates structured project ramp-up periods and milestone-based access to new projects. SU024
CU019 KTVU reported Mercors founding story emphasizing direct outreach to AI labs as the initial customer acquisition strategy. SU025
CU020 Mercors Research portal lists open-domain AI research evaluation as a customer-facing service, indicating diversification beyond annotation. SU022
CU021 A reported $450M ARR run rate in September 2025 implies concentration risk if even one or two top-10 customers reduce spend. SU003, SU010
CU022 OpenAI and Google both reduced spend with Scale AI within months of Mercors rapid growth, suggesting platform-switching risk exists at scale. SU008, SU009
CU023 No public churn rate, net revenue retention, or cohort data has been disclosed for Mercors AI-lab customer segment. SU003, SU012
CU024 Mercors enterprise product announcement in 2025 suggests the company is attempting to diversify beyond annotation into broader workforce management. SU018, SU022
CU025 The talent portal documentation suggests Mercor uses contractual milestone gates to control project access, a structural retention mechanism for workers. SU023, SU024
CU026 Mercors revenue per worker is not publicly disclosed, making it impossible to assess expansion revenue dynamics from existing accounts. SU014, SU030
CU027 Mercor launched Apex, a premium software-engineering evaluation product that benchmarks AI coding models using human expert assessors. SU031, SU032
CU028 The Apex SWE leaderboard publicly ranks AI coding models evaluated on real tasks by Mercor experts, functioning as a customer-facing proof of methodology rigor. SU032
CU029 Mercors Series C raised $350 million in new capital based on TechCrunch reporting of the round and pre- and post-money valuations. SU004, SU030
CU030 Turing AI, a comparable crowdwork and AI annotation platform, was valued at $2.2 billion in March 2025, roughly 20% of Mercors October 2025 valuation, indicating investor premium for Mercors scale and customer quality. SU033
CU031 Mercors company-disclosed blog post describes the revenue trajectory as going from $2M per month in early 2025 to $2M per day later in the year, a primary-source corroboration of the $450M ARR figure. SU016
CU032 Mercors Series C blog post describes the round as driven by customer momentum and demand from AI labs, confirming that customer growth was the primary raise catalyst. SU030
CU033 Forbes profiles of the Mercor founders note direct relationships with AI lab procurement teams, indicating a high-touch enterprise sales motion from inception. SU010, SU011
CU034 The Apex leaderboard evaluation data is produced from actual AI lab customer projects submitted for benchmarking, serving as indirect customer-proof evidence that frontier labs are active platform users. SU031, SU032
CU035 Mercors homepage documents multiple distinct product lines — Annotation, Evaluation, Experts, Apex, and Enterprise — confirming a multi-product customer engagement strategy targeting different buyer segments. SU021, SU020
CR001 The US Department of Labors 2024 independent contractor rule tightens the economic-reality test, increasing reclassification risk for platforms using gig workers. SR001, SR002
CR002 Californias AB 5 applies the ABC test to worker classification; Mercors annotator workforce likely faces scrutiny under this law if operating in California. SR003, SR004, SR005
CR003 The California Supreme Courts 2024 clarification of AB 5 scope in a major trucking case signals continued judicial willingness to expand gig-worker protections. SR007
CR004 The EU AI Act (2024) imposes obligations on providers of AI systems used in employment contexts; Mercors AI-assisted matching tools may fall within scope. SR010
CR005 Scale AI filed a trade secrets lawsuit against Mercor and a former employee in September 2025, alleging misappropriation of proprietary customer and pricing data. SR015, SR016, SR017
CR006 A class-action lawsuit was filed against Mercor in April 2026 alleging negligent data security practices following a confirmed cyberattack that exposed user personal data. SR011, SR012, SR013
CR007 TechCrunch confirmed in March 2026 that Mercor suffered a cyberattack that exposed personal data of some users; the company disclosed the incident publicly. SR014, SR021
CR008 Mercor maintains a Trust Center at trust.mercor.com, indicating some level of security-posture documentation and compliance program existence. SR026
CR009 NIST Cybersecurity Framework version 2 (2024) establishes best-practice controls for organizations handling sensitive personal data; Mercor has not disclosed conformance. SR008
CR010 Mercors talent portal contract policy documents indicate workers are engaged as independent contractors under written service agreements. SR028, SR029
CR011 Mercors tax and work-authorization policy requires workers to self-certify eligibility; this shifts classification and tax risk to workers rather than the platform. SR027
CR012 Scale AIs reduction in workforce by 14% following loss of OpenAI and Google contracts illustrates how customer concentration can cause rapid organizational stress. SR024, SR018
CR013 Rest of World documented quality-control challenges among AI annotation workers broadly, suggesting systematic quality risk across the annotation industry. SR022, SR023
CR014 Mercors blog post about handling 10x volume growth in one week reveals operational scaling risks and the absence of pre-built capacity buffers. SR031
CR015 Mercor has not disclosed whether it carries cyber liability insurance, errors-and-omissions coverage, or workers compensation insurance for its contractor base. SR026, SR030
CR016 The workers compensation implications of AB 5 are specifically addressed by Californias DIR; Mercors annotators may qualify for coverage under certain interpretations. SR006, SR004
CR017 Time magazine documented wage and working-condition concerns among AI data annotators working for Scale AI in India, raising analogous questions for Mercors global workforce. SR032
CR018 No public SOC 2 report, ISO 27001 certification, or third-party security audit has been published for Mercor; the Trust Center does not disclose certifications. SR026
CR019 CNBC reported Scale AIs founder departure in June 2025; this destabilization of the largest competitor creates both opportunity and execution risk for Mercor. SR025
CR020 Mercors Data and AI Usage policy at talent.docs.mercor.com indicates that annotator-produced data is owned by the customer, not the worker — a key IP and liability structure. SR030
CR021 The CourtListener docket for the 2026 class action shows the case was filed in the Northern District of California and remains active as of May 2026. SR012, SR013
CR022 TechCrunch noted in April 2026 that the data breach and Scale AI litigation arriving in the same month created compounding reputational risk for Mercor. SR021
CR023 The IRS worker-classification guidance requires multi-factor analysis; Mercors reliance on worker self-certification may not insulate it from federal reclassification. SR002
CR024 Mercors legal support documentation at talent.docs.mercor.com provides a dispute resolution pathway for workers, suggesting awareness of contractor-relation legal exposure. SR029
CR025 Worker misclassification penalties under California law can include back wages, benefits, and penalties; against a publicly disclosed roster of more than 30,000 experts, the exposure could still be material. SR003, SR004, SR006
CR026 Reuters reported that OpenAI wound down its Scale AI work in June 2025; Scale AIs subsequent 14% headcount reduction illustrates how a single customer decision can affect a platform at Mercors scale. SR018, SR024
CR027 Mercors PACER docket for the Scale AI trade-secret case shows ongoing discovery activity as of early 2026, indicating the litigation is not close to resolution. SR019, SR020
CR028 The EU AI Act risk classification for AI-assisted employment matching is likely high-risk, requiring conformity assessment before market deployment in the EU. SR010
CR029 Rest of World and Time reporting on annotation worker conditions suggest Mercor faces reputational risk from association with below-market pay for global contractors. SR022, SR032
CR030 Mercors Trust Center existence indicates basic security governance, but the absence of disclosed certifications leaves material uncertainty about actual security controls. SR026, SR008
CR031 The California AB 5 taxes and work-authorization FAQ from FTB specifically addresses multi-state workers, directly relevant to Mercors cross-state contractor base. SR003, SR027
CR032 Mercors rapid scaling post documents that the platform faced queue failures and worker-matching errors during a 10x volume spike, revealing infrastructure fragility. SR031
CR033 Mercor raised a $350M Series C in October 2025 at a $10B valuation; no burn rate, annual OpEx, or runway figure has been disclosed, creating opacity around financial model risk. SR033
CR034 TechCrunch reported a $450M annualized revenue run rate in September 2025, implying a 22x revenue multiple at the $10B Series C valuation — highly sensitive to any revenue deceleration. SR034, SR035
CR035 Mercors business model depends on sustained enterprise AI training budgets; a slowdown in AI capital expenditure by frontier labs would directly reduce demand for annotation services. SR034, SR033
CR036 A 22x revenue multiple creates significant valuation compression risk; even modest revenue deceleration could reset the valuation anchor and complicate future fundraising. SR033, SR034
CR037 Global contractor payroll at scale for a publicly disclosed roster of more than 30,000 experts creates working-capital demands and cross-border payment risks including FX volatility, sanctions exposure, and payment-rails failure. SR027, SR028
CR038 Legal defense costs for two simultaneous cases (Scale AI trade secrets and Gill class action) consume management bandwidth and cash without disclosed reserve allocation. SR005, SR006, SR015
CR039 A single large-customer revenue departure — analogous to OpenAI leaving Scale AI — could reduce Mercors ARR by an estimated 20-40%, based on industry concentration norms at this stage. SR018, SR024, SR026
CR040 Key monitoring indicators for Mercors thesis break include: AB 5 enforcement action opened, second material breach, an adverse court finding or new trade-secret dispute, ARR growth <50% YoY, or customer-concentration ratio >70%. SR033, SR035
CV001 Mercor announced a $30 million Series A at a $250 million valuation in 2024. SV001
CV002 Mercor announced a $100 million Series B at a $2 billion valuation in February 2025. SV002, SV003, SV004, SV005
CV003 Mercor announced a $350 million Series C at a $10 billion valuation in October 2025. SV006, SV007
CV004 Mercor's total disclosed primary capital across seed, Series A, Series B, and Series C is about $483.6 million. SV001, SV002, SV006
CV005 TechCrunch reported Mercor had reached a $75 million annual recurring revenue run rate by February 2025. SV003
CV006 TechCrunch reported Mercor was approaching a $450 million annualized revenue run rate in September 2025. SV008
CV007 Mercor's March 2026 engineering post said the company had crossed a $1 billion annualized revenue run rate earlier in 2026. SV009
CV008 TechCrunch reported Mercor's headline revenue includes the full amount customers pay before contractors receive their share. SV008
CV009 Mercor said in October 2025 that it had more than 30,000 contractors and was paying more than $1.5 million per day to them; its 2026 engineering post raised that daily payout figure above $2 million. SV006, SV007, SV009
CV010 TechCrunch Disrupt 2025 coverage named OpenAI, Anthropic, and Meta as Mercor customers and said the company had increased annualized recurring revenue to roughly $500 million while remaining profitable. SV010
CV011 TechCrunch reported that Mercor was affected by the LiteLLM supply-chain attack and said the company brought in third-party forensics experts. SV011
CV012 TechCrunch reported that Meta paused contracts with Mercor after the breach while other customers reviewed their relationships. SV012
CV013 Claim Depot said the breach litigation alleged exposure of personal data for more than 40,000 people and noted multiple federal class actions tied to the incident. SV013
CV014 Scale AI sued Mercor and former Scale employee Eugene Ling in September 2025 over alleged trade-secret and customer-material misuse. SV014, SV015, SV016
CV015 PacerMonitor shows the Scale AI lawsuit was voluntarily dismissed with prejudice in January 2026. SV017
CV016 Mercor's research page says the company is used by the top five AI labs and six of the Magnificent Seven. SV018
CV017 Mercor's experts page and contractor docs show a weekly payout system that relies on Stripe or Wise, time tracking, screenshot review, and human judgment over disputed hours. SV019, SV023
CV018 Mercor's supported-countries policy shows that payment coverage depends on Stripe or Wise jurisdiction support and that some countries are unsupported. SV024
CV019 Mercor's assessments page says assessments are now a primary entry point to work on the platform and can qualify talent for multiple roles. SV025
CV020 Mercor's RL Studio documentation describes an internal production system with projects, worlds, task states, reviewer flows, and approval tracking. SV026
CV021 Mercor's research, Enterprise AI, and APEX pages show the company is trying to move from pure expert supply toward benchmark, evaluation, and workflow infrastructure. SV018, SV020, SV021, SV022
CV022 Appen's public product pages show the category already converging around expert RLHF, agent trajectories, regulatory audits, and model-evaluation workflows rather than commodity labeling alone. SV028, SV029, SV030
CV023 Stanford's 2025 AI Index and MarketsandMarkets both point to ongoing AI-investment and data-annotation demand growth, which supports continued category expansion. SV031, SV040
CV024 As of May 2026, Appen's market cap was about $0.23 billion against roughly $0.23 billion of revenue, implying about a 1x revenue multiple. SV032, SV033
CV025 As of May 2026, Upwork's market cap was about $1.08 billion against roughly $0.79 billion of revenue, implying about a 1.4x revenue multiple. SV034, SV035
CV026 As of May 2026, Fiverr's market cap was about $0.39 billion against roughly $0.42 billion of revenue, implying about a 0.9x revenue multiple. SV036, SV037
CV027 As of May 2026, Palantir's market cap was about $328.14 billion against roughly $5.22 billion of revenue, implying roughly a 63x revenue multiple. SV038, SV039
CV028 Public labor-market and data-service comps trading around 1x revenue imply Mercor's $10 billion mark cannot be defended on marketplace economics alone. SV032, SV033, SV034, SV035, SV036, SV037
CV029 Palantir-like software multiples show how much upside exists if Mercor proves durable software-control characteristics, but that outcome requires very different evidence from a labor marketplace. SV038, SV039, SV020, SV021, SV022
CV030 Public evidence indicates Mercor still depends on a small number of frontier AI labs for most of its revenue and does not disclose retention metrics. SV008, SV010, SV012
CV031 Mercor's public materials and media coverage still do not disclose audited net revenue, take rate, gross margin, burn, or cash on hand. SV002, SV006, SV008, SV009
CV032 At $10 billion versus the September 2025 $450 million annualized run-rate figure, Mercor trades at about 22x gross revenue. SV006, SV008
CV033 If Mercor's own 2026 $1 billion annualized revenue claim were verified, the $10 billion mark would imply about a 10x gross revenue multiple, but that figure would still be unaudited and gross of contractor payouts. SV008, SV009
CV034 Breach fallout, customer concentration, and labor or legal exposure make Mercor's downside more asymmetric than pure software comp sets suggest. SV011, SV012, SV013, SV014, SV018, SV023, SV024
CV035 The bull case requires Mercor's benchmark and workflow products to become sticky enough to lift margin, reduce concentration, and support a partial software rerating. SV018, SV020, SV021, SV022, SV025, SV026
CV036 The base case assumes revenue keeps growing but the valuation multiple compresses somewhat because benchmark attach, net revenue quality, and post-breach trust are only partially proven. SV008, SV011, SV012, SV020, SV021, SV022
CV037 The bear case is a reset toward services or labor-platform multiples if a top customer is lost, security issues persist, or contractor and legal costs rise. SV011, SV012, SV013, SV014, SV023, SV024, SV032, SV033, SV034, SV035, SV036, SV037
CV038 Mercor's strongest public upside lever is benchmark and workflow evidence such as APEX, Enterprise AI, RL Studio, and assessments rather than simply adding more contractor volume. SV020, SV021, SV022, SV025, SV026
CV039 Given the current evidence set, the most sensible recommendation is TRACK rather than BUY, with medium confidence and a high risk rating. SV008, SV010, SV011, SV012, SV031, SV032, SV033, SV034, SV035, SV036, SV037, SV038, SV039
CV040 Mercor's valuation stance at the $10 billion mark is stretched until the company discloses cleaner net-revenue, concentration, and trust-remediation evidence. SV008, SV011, SV012, SV023, SV032, SV033, SV034, SV035, SV036, SV037
CV041 A price closer to roughly $6 billion to $7.5 billion, or audited proof of software-like economics at the current mark, would make the setup materially more investable. SV008, SV009, SV020, SV021, SV022, SV032, SV033, SV034, SV035, SV036, SV037, SV038, SV039
CV042 Final diligence should focus on net revenue and take rate, top-10 customer mix, post-breach security controls, and contractor or legal exposure before any buy call. SV011, SV012, SV013, SV017, SV023, SV024, SV027
来源
编号出版方标题引文
SO001 Mercor Mercor homepage
SO002 Mercor Mercor experts page
SO003 Mercor Mercor research
SO004 Mercor Mercor careers
SO005 Mercor Introducing Mercor: Redefining Hiring With AI
SO006 Mercor Announcing Mercor's Series B
SO007 Mercor Announcing Mercor's Series C
SO008 PR Newswire Mercor raises $30M Series A at a $250M valuation to create jobs with AI
SO009 TechCrunch Mercor, an AI recruiting startup founded by 21-year-olds, raises $100M at $2B valuation
SO010 CNBC AI hiring startup Mercor now valued at $2 billion after recent strong growth
SO011 Bloomberg AI Startup Led by 21-Year-Old Thiel Fellow Lands $2 Billion Valuation
SO012 TechCrunch Sources: AI training startup Mercor eyes $10B+ valuation on $450M run-rate
SO013 TechCrunch Mercor quintuples valuation to $10B with $350M Series C
SO014 CNBC AI startup Mercor now valued at $10 billion with new $350 million funding round
SO015 TechCrunch How AI labs use Mercor to get the data companies won't share
SO016 TechCrunch Scale AI is suing a former employee and rival Mercor, alleging they tried to steal its biggest customers
SO017 TechCrunch After data breach, $10B-valued startup Mercor is having a month
SO018 CourtListener Scale AI, Inc. v. Mercor.io Corporation docket
SO019 Justia Scale AI, Inc. v. Mercor.io Corporation et al
SO020 PacerMonitor Scale AI, Inc. v. Mercor.io Corporation et al
SO021 KTVU AI startup Mercor, valued at $2B, founded by college dropouts
SO022 Forbes Adarsh Hiremath profile
SO023 Time The people training AI in India
SO024 CNBC OpenAI is winding down its work with Scale AI, whose founder is joining Meta
SO025 The Times of India Who are Adarsh Hiremath and Surya Midha?
SM001 Mercor Mercor research
SM002 Mercor Mercor homepage
SM003 Mercor Mercor experts page
SM004 TechCrunch Sources: AI training startup Mercor eyes $10B+ valuation on $450M run-rate
SM005 TechCrunch How AI labs use Mercor to get the data companies won't share
SM006 Stanford HAI AI Index Report 2025
SM007 IBM Key findings from Stanford's 2025 AI Index report
SM008 MarketsandMarkets Data Annotation and Labeling Market by Component and Vertical
SM009 Labelbox An economic report on the human expertise fueling frontier AI
SM010 OpenAI Aligning language models to follow instructions
SM011 OpenAI Improvements to fine-tuning API and expanding custom models program
SM012 Anthropic Constitutional AI: Harmlessness from AI Feedback
SM013 Scale AI Scale RLHF
SM014 Labelbox Labelbox RLHF
SM015 Appen Frontier Alignment
SM016 iMerit Tools and automation platforms for RLHF
SM017 CloudFactory RLHF: How to align AI with human values
SM018 Snorkel AI Snorkel homepage
SM019 Invisible Technologies Invisible homepage
SM020 Toloka Toloka training data for AI agents and LLMs
SM021 TIME The people training AI in India
SM022 Rest of World The hidden labor force powering AI
SM023 Rest of World The people paid to train AI are outsmarted by it
SM024 NIST AI Risk Management Framework
SM025 Appen Human data for frontier AI
SM026 Surge AI Surge AI homepage
SP001 Mercor Mercor research
SP002 Mercor Mercor experts page
SP003 Scale AI Scale AI homepage
SP004 Scale AI Scale RLHF
SP005 Labelbox Why Labelbox
SP006 Labelbox Labelbox RLHF
SP007 Labelbox Labelbox expert network
SP008 Appen Appen homepage
SP009 Appen Appen investor relations
SP010 Appen Frontier Alignment
SP011 Appen Data capabilities for agentic AI
SP012 iMerit iMerit homepage
SP013 CloudFactory CloudFactory homepage
SP014 Surge AI Surge AI homepage
SP015 Snorkel AI Snorkel homepage
SP016 Snorkel AI Snorkel how it works
SP017 SuperAnnotate SuperAnnotate homepage
SP018 Invisible Technologies Invisible homepage
SP019 Toloka Toloka homepage
SP020 TechCrunch Mercor quintuples valuation to $10B with $350M Series C
SP021 CNBC Scale AI founder Wang announces exit for Meta as part of $14B deal
SP022 CNBC OpenAI is winding down its work with Scale AI, whose founder is joining Meta
SP023 MarketsandMarkets Data Annotation and Labeling Market
SP024 Appen Model evaluation and integrity
SP025 TechCrunch How AI labs use Mercor to get the data companies won't share
SP026 CNBC AI startup Mercor now valued at $10 billion with new $350 million funding round
SI001 Mercor Mercor homepage
SI002 Mercor Mercor experts page
SI003 Mercor Mercor careers
SI004 Mercor Announcing Mercor's Series B
SI005 Mercor Announcing Mercor's Series C
SI006 Mercor When you go from $2 million a month to $2 million a day
SI007 Mercor Docs Payments
SI008 PR Newswire Mercor raises $30M Series A at a $250M valuation to create jobs with AI
SI009 CNBC AI hiring startup Mercor now valued at $2 billion after recent growth
SI010 TechCrunch Mercor, an AI recruiting startup founded by 21-year-olds, raises $100M at $2B valuation
SI011 Bloomberg AI Startup Led by 21-Year-Old Thiel Fellow Lands $2 Billion Valuation
SI012 TechCrunch Sources: AI training startup Mercor eyes $10B+ valuation on $450M run-rate
SI013 CNBC AI startup Mercor now valued at $10 billion with new $350 million funding round
SI014 KTVU AI startup Mercor, valued at $2B, founded by college dropouts
SI015 Forbes Adarsh Hiremath profile
SI016 TechCrunch How AI labs use Mercor to get the data companies won't share
SI017 TechCrunch After data breach, $10B-valued startup Mercor is having a month
SI018 TechCrunch Scale AI is suing a former employee and rival Mercor, alleging they tried to steal its biggest customers
SI019 CourtListener Scale AI, Inc. v. Mercor.io Corporation docket
SI020 PacerMonitor Scale AI, Inc. v. Mercor.io Corporation et al
SI021 CNBC OpenAI is winding down its work with Scale AI, whose founder is joining Meta
SI022 CNBC Scale AI cuts 14% of workforce after Meta investment, hiring of founder Wang
SI023 Appen Investor Relations
SI024 Appen Data capabilities: model evaluation and integrity
SI025 California Legislature AB 5 worker status law text
SI026 Mercor Mercor blog index
SI027 CNBC Google, Scale AI's largest customer, plans split after Meta deal, sources say
SI028 Stanford HAI 2025 AI Index Report
SI029 Grand View Research Data annotation tools market report
SE001 Mercor Mercor homepage
SE002 Mercor Mercor experts page
SE003 Mercor Mercor research
SE004 Mercor Mercor careers
SE005 Mercor Announcing Mercor's Series B
SE006 Mercor Announcing Mercor's Series C
SE007 Mercor APEX Benchmarks
SE008 Mercor Introducing the AI Productivity Index for Software Engineering
SE009 Mercor APEX-SWE leaderboard
SE010 Mercor Introducing APEX-Agents
SE011 Mercor Expanding the Mercor AI Productivity Index
SE012 Mercor Introducing Mercor Enterprise AI
SE013 Mercor Engineering Monty: Scaling an AI Interviewer
SE014 Mercor Rebuilding a Critical Service in One Week
SE015 Mercor The Economy will Become an RL Environment Machine
SE016 Mercor Docs Documentation Index
SE017 Mercor Docs How Mercor Uses AI and Data
SE018 Mercor Docs LLM Usage Policy
SE019 Mercor Docs Background Check
SE020 Mercor Docs Use Insightful for Time Tracking
SE021 Mercor Docs Payments
SE022 OpenAI Aligning language models to follow instructions
SE023 OpenAI Improvements to fine-tuning API and expanding custom models program
SE024 Anthropic Constitutional AI: Harmlessness from AI Feedback
SE025 Appen Frontier Alignment
SE026 Appen Agentic AI data capabilities
SE027 Appen Data capabilities: model evaluation and integrity
SE028 iMerit Tools and automation platforms for RLHF
SE029 Toloka Toloka training data for AI agents and LLMs
SE030 Scale AI Scale RLHF
SE031 TechCrunch How AI labs use Mercor to get the data companies won't share
SE032 TechCrunch After data breach, $10B-valued startup Mercor is having a month
SE033 CNBC AI startup Mercor now valued at $10 billion with new $350 million funding round
SE034 CloudFactory RLHF: How to align AI with human values
SU001 TechCrunch Mercor raises $34M Series A to scale AI training marketplace
SU002 TechCrunch Mercor, an AI recruiting startup founded by 21-year-olds, raises $100M at $2B valuation
SU003 TechCrunch Sources: AI training startup Mercor eyes $10B valuation on $450M run rate
SU004 TechCrunch Mercor quintuples valuation to $10B with $350M Series C
SU005 TechCrunch How AI labs use Mercor to get the data companies wont share
SU006 CNBC AI hiring startup Mercor now valued at $2 billion after recent growth
SU007 CNBC AI hiring startup Mercor raises funding at $10B valuation
SU008 CNBC OpenAI is winding down its work with Scale AI; founder is joining Meta
SU009 CNBC Google, Scale AI's largest customer, plans split after Meta deal
SU010 Forbes Mercor reaches $10 billion valuation
SU011 Forbes Mercor makes the AI Cloud 100
SU012 Bloomberg Mercor, the $10 billion AI startup recruiting white-collar workers
SU013 Bloomberg AI startup led by 21-year-old Thiel Fellow lands $2 billion valuation
SU014 Mercor Mercor Series C announcement
SU015 Mercor Mercor Series B announcement
SU016 Mercor When you go from $2 million a month to $2 million a day
SU017 Mercor Mercor at TechCrunch Disrupt
SU018 Mercor Introducing Mercor Enterprise AI
SU019 Mercor Introducing Mercor
SU020 Mercor Mercor Experts
SU021 Mercor Mercor homepage
SU022 Mercor Mercor Research
SU023 Mercor Talent Docs Mercor talent portal overview
SU024 Mercor Talent Docs Project onboarding guide
SU025 KTVU Bay Area high school friends, college drop-outs behind $2B AI recruiting startup
SU026 Rest of World The people paid to train AI are outsmarted by it
SU027 Axios Scale AI sues Mercor over alleged trade secret theft
SU028 Times of India Who are Adarsh Hiremath and Surya Midha, the youngest self-made billionaires
SU029 PR Newswire Mercor raises $30M Series A at $250M valuation to create jobs with AI
SU030 Mercor Announcing Mercor Series C
SU031 Mercor Introducing Mercor Apex
SU032 Mercor Apex SWE leaderboard
SU033 CNBC Turing AI valuation reaches $2.2 billion
SR001 US DOL FLSA Misclassification Rulemaking
SR002 IRS Understanding Employee vs Contractor Designation
SR003 California FTB Worker Classification and AB 5 FAQ
SR004 California DIR AB 5 Worker Classification Overview
SR005 California Legislature AB 5 bill text (2019)
SR006 California DIR AB 5 Workers Compensation FAQ
SR007 Reuters California Supreme Court to clarify gig worker law in major trucking case
SR008 NIST NIST Cybersecurity Framework
SR009 NIST AI Risk Management Framework
SR010 European Commission EU AI Act proposal
SR011 Claim Depot Mercor data breach class action lawsuit
SR012 CourtListener Gill v. Mercorio Corporation (2026 data breach case)
SR013 Justia Docket 3:2026cv02831 — Gill v Mercorio Corporation
SR014 TechCrunch Mercor says it was hit by a cyberattack tied to compromise of open-source LiteLLM project
SR015 TechCrunch Scale AI is suing a former employee and rival Mercor, alleging they tried to steal its biggest customers
SR016 Axios Scale AI sues Mercor over alleged trade secret theft
SR017 Bloomberg Scale AI sues rival startup Mercor
SR018 Reuters OpenAI winds down work with Scale AI after Meta deal
SR019 CourtListener Scale AI Inc v Mercorio Corporation (2025 trade secrets)
SR020 PACER Monitor Scale AI Inc v Mercorio Corporation — case docket
SR021 TechCrunch After data breach, $10B-valued startup Mercor is having a month
SR022 Rest of World The hidden labor force powering AI
SR023 Rest of World The people paid to train AI are outsmarted by it
SR024 CNBC Scale AI cuts 14% of workforce after Meta investment, hiring of Wang
SR025 CNBC Scale AI founder Wang announces exit for Meta, part of $14 billion deal
SR026 Mercor Mercor Trust Center
SR027 Mercor Talent Docs Taxes and Work Authorization Policy
SR028 Mercor Talent Docs Contracts Policy
SR029 Mercor Talent Docs Legal Support Documentation
SR030 Mercor Talent Docs Data and AI Usage Policy
SR031 Mercor When volume grew 10x in a month and we had one week to fix it
SR032 Time AI data workers in India working for Scale AI
SR033 TechCrunch Mercor quintuples valuation to $10B with $350M Series C
SR034 TechCrunch Sources: AI training startup Mercor eyes $10B valuation on $450M run rate
SR035 Bloomberg Mercor: The $10 Billion AI Startup Recruiting White-Collar Workers
SV001 PR Newswire Mercor raises $30M Series A at a $250M valuation to create jobs with AI
SV002 Mercor Announcing Mercor's Series B
SV003 TechCrunch Mercor, an AI recruiting startup founded by 21-year-olds, raises $100M at $2B valuation
SV004 CNBC AI hiring startup Mercor now valued at $2 billion after recent growth
SV005 Bloomberg AI Startup Led by 21-Year-Old Thiel Fellow Lands $2 Billion Valuation
SV006 Mercor Announcing Mercor's Series C
SV007 CNBC AI startup Mercor now valued at $10 billion with new $350 million funding round
SV008 TechCrunch Sources: AI training startup Mercor eyes $10B+ valuation on $450M run-rate
SV009 Mercor When you go from $2 million a month to $2 million a day
SV010 TechCrunch How AI labs use Mercor to get the data companies won't share
SV011 TechCrunch Mercor says it was hit by cyberattack tied to compromise of open-source LiteLLM project
SV012 TechCrunch After data breach, $10B-valued startup Mercor is having a month
SV013 Claim Depot Mercor class action alleges AI startup failed to protect data of more than 40,000 people
SV014 CourtListener Scale AI, Inc. v. Mercor.io Corporation docket
SV015 Axios Scale AI sues rival "unicorn" Mercor
SV016 Bloomberg Scale AI Sues Rival Startup Mercor Alleging Trade-Secret Theft
SV017 PacerMonitor Scale AI, Inc. v. Mercor.io Corporation et al
SV018 Mercor Mercor research
SV019 Mercor Mercor experts page
SV020 Mercor Introducing Mercor Enterprise AI
SV021 Mercor Introducing APEX-Agents
SV022 Mercor APEX-SWE leaderboard
SV023 Mercor Docs Time Tracking & Pay Policies
SV024 Mercor Docs Supported Countries for Payment
SV025 Mercor Docs Assessments
SV026 Mercor Docs RL Studio (RLS)
SV027 Appen Investor relations
SV028 Appen Frontier Alignment
SV029 Appen Agentic AI
SV030 Appen Data capabilities: model evaluation and integrity
SV031 Stanford HAI 2025 AI Index Report
SV032 CompaniesMarketCap Appen market capitalization
SV033 CompaniesMarketCap Appen revenue
SV034 CompaniesMarketCap Upwork market capitalization
SV035 CompaniesMarketCap Upwork revenue
SV036 CompaniesMarketCap Fiverr market capitalization
SV037 CompaniesMarketCap Fiverr revenue
SV038 CompaniesMarketCap Palantir market capitalization
SV039 CompaniesMarketCap Palantir revenue
SV040 MarketsandMarkets Data Annotation and Labeling Market