AfterQuery
专攻专家推理数据平台,增速极快但客户高度集中,定价面临估值拉伸压力
AfterQuery 展现出专家级数据平台的领跑身位,营收增长迅猛,但其估值已被打到极满,同时背负着大客户高度集中、利润率不透明以及公开经营数据披露有限的现实风险。
覆盖范围与披露说明
公开报道证实了公司概况、业务增速与估值水平,但经审计的利润率、留存率及客户集中度数据仍未公开披露。
封面要素
公司概况
AfterQuery 是一家获风投支持的私营 AI 数据平台,向前沿模型实验室及相邻企业级 AI 买家销售专家推理轨迹、强化学习环境与评测工具。公司自 2024 年成立以来规模迅速扩张,但公开记录针对续约持续性、毛利率以及高估值在多大程度上反映老股转让定价而非一级市场增资等关键问题,仍存疑问。
- 成立时间
- 2024-11-01
- 创始人
- Spencer Mateega, Carlos Georgescu, Danny Tang
- 创立地点
- San Francisco, California, USA
- 总部
- San Francisco, California, USA
- 产品
- 面向基础模型后训练,提供专家生成的推理数据集、强化学习沙盒、基准评测套件与定制工作流工具。
- 客户
- 前沿 AI 实验室、模型评测团队,以及受监管领域中早期采用 AI 的企业买家
- 商业模式
- 针对定制数据包、强化学习环境及诊断咨询的按项目计费 B2B 合同
- 阶段
- Late-stage private
- 融资情况
- 自 Y Combinator 孵化起步,以 $300M 投后估值完成 $30M A 轮融资,后续据媒体报道以 $3.2B 估值获得新一轮融资。
执行摘要
主要优势
- 深耕专家推理数据与强化学习环境,确立品类领跑的产品定位
- 营收规模放量迅速,据报道来自前沿 AI 实验室的客户需求强劲
- 主打软件优先的工作流工具链,摆脱单纯依赖劳动力套利的标注模式
主要风险
- 业务严重依赖少数前沿 AI 实验室大客户,客户集中度居高不下
- 毛利率缺乏透明度,且专家级人工交付成本可能居高不下
- 估值隐含了对持续超高速增长与极低执行风险的苛刻假设
未决问题
- 经审计的客户集中度、续约条款以及平均合同期限均未公开披露
- 各产品线的毛利率与边际贡献率均未公开验证
- 据报道的 $3.2B 融资中,新股与老股的具体构成尚不明确
目录
01公司概况
1.1 公司定位、创立渊源与核心使命
AfterQuery 是一家应用人工智能研究实验室兼专业训练数据提供商,成立于 2024 年,总部位于加利福尼亚州旧金山。公司旨在解决前沿基础模型开发者面临的核心运营瓶颈:尽管最先进的大语言模型在标准基准测试中表现优异,但在执行需要复杂多步推理、语境权衡和工具调度的专业工作流时仍显吃力。AfterQuery 并不依赖公开互联网抓取或通用众包标注,而是紧扣经过认证的从业人员如何做出专业决策,构建专有数据集与强化学习环境。公司核心产品涵盖监督微调(SFT)提示-响应轨迹、将定性专业判断转化为可扩展奖励函数的强化学习评分准则,以及依托模型上下文协议(MCP)服务器与桌面环境构建的模拟 Agent 沙盒。[CO001, CO007]
| 指标 | 数值 / 状态 | 日期 | 置信度 | 尽调缺口 / 局限 |
|---|---|---|---|---|
| 累计融资 | $30.5M ($30M Series A + $500K Seed) | 2026-04-09 | 高 | 据报 $3.2B 轮次是否完成交割尚未确认 |
| 估值 | $300M 确认投后;$3.2B 未确认 | 2026-09-01 | 中 | Forbes 报道了 2026 年 9 月 $3.2B 估值,但公司尚未确认 |
| 年经常性收入(ARR) | $100M 运行率 | 2026-04-09 | 中 | 公司自述运行率;仍须经审计的财务验证 |
| 员工数 | 30 名员工(已披露)至 212 名员工(Tracxn 预估) | 2026-07-26 | 中 | 核心团队人数(30)与平台专家网络人数存在出入 |
| 专家贡献者网络 | 近 100,000 名认证专业人士 | 2026-04-09 | 中 | 外包网络规模;月活跃贡献者人数未披露 |
数据反映截至 2026 年第三季度经核实的公司披露、风投公告与分析师预估;未确认融资轮次已特别标注。
[CO003, CO004, CO005, CO008, CO011, CO012]1.2 创始团队、领导架构与关键人风险
AfterQuery 由首席执行官 Spencer Mateega、Carlos Georgescu 与 Danny Tang 联合创立。创始人在入读 University of Pennsylvania 的 Wharton School 本科之前,便在早期技术项目与高中创业中建立了合作关系。Spencer Mateega 取得计算机科学与金融双学位,其间曾在 Google、Meta、Morgan Stanley 及 Silver Lake 完成软件与私募股权实习。Carlos Georgescu 带来 Citadel Securities 的量化软件工程经验及过往软件创业经历,Danny Tang 则拥有机器学习神经科学研究与商业地产资本市场背景。由于 AfterQuery 仅保留约 30 名全职员工的精干核心团队,运营与研发领导权高度集中于三位创始人手中。技术架构、前沿实验室客户关系以及平台专家网络管理均存在明显的关键人依赖。[CO001, CO009, CO011]
| 成员 | 职务 | 背景 | 创始人契合度与职能覆盖 | 关键人依赖度 |
|---|---|---|---|---|
| Spencer Mateega | CEO 兼联合创始人 | Meta 与 Google 软件工程师、Silver Lake 私募股权、Wharton 金融与计算机 | 高管领导力、技术愿景与前沿实验室销售 | 关键高管依赖 |
| Carlos Georgescu | 联合创始人 | Citadel Securities 软件工程师、教育科技创业公司创始人、奖学金得主 | 后训练架构、算法设计与评估框架 | 关键技术依赖 |
| Danny Tang | 联合创始人 | Wharton 毕业生、机器学习神经科学实验室、商业地产 IPO 经验 | 运营规模化、专家网络供给与战略合作 | 重大运营依赖 |
管理团队履历与职责汇总自 Y Combinator 目录、官方公告与专业职业履历。
[CO001, CO009]1.3 融资历史、估值演进与股权架构
AfterQuery 作为 Y Combinator 2025 年冬季批次成员启动商业运营,最初筹得 $500,000 种子前轮投资。2026 年 4 月,公司完成 $300M 投后估值的 $30M A 轮融资。该轮融资由 Altos Ventures 领投,The Raine Group、Y Combinator、BoxGroup、Latitude Capital,以及来自 Google DeepMind、OpenAI、Anthropic 和 Microsoft AI 等顶尖 AI 实验室的知名天使投资人跟投。各项已确认轮次累计披露的风投融资金额达 $30.5M。2026 年 9 月,包括 Forbes 和 TechCrunch 在内的全国性科技媒体报道称,AfterQuery 完成新一轮融资,公司估值达 $3.2B,相当于 5 个月内估值跃升十倍,创下 Y Combinator 史上据报道最快晋升独角兽的纪录。然而,有关交割条款、领投方及保护性条款的正式文件目前仍未公开。[CO002, CO004, CO008, CO010, CO012]
| 利益相关方 | 角色 | 控制权或经济重要性 | 尽调诉求 |
|---|---|---|---|
| Altos Ventures | 领投方(Series A) | 持有重大少数经济权益并拥有董事会席位 | 审查表决权与保护性条款 |
| The Raine Group | 战略投资方(Series A) | 战略增长伙伴兼企业网络渠道方 | 评估商业协同效应与业务合作协议 |
| Y Combinator | 种子轮投资方(W25 期) | 持有标准股权与按比例优先认购权的孵化器投资方 | 查验初始持股比例与按比例优先认购权行权记录 |
| BoxGroup | 参投风投机构(Series A) | 早期联合投资财团跟投方 | 核实配额与知情权 |
| Latitude Capital | 联合投资方(Series A) | 成长资本联合投资方 | 核对跟投条款与老股转让限制约定 |
机构投资方与财团参投方梳理自 Series A 披露文件与风投数据库。
[CO004, CO010, CO012]1.4 业务演进、产品发展路径与尽调风险
AfterQuery 自 2025 年初上线后商业变现迅速提速,收入运行率从 2025 年 9 月预估的 $6.5M 增长至 2026 年 4 月年化收入运行率突破 $100M。这一发展势头源于顶尖基础模型开发者对软件工程、企业金融及法律分析等专业领域人工策划推理数据与评测基准的爆发式需求。该平台统筹近 100,000 名认证专业贡献者,负责生成示范轨迹、代码重构轨迹及奖励模型评分准则。尽管账面增速亮眼,尽职调查仍揭示出重大运营风险。AfterQuery 的客户收入高度集中于少数前沿 AI 实验室,这些实验室可能将数据策划转向自研,或转向 Scale AI 等平台型老牌厂商。此外,管理定制化专业工作流相较初级标注交易市场带来了更高的成本结构,必须大幅借助标准化现成数据目录才能拉升毛利率。[CO005, CO006, CO007, CO013, CO014]
AfterQuery 自成立起,历经快速 A 轮融资及数十亿美元估值报道的时间线演进。
日期与财务数值反映公开报道和经过核实的公司新闻稿;据报道的 $3.2B 融资轮次仍以最终交割文件为准。
2024-11: 公司成立 [CO001]
2025-01: Y Combinator W25 期 [CO002]
2025-09: 早期收入起量 [CO003]
2026-04: 官宣 $30M A 轮融资 [CO004]
2026-04: $100M ARR 运行率里程碑 [CO005]
2026-04: 平台专家规模达 100K [CO006]
2026-06: 拓展企业级沙箱 [CO007]
2026-09: 据报完成 $3.2B 估值融资 [CO008]
1.5 重点图表
02市场分析
2.1 市场定义与业务边界
人工智能训练数据市场正在经历结构性转型,从低成本众包数据标注转向高置信度专家推理与交互式强化学习环境。在深度学习落地的初期,数据供应商主要比拼计算机视觉与对话文本的单位标注吞吐量。然而,前沿基础模型开发者已基本耗尽通用网络抓取与浅层标注所能带来的性能红利。为攻克关键任务应用中的模型推理瓶颈与幻觉问题,前沿实验室正将大量资金投向持证医师、执业企业律师、金融分析师及资深软件架构师等经认证的行业专家,由其生成逐步推理轨迹与严格评测准则。AfterQuery 正好卡位这一分水岭:其产品涵盖监督微调数据、定制强化学习沙盒及特定领域评测基础设施,同时明确排除同质化的基础标注与通用基础设施运维。[CM006, CM007, CM013]
| 细分板块 / 类别 | 纳入支出 | 排除支出 | 买方 / 付款方 | 战略相关性 |
|---|---|---|---|---|
| 前沿后训练数据 | 垂直领域专用 SFT 推理轨迹与专家示范数据 | 通用公开网络抓取与基础文本过滤 | 前沿 AI 实验室后训练负责人 | 推动模型能力爬坡的近期核心收入引擎 |
| RL 训练场(Gyms)与工具环境 | 定制沙箱、API 模拟桩、Docker 容器与测试套件 | 大宗算力托管与通用虚拟机配置 | 强化学习(RL)研究科学家与评估工程团队 | 支持行为验证的高留存软件层 |
| 领域评分规则(Rubrics)与基准测试 | 定制评估框架、评分规则工程与经过验证的测试集 | 标准开源静态基准(MMLU、GSM8K) | 模型评估负责人与企业信任安全团队 | 售前能力诊断与销售赋能资产 |
| 企业级智能体集成 | 企业专属工作流数据、定制业务流程建模与专有 CAD/代码摄入 | 标准企业级 SaaS 席位授权与通用横向工作流自动化 | 企业首席数据官(CDO)、AI 副总裁与业务部门主管 | 跳出集中基础模型实验室圈子的长期 TAM 拓展 |
分类基于 AfterQuery 产品披露与行业分析;排除支出着重标出高阶数据方案不涉及的大宗通用服务。
[CM006, CM007, CM013]2.2 市场规模测算视角及自上而下与自下而上的分歧
测算专家后训练数据的潜在市场空间,揭示出传统自上而下分析师预测与自下而上实际商业表现之间的显著脱节。Grand View Research 与 Fortune Business Insights 发布的权威市场研究报告将 2026 年全球 AI 训练数据集市场规模预计为约 $3.9B 至 $4.4B,Mordor Intelligence 更宽泛的分类口径则延伸至 $11.9B。然而,这些自上而下的市场测算模型普遍低估了前沿人工智能实验室预算转移的速度。对 Scale AI、Mercor、Handshake、Micro1、Surge AI 和 AfterQuery 等核心商业供应商披露的年化收入运行率进行自下而上梳理后发现,全行业累计开票金额每年已达 $6B 至 $8B。随着基础模型开发者展开白热化的能力角逐,此前投向通用算力集群或内部启发式数据流的预算,正被大规模调往外部高置信度人类数据生成与专用强化学习训练场(gym)。[CM001, CM003, CM004, CM005, CM012]
| 测算视角 | 发布方 / 信源 | 年份 | 地域 | 估计值 | 复合年均增长率(CAGR)/ 增速背景 | 测算方法与局限 |
|---|---|---|---|---|---|---|
| 全球 AI 训练数据集 | Grand View Research | 2026 | 全球 | $3.9B | 年均增长 20% 中高位 | 自上而下的市场模型汇总了宽口径数据服务;低估了前沿专用强化学习环境 |
| AI 训练数据与服务 | Fortune Business Insights | 2026 | 全球 | $4.4B | 预计年均复合增长率超过 20% | 二手市场预估,整合了商业标注与数据准备平台 |
| 广义 AI 数据与模型服务 | Mordor Intelligence | 2026 | 全球 | $11.9B | 复合年增长率超过 25% | 宽口径分类,涵盖合成数据工具、语音转录与标注硬件 |
| 前沿供应商实际开票口径 | Signals Inbox / 行业数据 | 2026 | 全球(美国及跨境) | $6B-$8B+ 合计运行率 | 头部供应商同比增速超 100% | 自下而上加总 Scale AI、Mercor、Handshake、Micro1、Surge AI 与 AfterQuery 据报的总运行率 |
各项预估反映出各家统计口径并不一致;自上而下的行业报告滞后于自下而上供应商收入所体现的 2025-2026 年采购暴增态势。
[CM003, CM004, CM005]多层目标市场规模测算,对比泛 IT 与 AI 支出以及前沿后训练数据可触达预算。
各层宽度仅表示层级,不代表数值比例。
宏观市场层级基于已公开的行业分析师估算区间;前沿供应商层级代表 2026 年化总采购运行率。
[CM003, CM004, CM005, CM012]2.3 买家细分、采购动态与业务工作流
后训练生态中的采购动态具有鲜明的客群两极化格局,买家涵盖前沿 AI 科研实验室与前瞻性企业机构。在基础模型实验室中,核心决策者是研究科学家、评测负责人与后训练主管,其核心职责在于攻坚核心推理基准指标并防范灾难性模型失效模式。此类技术买家采用咨询式、结果导向的采购流程,合同金额取决于专业领域难度、验证复杂度及数据集规模,而非可预测的按席位软件许可模式。与此同时,第二条企业级应用渠道正在萌芽。行业调查显示,尽管超过 60% 的大型企业正在积极试验自主 AI Agent,但将规模化 Agent 工作流成功部署至生产环境的企业不足四分之一。弥合这一企业落地鸿沟,需要将现成基础模型与企业专属专有数据、合规边界及确定性工具调用环境深度结合,这也为 AfterQuery 打开了不断扩大的第二增长曲线市场机遇。[CM002, CM008, CM009, CM013]
| 目标客群 | 主要买方 | 主要使用者 | 付款方 / 预算归属方 | 目标工作流 | 采购触发因素 |
|---|---|---|---|---|---|
| 前沿模型实验室(美国) | 研究副总裁 / 后训练负责人 | 研究科学家与评估工程师 | 基础模型训练预算 | 推理基准、工具调用链路与 RLHF | 模型能力遭遇瓶颈及与同业前沿模型的基准竞争 |
| 超大规模云厂商与硬件芯片厂商 | AI 生态负责人 / 基础实验室负责人 | 应用 AI 工程师与解决方案架构师 | 战略 AI 基础设施预算 | 领域专用智能体示范与合作伙伴展示方案 | 在专有芯片上打造差异化模型权重的需求(例如 Nvidia Nemotron) |
| 海外区域模型实验室 | 首席科学家 / AI 实验室总监 | 后训练与对齐团队 | 国际研发与主权算力预算 | 双语推理数据集与现成领域数据包 | 本土模型竞争力诉求及补齐特定垂直领域能力短板 |
| 早期试水企业 | 首席数据官 / 企业 AI 负责人 | 内部软件工程师与业务专家 | 业务部门数字化转型预算 | 金融、法律与医疗领域的高价值垂直工作流 | 由实验性聊天机器人试点向生产级自主智能体工作流演进 |
买方画像与采购动态来自客户公开披露及前沿 AI 部署的 B2B 销售分析。
[CM002, CM008, CM009, CM010, CM011]2.4 采用驱动因素、结构性制约与证据缺失
AfterQuery 目标市场的扩张受到多重结构性顺风的叠加驱动,但也受到重大运营与商业落地瓶颈的制约。从增长驱动力来看,基础模型预训练架构迅速同质化,促使后训练数据与评测环境成为差异化竞争的核心支点。公司与 Nvidia 在 Nemotron 3 Ultra 训练数据集上达成的战略合作印证了这一点:硬件生态与模型开发者正日益依赖专业数据合作伙伴。然而,潜在的结构性风险同样突出。商业需求目前高度集中于极少数顶尖前沿实验室,导致单笔合同体量庞大且交付节奏不均,加剧了营运资金波动。此外,针对数据出口的潜在地缘政治与监管限制(例如据推算来自中国 AI 科研实验室的经常性收入)可能冲击国际收入线。尤为关键的是,AfterQuery 的真实毛利率概况与承包商转包分成机制在现有商业报道中均未公开,留下了重大的分析空白。[CM001, CM002, CM010, CM011, CM014]
2.5 重点图表
03竞争格局
3.1 竞争格局与直接同行
前沿 AI 训练数据市场已分化为走量型数据工厂与专业化专家数据平台两大阵营。Scale AI 等老牌数据供应商依托覆盖广泛的多模态标注业务、深厚的联邦政府及商业企业客户关系,以及超过 $1.6B 的庞大融资规模牢牢占据市场底座。2026 年初,Scale AI 推出专用强化学习环境,印证了全行业从静态提示-响应标注向交互式工作流执行的转变。与此同时,Mercor 与 Surge AI 等专业挑战者作为 AfterQuery 的直接同行迅速崛起,直接向顶尖人工智能实验室输送顶尖专业人才及特定领域评测环境。Mercor 据报已渗透多数顶级基础模型开发者,将自身定位为承接复杂推理任务的核心专家数据集供应商。此外,Labelbox 等更广泛的平台依托 Alignerr 等专业人才分支加速拓展,Turing 等人才交易平台亦提供庞大的开发者资源池与合成数据生成流程。Tracxn 的综合市场情报显示,数据及数据集领域活跃的竞争对手超过 240 家,既反映出旺盛的商业需求,也表明基础标注服务准入门槛极低。[CP001, CP002, CP003, CP007, CP010]
| 竞争对手 | 类别 | 规模与融资 | 目标客群 | 核心差异化 | 已知局限 |
|---|---|---|---|---|---|
| Scale AI | 头部老牌平台 | $1.6B+ 累计融资 / $13.8B 估值 | 前沿实验室与企业级 AI | GenAI Data Engine 及全面的企业客户布局 | 传统众包标注的成本结构与同质化大宗商品化趋势 |
| Mercor | 直接垂类竞争对手 | 获 Series A 轮融资 / $250M+ 估值 | 顶级基础模型实验室 | 高水准人类专家人才与评估环境 | 较少侧重定制自研沙箱软件工具链 |
| Surge AI | 直接垂类竞争对手 | 风投支持 / 非公开 | 模型实验室与科技企业 | 精英级 RLHF 与基准排行榜影响力 | 财务披露不透明且精品项目吞吐量有限 |
| Labelbox (Alignerr) | 相邻领域平台 | 累计融资 $189M(Series D) | 模型实验室与企业级 AI | 旗下 Alignerr 平台汇聚 1.5M+ 知识工作者 | 历史积累主要集中在计算机视觉与标注工具链 |
| Turing | 人才市场平台 | 累计融资 $140M+(Series D) | 企业级工程团队与 AI 实验室 | 庞大的全球开发者资源池与合成数据包 | 非代码类专业垂直领域的沉淀较浅 |
规模与融资指标汇总自 Sacra 股票研究、Tracxn 风投资讯及 TechCrunch 截至 2026 年的报道。
[CP001, CP002, CP003, CP007, CP010]3.2 能力差异化与定价机制
区别于传统数据供应商,AfterQuery 将业务重心放在专业过程推理与多轮工作流执行上,而非简单的文本分类或事实问答。传统供应商依赖宽泛的众包标注,AfterQuery 则依托自研软件基础设施,将经过认证的执业医生、企业律师及量化工程师等专业人员组织进模拟运行环境。公司公布的研究成果强调,若缺乏特定领域的行为约束,标准基础模型在 FinanceQA 等评测的真实专业工作流中,失误率最高可达 60%。公司平台借助 Docker 容器、浏览器界面与开发者 API 构建沙盒环境,供模型学习工具调用时序、输入验证及错误恢复。与按席位计费的标准 SaaS 模式不同,专家数据行业的商业变现主要以程序化与项目制为主。合同定价由专业领域复杂度、专家资历深浅以及环境模拟要求共同决定。尽管这种机制为来自 Nvidia、Legora 及 Motif Technologies 等客户贡献了可观的单笔合同金额,但相比纯软件授权,也带来了高昂的实际交付成本与收入不稳定性。[CP004, CP005, CP006, CP008, CP009, CP012, CP013, CP014]
| 采购考量标准 | AfterQuery | Scale AI | Mercor | Labelbox (Alignerr) |
|---|---|---|---|---|
| 定制 RL 沙箱(Docker/API) | 核心重点(Docker、API、浏览器环境) | 专属 RL 环境(2026 年初上线) | 定制评估与 RL 环境 | 定制基准测试环境 |
| 专业垂直领域专家(医疗 / 金融) | 100k 认证在职专业人士 | 经过严格审核的专业领域专家网络 | 顶尖人类专业知识网络 | 聚集博士与专家的 Alignerr 网络 |
| 公开基准测试支持(FinanceQA/VADER) | 已在 FinanceQA、VADER、IDE-Bench 得到验证 | 广泛的 SEAL 排行榜与评估测试 | 布局公开基准测试与竞技场(Arena) | 专属 AGI 基准测试与竞技场评估 |
| 自研工作流工具链 | 自研软件优先型管理平台 | GenAI Data Engine 工具套件 | 定制沙箱与交付框架 | 数据工场运营软件 |
| 自助式企业级 SaaS 订阅 | 无(仅提供顾问式企业级大客户合作) | 企业平台与开发者自助服务 | 直接高接触企业大客户销售 | 可提供软件平台授权 |
能力对齐情况综合自各厂商产品公告、Sacra 市场报告及 PromptLoop 平台分析。
[CP001, CP002, CP003, CP004, CP006, CP008, CP013, CP014]| 供应商 | 合同模式 | 包含能力 | 折扣与未披露信息 | 战略含义 |
|---|---|---|---|---|
| AfterQuery | 规划性与按项目确定交付范围 | 定制 SFT 轨迹、RL 环境、领域评估 | 阶梯采购定价未披露;按项目收取定制费用 | 结果导向型定价带来单笔大额合同,但收入波动较大(lumpy) |
| Scale AI | 年度企业级保底承诺 + 按量计费 | Data Engine、RL 环境、全托管标注 | 大规模运行任务享受企业级分阶折扣 | 多年度平台承诺绑定带来极高的切换成本 |
| Mercor | 按项目计费与常年专家顾问服务 | 基准测试、RL 环境、人类专家数据 | 内部费率表;未披露按小时 / 按任务加价比例 | 在专业领域知识合作上展开直接价格竞争 |
| Labelbox | 软件订阅 + 数据服务增购附加项 | 平台工具链、Alignerr 网络准入、评估测试 | 平台阶梯折扣;弹性浮动的专家人工费率 | 软件与服务混合模式提供可预期的保底基准费用 |
商务条款来自 Sacra 市场分析及行业采购基准;具体客户费率表仍属保密范畴。
[CP001, CP002, CP003, CP009, CP012]3.3 护城河韧性与战略替代风险
AfterQuery 竞争优势的持久性,取决于其专家网络的稳固程度以及强化学习沙盒的技术复杂度。鉴于各领域专家属于非排他性脑力工作者,可在 Mercor、Surge AI 或 Scale AI 等多家平台之间兼职,单纯的贡献者准入渠道并不构成难以攻破的护城河。公司的防御壁垒实则源自专有软件工具链、工作流插桩机制与验证规则体系,正是这些要素构筑了将专家推理转化为高信噪比奖励模型的能力。尽管如此,AfterQuery 仍面临双重结构性替代威胁:其一,前沿人工智能实验室既是公司最大的客户群,也是最大的脱媒风险来源;顶尖科研实验室具备充足的技术储备,一旦认为外部供应商利润率过高,便有能力将专家标注流程收归自研。其二,合成数据生成与 Agent 评测框架快速成熟,使竞争对手得以以更低的边际成本批量生成训练样本。化解这些挑战,要求 AfterQuery 持续向更复杂、受监管的专业领域攻坚,而在这些深水区,合成模拟依然无法复现真实严谨的专业判断。[CP001, CP002, CP004, CP005, CP011]
AI 数据提供商在领域工作流专业化(X 轴)与 RL 环境复杂度(Y 轴)维度的序数定位。
| 条目 | X (5.5–9) | Y (5–8.5) | 说明 |
|---|---|---|---|
| 1. AfterQuery | 9 | 8.5 | 高度专业化的专家工作流与定制 API/终端 RL 沙箱 |
| 2. Scale AI | 7.5 | 8 | 拥有广泛的平台规模,并于 2026 年推出专用 RL Environments |
| 3. Mercor | 8.5 | 7.5 | 精英人才市场,正向 RL 环境与评估扩展 |
| 4. Surge AI | 8 | 6.5 | 高质量 RLHF 与基准设计,具备精品服务能力 |
| 5. Labelbox | 6.5 | 6 | Alignerr 专家网络与数据工厂软件深度整合 |
| 6. Turing | 5.5 | 5 | 海量开发者人才库与合成数据流水线 |
评分为 1-10 分序数评估,基于 Sacra 分析、PromptLoop 画像及技术产品公开披露。
[CP001, CP002, CP004, CP005, CP011]3.4 重点图表
04财务情况
4.1 收入来源与定价体系
AfterQuery 采用 B2B 商业模式,主要服务于前沿基础模型机构内部的后训练团队、评测工程师及企业 AI 部门。公司不提供原始网络抓取数据或自动化粗标,而是将复杂的人类推理轨迹与专用沙盒环境打包交付。独立行业研究机构 Sacra 的报告显示,AfterQuery 的收入来源细分为四大产品层:定制强化学习环境、定制监督微调(SFT)推理数据集、标准化评测套件以及企业诊断咨询。定制强化学习环境启动容器化软件栈(如终端沙盒和浏览器界面),供基础模型依据自动化奖励信号执行工具调用序列。由于这些项目针对金融、法律和医疗等领域的复杂失效模式,合同变现主要采取基于项目的程序化收费,而非纯粹按席位计费的 SaaS 模式。业务范围涵盖专业难度、所需专家资历、环境复杂度以及交付 token 数量,促成了高客单价的定制化交付。[CI002, CI004, CI005, CI006]
| 收入来源 | 变现机制 | 销售单元 | 当前商业化现状 | 收入质量评估 | 尽调核查要求 |
|---|---|---|---|---|---|
| 定制 RL 环境 | 按里程碑结算的项目合同 | 按环境部署与 API 访问量计费 | 前沿实验室已形成实质商业采购 | 单价高但属于非经常性工程服务 | 审查合同条款、续约频率与知识产权(IP)归属约定 |
| 监督微调(SFT)数据 | 按数量与复杂度分级的数据包 | 按提示-响应对(prompt-response pair)与推理轨迹量计费 | 主要大客户的核心收入来源 | 总收入规模大,但专家薪酬支出构成可变营业成本(COGS) | 调阅专家报酬方案与单位数据交付毛利率 |
| 标准化现成评估数据包 | 标准化软件授权与基准测试订阅 | 按基准测试套件或评估席位收费 | 尚处起步阶段的标准化产品变现层 | 高毛利的复用型数字资产库存 | 核验经常性续约率与产品目录扩充速度 |
| 企业级 AI 诊断咨询 | 常年顾问与咨询服务收费 | 单次项目制的诊断审计与实施交付 | 特定企业试点项目(如 The Raine Group) | 人力密集型服务收入,可扩展性偏低 | 评估诊断试点向程序化数据交易的转化率 |
收入流分类与变现结构汇编自 Sacra 研究报告及公司公告;合同级定价仍属机密。
[CI004, CI005, CI006, CI012]4.2 市场拓展打法与合同商业模型
AfterQuery 的市场拓展(GTM)打法显著有别于传统企业软件销售及初级数据标注平台。其需求挖掘立足于实证 AI 科研论文以及 FinanceQA、IDE-Bench、Terminal-Bench 等公开基准评测套件。公司发表严谨的技术研究,证实主流商用基础模型在复杂的真实专业工作流任务中失误率高达 60% 以上,直接与前沿 AI 实验室的研究科学家及后训练负责人建立起技术互信。这种以科研为主导的销售模式,将采购数据集包装为攻坚已知基准缺陷的量化解决方案,大幅降低了采购阻力。企业级交付依托近 100,000 名经过认证的软件工程、医疗和法律执业专业人员网络。至关重要的一点是,AfterQuery 安排内部研究人员在交付前对所有人为策划的数据集开展预校验,涵盖模型微调与基准指标验证。这种软件先行的校验闭环确保训练数据处于严格的难度区间,使前沿模型能够顺利吸收,且不破坏原有的推理稳定性。[CI005, CI007, CI008, CI014]
| 变现维度 | 合同定价模式 | 刊例价与实际成交透明度 | 折扣与未知因素 | 信息来源归属 |
|---|---|---|---|---|
| 领域技术复杂度 | 按专业领域(金融、医学、法律)分档定价 | 依据内部费率表定制项目范围 | 受专家薪资溢价与稀缺专才供给限制 | Sacra 行业分析与 Y Combinator 档案 |
| 专家资质构成 | 按从业者资质实行差异化报酬 | 采用非公开项目报价,未设公开刊例价 | 博士与执业资格专家的时薪成本存在差异 | 公司官方博客与新闻稿 |
| 交互式沙箱深度 | 基础设施算力费加测试套件复杂度费 | 计入定制环境合同总报价 | 云容器算力成本与延迟 SLA | Sacra 产品拆解分析 |
| 公开基准发布 | 作为销售赋能资产,不直接向客户收费 | 完全公开的研究发表物 | 辅助客户线索生成的间接成本中心 | Forbes 报道与公司公告 |
据 Sacra 报道,定价条款反映了咨询式 B2B 合同范围;每专家工时费率表与每 token 基准的确切数据未见公开披露。
[CI005, CI007, CI008, CI014]4.3 商业转化流程与交付桥梁
将客户需求转化为商业收入与毛利遵循多阶段技术流程。客户合作始于失效模式诊断分析,由 AfterQuery 与实验室研究人员共同锁定目标专业领域的特定推理瓶颈。技术参数明确后,公司将工作流需求与经过认证的行业专家进行匹配,由专家撰写结构化推理轨迹与多步决策序列。沙盒强化学习环境运行自动化验证器与测试套件,确保生成的数据符合专业标准并提供明确的奖励信号。验证完成后,数据集与环境既可以作为专有定制包交付,也可以并入可复用的评测基准。开票节点随数据交付与基准验证触发,对照既定交付节点产生即时现金流。然而,支付给执业律师、金融分析师及专科医生等资深行业专家的直接报酬,构成了直接主营业务成本(COGS),并随合同规模同比例增加。[CI006, CI007, CI008, CI009, CI011]
梳理商业转化流程:前沿实验室研发需求如何转化为定制专家数据清洗、里程碑验证以及最终实现的毛利。
| 节点 / 连接 | 节点 / 起点 | 终点 | 说明 |
|---|---|---|---|
| 节点 1 | 1. 前沿实验室失效模式识别 | ||
| 节点 2 | 2. 圈定领域、专家层级与环境范围 | ||
| 节点 3 | 3. 在 100k 专家网络中匹配对口专家 | ||
| 节点 4 | 4. 沙箱内 RL 工具调用与合成验证 | ||
| 节点 5 | 5. 向客户实验室交付高保真数据集 | ||
| 节点 6 | 6. 按项目制及里程碑合同开票 | ||
| 节点 7 | 7. 确认毛利(扣除专家人力支出) |
展示 Sacra 梳理的商业服务交付工作流;利润率实现比例代表预估区间,受专家报酬浮动影响。
[CI006, CI007, CI008, CI011]4.4 资本充裕度、单体经济模型与尽调结论
AfterQuery 展现出亮眼的账面资本效率,从 2025 年初上线起步,仅凭截至 A 轮累计获得的 $30.5M 一级市场风投资金,年化收入运行率据报道便达到 $100M。2026 年 4 月 9 日,公司完成由 Altos Ventures 领投,The Raine Group、Y Combinator 与 BoxGroup 跟投的 $30M A 轮融资,投后估值达 $300M。5 个月后的 2026 年 9 月,媒体报道显示新一轮融资对公司的估值达到 $3.2B;联合创始人 Spencer Mateega 声称经常性收入已加速迈入数亿美元区间,且经营性现金流已转正。尽管增长势头亮眼,对 AfterQuery 进行投资判断仍须审视实质性的单体经济模型缺陷与结构性隐忧。首当其冲的是成本结构:不同于受益于全球劳动力套利的众包数据标注平台,AfterQuery 对资深专业精英的依赖带来了高昂的边际交付成本,制约了毛利率的扩张空间。此外,收入依然高度集中于少数前沿模型实验室,导致业务对实验室训练预算及自研决策产生严重依赖。[CI001, CI002, CI003, CI010, CI011, CI012, CI013]
| 经济指标 | 披露值 / 估算值 | 评估置信度 | 投资判断影响 | 尽调核查诉求 |
|---|---|---|---|---|
| 年化收入运行率 (ARR) | 截至 2026 年 4 月据报达 $100M+ | 高(公司披露,多方信源报道) | 展现出前所未有的商业化推进速度 | 核查历史月度收入确认情况与合同排期 |
| 毛利率 | null | 低(机密非公开指标) | 受人类专家报酬与计算沙箱开销挤压 | 获取按收入流细分的 GAAP 审计毛利率明细表 |
| 全职员工人数与杠杆人效 | 30 名员工(2026 年中估算) | 中(分析师追踪,非监管申报信源) | 依托外包专家网络撬动了极高的人均 ARR 产出 | 对比审查全职组织架构图与活跃 1099 专家网络 |
| 累计风险投资额 | 截至 A 轮披露为 $30.5M | 高(监管文件与公司公告确认) | 相比 $100M+ 收入规模,股权资金消耗极低 | 核实账面现金、信贷额度以及 2026 年 9 月融资款项 |
指标反映公开披露与分析师估算;毛利率、净收入留存率 (NRR) 及获客回本周期属于机密指标,需签署保密协议方可获取。
[CI001, CI002, CI009, CI010, CI011, CI013]4.5 重点图表
05产品与技术
5.1 产品定义与工作流整合
AfterQuery 专门构建数据基础设施层,旨在训练基础模型像资深专业人士一样行动,而非沦为静态的问答引擎。公开网络数据虽能让模型完成文本自动补全并通过标准单选题测试,但在面对复杂的多步专业工作流时,模型往往频繁出错。在缺乏场景约束的环境中,模型容易臆造事实参数、编造标识符,或打乱外部工具的调用次序。针对这一结构性缺陷,AfterQuery 策划数据集与交互式训练环境,教导模型掌握专业工作的实际展开方式:搜集初始识别参数、确立正确的工具调用顺序、在模糊情境下维护状态,以及从系统异常中平稳恢复。这种过程化训练方法,为量化金融、复杂代码重构以及企业交易尽调等核心用例提供了坚实底层支撑。[CE001, CE004, CE005, CE006]
| 领域与工作任务 | 传统模型失效模式 | AfterQuery 解决方案 | 可量化收益 | 技术局限 |
|---|---|---|---|---|
| 财务分析与尽职调查 | 伪造数值且财务计算顺序错乱 | 提示词-响应推理轨迹与 FinanceQA 验证 | 减少多步计算与电子表格错误 | 需持续更新以适应不断演进的会计准则 |
| 软件工程与代码重构 | 生成语法有效但在运行时边界场景失败的代码 | Terminal-Bench 与 Docker 容器化测试套件执行环境 | 经真实编译器与单元测试套件检验的执行验证 | 容器启动延迟与沙箱隔离资源争用 |
| 客户业务流程导航 | 伪造订单编号与占位邮箱地址 | 交互式 API 与 MCP 服务器工具调用轨迹训练 | 模型学会执行工具调用前先查询身份识别信息 | 依赖客户 API 稳定性与逼真的 Mock 数据库状态 |
| 法律文档尽调 | 总结法律文本却未能识别不利合同条款 | 专家整理的修订标记、合同批注数据集与法律问答 | 识别复杂交易记录中决定交易成败的重大阻碍条款 | 高度依赖稀缺持证执业律师完成初始数据标注 |
工作流失效模式与纠偏训练机制记载于 Sacra 分析报告及官方产品披露中。
[CE001, CE004, CE005, CE006]5.2 模块化架构与沙盒设计
核心产品体系划分为三个互补的结构层级:监督微调(SFT)推理数据集、强化学习(RL)执行沙盒以及特定领域评测基准。SFT 层提供密集提示-响应对,并辅以详尽的逐步推理轨迹,在进入强化学习前为模型注入基础专业规范。强化学习层构成了 AfterQuery 的核心技术护城河,包含依托真实操作系统接口、开发者工具、浏览器、外部 API 及模型上下文协议(MCP)服务器构建的动态隔离沙盒。这些环境并未依赖虚构的合成模拟,而是启动崭新的 Docker 容器,供模型执行终端命令并对照实时单元测试套件编译代码,将真实的执行结果转化为确定性奖励信号。评测层则引入了 FinanceQA 与 IDE-Bench 等专有基准,用以诊断模型在真实企业任务中的失效模式。[CE002, CE003, CE004, CE005, CE006, CE011]
| 模块 / 产品 | 目标买家 / 用户 | 成熟度与部署情况 | 核心技术差异化优势 | 尽调盲区 |
|---|---|---|---|---|
| 监督微调 (SFT) 数据集 | 前沿 AI 实验室后训练研究团队 | 生产环境(金融、代码、法律、医学) | 复杂多步推理轨迹与专业领域问题解答 | 样本级颗粒度单价与专有评测评分标准归属 |
| 强化学习 (RL) 沙箱 | Agent 开发与后训练团队 | 跨终端与 API 环境的生产部署 | 配备实时测试套件的 Docker 容器化执行沙箱,用于提供奖励反馈 | 持续长步长推演带来的算力开销与扩展成本 |
| 领域评估基准 | 模型评估与对齐团队 | 生产环境(FinanceQA、IDE-Bench、Market-Bench) | 替代死记硬背公开数据集的真实专业任务故障诊断 | 随着前沿模型指标饱和,基准重置基准线的频率 |
| 企业定制工具与工作流 | 企业 AI 团队(如 The Raine Group) | 早期企业生产部署 / 高接触试点 | 针对企业特定知识瓶颈定制的 API 与数据库实据对接 | 咨询服务模式向标准化现成软件转型的可扩展性 |
模块分类综合自 Sacra 行业研究与 AfterQuery 官方技术出版物;定价与交付条款仍属保密。
[CE002, CE003, CE004, CE006]分层架构,涵盖专家搜寻、定制工作流搭建、隔离执行沙箱与前沿模型评估。
- 专家网络层
- 涵盖法律、金融、医疗与工程领域的约 100k 名经认证从业者
- 数据摄取与清洗
- 自研人在回路任务创建与审核软件
- 程序化 SFT 引擎
- 结构化推理轨迹与多轮专业对话数据对
- 交互式 RL 环境
- Docker 容器化执行沙箱、终端控制台与 MCP API 服务器
- 自动化奖励套件
- 确定性测试套件执行、实时编译器与反馈闭环
- 评估与基准测试
- 领域专属测试套件,涵盖 FinanceQA、IDE-Bench 与 Market-Bench
- 企业级安全与基础设施
- 部署于 GCP 与 AWS 之上、符合 SOC 2 Type II 及 ISO 27001 标准的云基础设施
分层架构提炼自技术招聘要求及研究报告中的系统架构概念。
[CE002, CE004, CE005, CE009, CE010, CE011]5.3 技术架构、安全性与合规性
交付高保真数据方案需要一套具备弹性的技术基础设施,连接全球专家贡献者网络与多租户模型实验室部署。AfterQuery 借助专属专家门户,统筹近 100,000 名跨越医疗、法律、软件工程和投资银行领域的认证专业人士。内部工程技术栈采用 Python、Go、TypeScript,并依托 AWS 与 Google Cloud Platform 上的现代化容器编排,管理数据流水线与人机协同(human-in-the-loop)标注工作流。由于模型实验室项目涉及高度机密的客户代码库、专有量化交易策略以及尚未发布的前沿模型架构,安全性是决定能否通过采购审核的生死门槛。公司持有 SOC 2 Type II 与 ISO 27001 认证,同时部署程序化身份与访问控制管理、严格的网络隔离边界以及自动化漏洞监控,确保企业级项目间实现彻底的数据隔离。[CE007, CE008, CE009, CE010, CE012, CE014]
| 架构层 | 核心职能 | 底层技术与依赖项 | 关键技术 / 运营风险 |
|---|---|---|---|
| 专家贡献者界面 | 入职引导、任务分配及依据评分标准的标注 | Web 门户 (experts.afterquery.com) 与定制工作流工具 | 分散的专业贡献者之间存在质量差异 |
| 执行与沙箱层 | 适用于工具调用 Agent 轨迹的容器化运行时环境 | Docker 容器、GCP/AWS 基础设施、终端界面、MCP 服务器 | 运行不受信任代码时的安全漏洞或沙箱逃逸 |
| 安全与基础设施管控 | 访问权限管理、遥测与客户知识产权隔离 | Python、Go、TypeScript 服务;IAM 访问边界;SOC 2 Type II 控制措施 | 跨竞争前沿实验室业务发生多租户数据隔离失效 |
| 评估与奖励基础设施 | 自动化单元测试验证与基于评分标准的打分 | 自动化测试框架、单元测试运行器、领域打分验证器 | 模型利用自动验证器启发式规则实施奖励作弊 (Reward hacking) |
架构组件汇总自 Sacra 分析档案与技术基础设施工程披露信息。
[CE005, CE007, CE009, CE010, CE012]5.4 竞争差异化与技术壁垒
AfterQuery 摒弃同质化的海外众包模式,坚持采用由经过认证的行业精英驱动、以自研软件为核心的路线,从而与传统数据标注平台拉开差距。多数竞品依赖劳动力套利与宽泛的人才资源池,AfterQuery 则打造定制化内部工具,用以捕捉纯合成数据生成所无法复制的隐性专业判断与深层领域推理。然而,这一市场定位也蕴含显著的战略与技术风险。公司的客户群高度集中在前沿 AI 实验室,这些实验室坐拥庞大的算力资源与内部后训练研发团队。倘若前沿实验室自行研发沙盒工程,或转向基于自我博弈的合成数据流水线,AfterQuery 必须证明人机协同的专家行为约束依然是高风险企业级 Agent 落地不可或缺的核心卡点。[CE011, CE013]
5.5 重点图表
06客户情况
6.1 客户细分与买家画像
AfterQuery 的商业合作模式主要构建于两大客群层级之上:前沿人工智能基础模型实验室,以及部署垂直领域 Agent 的大型企业软件机构。在基础模型层,买家是专业的后训练团队、强化学习研究小组及模型评测部门。这些买家需要密集、高水准的专业推理轨迹,以突破网络抓取预训练语料的固有极限。前沿科研买家并不采购通用数据标注服务,而是需要量化金融、软件工程、医疗和公司法等复杂垂直领域的定制任务完成数据与模拟强化学习环境。在企业客群层,买家多为打造内部开发者 Copilot 与工作流自动化工具的首席技术官及产品工程负责人。该细分领域的商业合作聚焦于降低特定领域的模型幻觉,并规范多步工具调用的执行纪律。尽管企业端需求增长迅速,但企业买家的采购阻力更大,在向外部专家评审开放内部专有工作流之前,往往要求获得严格的 SOC 2 Type II 与 ISO 27001 数据合规认证保障。[CU003, CU006, CU007, CU014]
| 客户群体 | 买方 / 付费方 | 主要用例 | 规模与采纳指标 | 收入 / 战略价值 | 尽调盲区 |
|---|---|---|---|---|---|
| 前沿 AI 实验室 | 后训练与基础模型研究团队 | 为核心推理基准构建专业推理轨迹与 RL 环境 | 参与度高;据报道各大主流 AI 实验室均在数十个领域使用其数据集 | 核心收入引擎;合同金额高($1M-$10M+ 定制项目) | 客户集中度风险;实验室可能自建专家标注管线 |
| 企业 AI 与开发者工具研发商 | 企业 AI 产品负责人与工程技术管理层 | 领域专属 Agent 工具集、代码重构与自动化测试评估环境 | 采纳度持续提升;据报道已在开发者工具栈中落地早期标杆部署 | 单客合同金额适中;具备拓展常态化软件工作流的潜力 | 实际生产部署规模与概念验证 (PoC) 实验界限不明 |
| 专业金融与法律机构 | 企业创新负责人、量化基金与并购法律合伙人 | 复杂交易尽调、并购条款分析与财务推理数据集 | 试点项目与领域合作,涵盖财务顾问网络 | 对经验证的准确率付费意愿强;行业准入门槛高 | 未见公开披露的常态化软件许可协议;高接触咨询交付存在阻力 |
| 医疗健康与临床 AI 实施方 | 临床 AI 负责人与医学信息学总监 | 诊断推理轨迹、临床工作流模拟与多模态医学评估 | 处于早期探索阶段;对经临床医生验证的标注数据集需求迫切 | 垂直领域长期目标市场空间 (TAM) 价值高;合规监管壁垒深厚 | 面临严苛的合规与验证障碍;采购决策周期漫长 |
客户细分综合了据报道的前沿实验室合作、产品文档及分析师市场数据。财务贡献预估属于定性评估。
[CU003, CU006, CU007, CU014]6.2 业务采纳路径与营收增速
自入选 Y Combinator 2025 年冬季批次以来,AfterQuery 展现出前所未有的商业扩张速度。公司宣布在 2026 年 4 月年化收入运行率突破 $100M,距其创立仅过去不足 14 个月的商业化运营时间。2026 年 9 月,风险投资市场对这种极高的扩张速度给予了回报,据报道 AfterQuery 完成新一轮私募融资,公司估值达到 $3.2B。该笔交易使 AfterQuery 成为 Y Combinator 历史上自创立起最快达到独角兽估值的初创公司,将传统企业级软件的扩张周期缩短了一个数量级。然而,对于这种高调的营收增长势头,在分析时必须保持审慎。财务分析师与数据追踪机构强调,年化收入运行率仅属于管理层基于当前业务的外推预测,而非经审计的 GAAP 历史营收。至关重要的一点是,公开商业追踪平台证实,AfterQuery 尚未披露任何经审计的客户数量数据、活跃企业客户总数或平均客单价(ACV),这使得支撑这笔 $100M 运行率的实际付费客户规模仍未经证实。[CU001, CU002, CU008, CU011]
| 采纳指标 | 披露数值 | 截至日期 | 主要信源 | 置信度 | 战略意涵 | 缺失基数 |
|---|---|---|---|---|---|---|
| 年化收入运行率 (ARR) | $100M+ | 2026-04 | TechCrunch / AfterQuery 官方博客 | 高 | 极高的收入爆发速度,印证了对高端后训练数据的迫切需求 | 未说明总客户账户数与合同期限 |
| 投后估值基准 | $3.2B | 2026-09 | TechCrunch / The Entrepreneur Story | 高 | YC 史上最快晋升独角兽(距 A 轮仅 5 个月);反映出投资者对 AI 数据护城河的信心 | 净现金消耗与持续转化为 GAAP 收入的能力未经证实 |
| 经认证专家网络规模 | ~100,000 | 2026-04 | AfterQuery 官方博客 | 中 | 具备跨工程、法律、医学承接并行模型评估的广泛专业能力 | 未披露月度活跃贡献者与注册资源池的比例 |
| 活跃企业客户数 | 2026-08 | Latka 分析师档案 | 高 | 公开信源称客户数量未作记录;折射出极高的客户集中度 | 独立付费客户数量与持续扩张的前沿实验室合同对比不明 |
增长轨迹指标结合了公司公布的运行率里程碑与第三方分析师追踪数据。Null 表示公开数据未作记录。
[CU001, CU002, CU008, CU011]6.3 标杆客户印证与企业级落地
公开披露与独立科技媒体的报道证实了公司在硬件系统和基础模型研发领域的数个重磅客户关系。半导体与 AI 算力巨头 Nvidia 已被证实为正式付费客户,采用 AfterQuery 的推理数据赋能企业级 Agent 开发。同样,企业软件平台 Legora 与韩国知名顶尖基础模型实验室 Motif Technologies 也已完成经过验证的生产环境部署。除技术类客户外,AfterQuery 还在金融咨询领域建立起商业战略合作,例如与 The Raine Group 持续合作诊断内部金融交易工作流,并搭建针对该机构的专用评测环境。尽管斩获了上述标杆客户,更广泛的客户名录仍受制于严格的商业保密协议。公开证据尚未披露公司与前沿实验室的合作究竟属于企业级主服务协议(MSA),还是仅为孤立的、按项目计费的功能突击开发。由于缺少包含量化投资回报率(ROI)指标且由客户背书的案例研究,实际生产工作流的渗透深度与后续使用率仍是核心尽调焦点。[CU004, CU005, CU008, CU014]
| 客户名称 | 客户群体 | 部署 / 用例 | 生产部署与试点阶段 | 据报成效 / 交付价值 | 采纳局限 / 风险 |
|---|---|---|---|---|---|
| Nvidia | 企业 AI 与硬件系统 | 面向企业基础设施的模型训练与 AI Agent 推理数据 | 生产部署 | 主流科技行业报道引述了经核实的客户部署案例 | 未披露具体合同金额、项目范围及多年期承诺协议期限 |
| Legora | 企业级 AI 软件 / 应用程序 | 针对专业工作流的领域专属任务仿真与任务达成数据 | 生产部署 | 报道 A 轮扩张的独立科技媒体提及了该具名签约客户 | 平台集成深度及向多条产品线延伸的情况未经证实 |
| Motif Technologies | 前沿 AI 研究实验室(韩国) | 基础模型能力基准与后训练数据集 | 生产部署 | 获得亚洲前沿 AI 市场跨国研究实验室客户的验证 | 未披露合同总金额及对单一研究项目的依赖程度 |
| The Raine Group | 财务顾问 / 投资合作伙伴 | 财务工作流诊断、尽调自动化以及机构专属 Agent 基础设施 | 商业合作 / 试点 | 验证了知识优先的诊断打法,可跨金融行业复用 | 相较公平交易客户,战略投资者关系可能对商业条款产生倾斜 |
具名客户引用代表独立报道 (TechCrunch) 中经核实的信息披露。受商业机密限制,覆盖案例仅为样本。
[CU004, CU005, CU009, CU014]6.4 业务持久度、扩张机制与客户集中风险
AfterQuery 客户关系的长期持久度,取决于其能否从咨询式项目交付商转型为深度嵌入的经常性基础设施层。当前,商业变现主要依托基于项目的定价机制,范围取决于专业领域难度、专家薪酬、环境复杂度及交付数据集规模。这一定价模式虽然造就了庞大的初期合同体量,却也使公司面临极高的收入集中度。公司的营收高度依赖少数资金充裕的前沿 AI 实验室,这些机构既具备支撑八位数美元合同的资金实力,也拥有将数据策划流程自研收编的技术储备。一旦主要基础模型实验室选择自主开发后训练数据策划,或转向混合合成数据管道,AfterQuery 将面临严峻的客户流失风险。此外,Scale AI 等平台型老牌企业,以及 Mercor、Surge AI 和 Labelbox 等灵活的专家网络,也正在加剧竞争压力。为抵御对前沿实验室的依赖,AfterQuery 必须系统推进其「拓展渗透(land-and-expand)」战略,将初期的基准评测套件逐步转化为永久性、多领域的深度工作流模拟环境,进而深度融入企业工程技术体系。[CU006, CU007, CU009, CU010, CU012, CU013]
客户从识别能力短板、部署定制强化学习环境到多领域拓展的推进路径。
综合了公司披露与 Sacra 分析中所述的客户导入工作流。
[CU003, CU007, CU008, CU012]6.5 重点图表
07风险
7.1 监管、法律与合规风险登记册
AfterQuery 处于行业专业知识与生成式模型后训练的交汇处,这在知识产权、数据保护及专业执业规范边界上带来了独特的法律与监管风险。在公司法、证券和医疗健康等敏感垂直领域,公司必须确保严格的引用完整性与司法管辖权准确性,防范生成式模型产生看似言之凿凿的幻觉。随着企业级客户规模不断扩大,保持 SOC 2 Type II 与 ISO 27001 认证已成为不可或缺的准入门槛,而非可有可无的常规合规。此外,管理分布于全球的专家贡献者网络,要求公司持续符合独立承包商分类规范,并建立清晰的知识产权转让框架,以确保所有交付的数据集与评测套件拥有完整无可争议的权属链(chain-of-title)。[CR001, CR002, CR003, CR004]
| 规则 / 要求 | 司法管辖区 | 现状 | 发生概率 | 严重程度 | 缓解措施 | 残余风险敞口 | 尽调核查路径 |
|---|---|---|---|---|---|---|---|
| 多司法管辖区法律引文准确性与评测 | 美国(联邦与州) | 积极招募持证执业律师以构建验证评分标准 | 中 | 高 | 聘请专业法律负责人审计法律推理数据集与评分标准 | 法律数据集中模型幻觉可能给企业客户带来法律责任 | 审查法律任务数据集的评分标准与错误率基准 |
| SOC 2 Type II 与 ISO 27001 数据安全认证 | 全球 / 企业商业 | 已作为企业采购交易的前置硬性准入条件落地 | 低 | 高 | 维持处理专有代码库及客户工作流的认证管控体系 | 年度审计失效或访问控制失陷将阻断企业级业务扩张 | 获取当前的 SOC 2 Type II 审计报告及 ISO 27001 证书 |
| 专家贡献者数据隐私与外包雇佣定性 | 加州 / 美国 / 国际 | 通过 AfterQuery Experts 专属隐私框架与门户进行管理 | 中 | 中 | 标准化专家协议、知识产权转让条款及体系化隐私政策 | 劳务用工定性诉讼或跨境专家数据传输挑战 | 审查外包劳务合作协议与知识产权转让条款 |
| 模型训练版权与专有数据摄入 | 美国联邦 / 欧盟人工智能法案 (EU AI Act) | 重点聚焦人类原创推理轨迹与仿真 RL 环境 | 低 | 中 | 内部自研工作流生成,避免使用第三方抓取的训练数据 | 监管层对企业训练数据来源与授权许可的严格审查 | 审查第三方评估数据集的权属链条证明文件 |
按严重程度排序的监管、法律与合规义务登记表;结合公开合规态势、招聘信息与数据保护架构综合评估。
[CR001, CR002, CR003, CR004]7.2 运营失效模式、机器人硬件与质量保障
运营稳健性是 AfterQuery 科研驱动型定位的核心基石。在认知智能领域,主要失效模式集中于标注评分噪声,即细微的专业差异逃脱了人工审核者的捕捉,进而损害了模型奖励信号。在方兴未艾的具身智能与物理 AI 产品中,由于对机器人硬件、遥控操作台和摄像机配置等物理设备存在依赖,运营复杂度成倍增加。AfterQuery 开源机器人研究的工程文档列举了常见的故障点,包括菊花链接线故障、伺服电机过载以及电源电压不匹配,这些故障会直接中断数据集录制流程。与此同时,仅靠约 30 名员工的精干核心团队运作,在统筹数以万计的分布式贡献者以及管理用于强化学习评测的独立 Docker 沙盒时,容易引发潜在的协调摩擦。[CR005, CR006, CR007, CR008]
| 失效模式 | 发生概率 | 严重程度 | 缓解措施成熟度 | 残余风险敞口 | 待解短板 |
|---|---|---|---|---|---|
| 专家推理评分噪声与幻觉推理轨迹 | 中 | 高 | 自研评分标准与软件工具 | 专业领域的细微错误逃过人工标注员审查流入微调数据集 | 医学与金融领域未公开评分标准的验证错误率 |
| 机器人硬件标定与遥操作超时错误 | 高 | 中 | 标准化 SOP 与电压检测 | 物理菊花链接线与电压不匹配导致数据采集停摆 | 多臂遥操作系统平台的硬件故障率 |
| 核心团队运营杠杆与协同瓶颈 | 中 | 中 | 定制工具自动化 | 30 人核心团队在管理 100k 分布式贡献者时面临极大负荷 | 内部全职员工 (FTE) 在工程与运营之间的编制分配情况 |
| RL 沙箱容器逃逸与模拟 API 执行失败 | 低 | 高 | 隔离的 Docker 沙箱 | 智能体发起对抗性工具调用时,模拟执行环境崩溃或泄露 | 针对客户动态 RL 环境部署自动化安全监控 |
运营失效模式按严重程度排序,覆盖人工标注质量、硬件在环机器人数据采集与核心团队杠杆。
[CR005, CR006, CR007, CR008]7.3 客户集中度、利润率承压与平台依赖
商业变现表现亮眼,但客户高度集中于前沿 AI 实验室构成重大战略风险点。由于头部前沿实验室既是 AfterQuery 的主要收入来源,又是具备自研数据流程实力的成熟实体,一两家标杆客户一旦决定收回后训练业务自研,或将合同打包给老牌数据供应商,都将严重挫伤其商业增长动能。此外,AfterQuery 坚持采用经认证的行业专家与定制自研软件,相比依赖劳动力套利的众包交易平台竞争对手,承担着明显更高的成本结构。若要拉升毛利率,必须将定制的程序化合作项目转化为可复用的标准化现成基准套件,并在更广阔的企业工作流中完成规模化部署。[CR009, CR010, CR011, CR012]
| 依赖类别 | 交易对手 / 类型 | 工作流角色 | 集中度水平 | 失效情景 | 严重程度 | 缓释措施 | 残留敞口 |
|---|---|---|---|---|---|---|---|
| 客户需求 | 前沿模型实验室(如 OpenAI、Anthropic、DeepMind) | 定制后训练数据与 RL 环境的核心买家 | 高 | 实验室转向自研数据管线,或将支出向老牌平台集中 | 高 | 向企业工作流多元化拓展,并与 Raine Group 建立垂直行业合作 | 流失任何一家前沿实验室客户,都会直接重创年化收入运行率 |
| 人才供给 | 100k 名经认证的行业专业人士 | 在法律、金融与医疗领域生成专家推理轨迹 | 中 | 合同工薪资上涨,或供给流失至竞品平台 | 中 | 自研专有工具链与灵活的按项目计酬机制 | 专家人工数据的高边际成本限制了毛利率扩张 |
| 算力与基础设施 | 云 GPU 供应商与 Hugging Face | 承载训练任务、基准测试与多模态评估作业 | 中 | 针对 20+ GB VLA 模型的显存配额受限,或峰值内存成本过高 | 中 | 采用多云托管,并借助梯度累加优化批大小 | 算力价格飙升会压缩项目制业务的毛利率 |
| 资本悬顶 | 成长期投资人 | 继 $300M A 轮后,据报道完成 $3.2B 估值融资 | 高 | 估值倍数压缩或增速放缓引发下调轮次摩擦 | 高 | 聚焦具备造血能力的企业合同,以及 $100M+ 收入运行率 | 一旦增速放缓,极高估值将造成严重的倍数悬顶风险 |
剖析买方集中度、专家劳动力供给、算力基础设施及估值预期等外部依赖项。
[CR009, CR010, CR011, CR012]7.4 剩余风险热力图、应对成熟度与逻辑破损触发条件
综合战略、运营与财务各项风险敞口来看,前沿实验室客户集中度高企与高估值压力构成了最大的剩余投资风险。A 轮融资后仅 5 个月估值据报道便膨胀至 $3.2B,极大压缩了运营容错空间,倒逼公司必须迅速突破核心基础模型实验室圈子向企业级客户扩张。对投资者而言,核心逻辑破损的触发条件包括:任何主要前沿实验室将后训练流程收归自研并导致实质性客户流失;专家承包商薪酬上涨导致毛利率持续受到侵蚀;或者发生重大数据安全泄露事故,危及 SOC 2 合规地位与企业数据共享协议。[CR013, CR014, CR015]
从发生概率、影响严重程度、缓释成熟度以及残留投资敞口多维度,对比评估关键战略、运营与财务风险类别。
定性排序综合自分析师估算、职位招聘说明与据报道的运营披露。
[CR013, CR014, CR015]7.5 重点图表
08估值
8.1 投资逻辑与融资背景
AfterQuery 抓住基础模型后训练(post-training)瓶颈,成为人工智能领域扩张最快的基础设施供应商之一。基础模型开发商越发需要特定垂直领域的推理数据和复杂工作流轨迹,而这类高阶数据无法靠公开网络抓取获得。对此,AfterQuery 开发了定制化强化学习环境,并组织起覆盖法律、金融、软件和医疗等领域且经过认证的在职专业人士网络,负责精选高信号训练数据。凭这一价值主张,公司于 2026 年 4 月宣布完成由 Altos Ventures 领投的 $30M Series A 轮融资,估值达 $300M;随后披露在成立 14 个月内,年化收入运行率已突破 $100M。2026 年 9 月,老股交易与媒体报道显示,AfterQuery 敲定由 Andreessen Horowitz 和 Sequoia Capital 领投的 Series B 轮融资,估值达 $3.2B。不过,尽管营收端增速亮眼,核心反向逻辑依然在于客户高度集中于少数前沿模型开发商;这些客户既有能力、也有动力自建后训练数据管线。[CV001, CV002, CV003, CV004, CV005, CV006, CV007]
| 核心维度 | 正向投资逻辑 | 反向风险论点 | 观点转向触发条件 |
|---|---|---|---|
| 市场地位 | 前沿 AI 实验室专家人工数据与 RL 环境的核心供应商。 | 前沿实验室可能自建数据清洗团队,或转向合成数据。 | 前 3 大实验室宣布全面自研后训练数据。 |
| 单体经济模型与利润率 | 软件化管理的工作流支撑起近 100,000 名专家的高效编排。 | 高额专家薪酬压缩毛利率,使其低于标准 SaaS 水平。 | 借助可复用数据包,毛利率证明能够扩张至 65% 以上。 |
| 商业势头 | 从成立到据报道达到 $100M ARR 运行率,仅用时 14 个月。 | 客户高度集中;一旦单家实验室流失,将引发单点故障式的收入风险。 | 向企业垂直行业成功扩张,占 ARR 比例超过 40%。 |
| 估值与退出 | Altos、a16z 与 Sequoia 等顶级机构支持,为公司争取溢价退出铺平道路。 | 32.0x 收入倍数没有给执行失误或倍数收缩留出任何容错空间。 | 市场向可比公司倍数(10x-15x)重估,或出现保持 100%+ 持续增长的证据。 |
正向论据与反向风险综合自公司陈述、Sacra 分析以及据报道的融资参数。
[CV001, CV002, CV005, CV006, CV007, CV014]8.2 可比估值与倍数动态
按 $100M 年化运行率收入对应 $3.2B 账面估值计算,AfterQuery 隐含市销率达到 32.0x。相较五个月前 Series A 轮确立的 $300M 估值,该估值扩张超过 10x。对照更广泛的软件与数据基础设施板块,AfterQuery 享有极高的稀缺性溢价。成熟企业级软件可比公司如 Ncino 的市销率约 4.0x,而 Motive 等成熟基础设施平台为 6.1x。即便在快速增长的生成式 AI 模型与基础设施同业中(如 Black Forest Labs 市销率为 33.9x),也能看出市场只对公认拥有结构性护城河的前沿 AI 公司给出 30x 以上倍数。市场分析师在行业评述中强调,鉴于官方未公开披露实际募集的新股金额,$3.2B 账面数字反映的是交易定价,而非真正打入资产负债表的现金。有分析博客援引 30x 至 50x 的主流估值区间,凸显出一级市场对该轮融资的追捧,但也导致其几乎没有缓冲空间来抵御宏观倍数收缩。[CV004, CV008, CV009, CV010]
8.3 情景分析与下行保护
为评估当前定价下投资回报的非对称性,结合商业化推进与估值倍数坚挺度,测算 2027 年的三种远期情景。在乐观情景下,AfterQuery 成功将客户群扩展到 Fortune 500 强企业的智能体部署,并做大标准化基准测试包的复用规模,2027 年年化收入运行率增至 $300M。若维持坚挺的 25.0x 远期倍数,公司估值将达 $7.50B,对应 2.3x 投资回报。在基准情景下,前沿实验室需求持续,收入增至 $220M,但估值倍数向高增长软件基础设施基准回落至 18.0x,对应企业价值 $3.96B,稀释前仅带来 24% 的温和回报。在悲观情景下,前沿实验室自建专家评估管线,企业端落地迟滞,收入增长停留在 $150M。估值倍数收缩至 10.0x,对应估值仅为 $1.50B,后期资金面临 53% 的严重回撤。鉴于超额收益高度集中在乐观情景、且下行风险敞口巨大,当前的风险收益比要求投资者坚守严格的入场条款。[CV003, CV004, CV009, CV011, CV012]
| 情景 | 2027E ARR($M) | 退出 / 远期倍数 | 隐含估值($B) | 下行触发条件与核心假设 | 概率信号 |
|---|---|---|---|---|---|
| 乐观情景 | $300M | 25.0x | $7.50B | 企业级智能体扩张成功;可复用的标准化数据推动毛利率扩张。 | 25% |
| 基准情景 | $220M | 18.0x | $3.96B | 实验室支出持续增长,但倍数从 32.0x 回落至高增长基础设施常态。 | 50% |
| 悲观情景 | $150M | 10.0x | $1.50B | 实验室集中度导致收入见顶;合成数据削弱专家人工需求;存在下调轮次风险。 | 25% |
远期估值预测假设达成 2027 年收入运行率目标,且倍数自当前 32.0x 进入水平收缩。测算未计入未来股权稀释。
[CV003, CV004, CV009, CV011, CV012]8.4 投资建议与尽调触发条件
综合各项实证依据,对 AfterQuery 给出继续研究(research-more)建议,估值态度为偏高(stretched),风险评级为高(high)。尽管多家独立行业信源证实了公司的产品市场契合度与营收提速,但在 32.0x 市销率高位投入成长期资金,仍须验证公开披露未包含的关键内部运营指标。确认性尽调须把住三道关口:第一,核实 Series B 轮中真正注入公司的新股现金与老股套现金额;第二,查验经审计的边际贡献率,判断专业专家网络成本是否呈线性扩张,还是保留了类似软件的高毛利率扩张空间;第三,评估头部 AI 实验室的客户集中度及续约承诺。可量化的投资逻辑破坏触发条件包括:任一顶级 AI 实验室宣布自建专家网络、单一大客户集中度超过 60%,或老股交易出现估值下调(down-round)。[CV004, CV006, CV008, CV013, CV014]
| 指标 / 参数 | 立场 / 输出 | 依据与决策影响 |
|---|---|---|
| 投资建议 | 继续研究 | 出资前,须对新股到账现金余额、单体经济模型与合同续约完成确认性尽调。 |
| 置信度 | 中 | 营收增长势头与估值定价已获多家渠道证实,但客户合同条款和毛利率仍未披露。 |
| 风险评级 | 高 | 客户高度集中在前沿 AI 实验室,叠加实验室潜在自研替代的风险敞口,导致收入面临剧烈波动。 |
| 估值立场 | 偏高 | 按 32.0x 收入倍数(对应 $100M 收入运行率下的 $3.2B 估值),当前定价计入了全部完美预期,安全边际极窄。 |
截至 2026 年 9 月,投资建议与风险画像综合了公开披露、据报道的 B 轮定价及分析师基准。
[CV004, CV006, CV008, CV013]决策流向图展示从营收规模高速扩张,走向估值偏高判断与继续研究建议的推导路径。
| 节点 / 连接 | 节点 / 起点 | 终点 | 说明 |
|---|---|---|---|
| 节点 1 | 前沿实验室需求与专家数据定位 [n1] | ||
| 节点 2 | 成立 14 个月内突破 $100M ARR 运行率 [n2] | ||
| 节点 3 | 据报道 $3.2B 估值(32.0x 倍数) [n3] | ||
| 节点 4 | 实验室集中度过高与高专家成本结构 [n4] | ||
| 节点 5 | 估值悬顶与下调轮次脆弱性 [n5] | ||
| 节点 6 | 投资建议:继续研究 / 估值判断:偏高 [n6] | ||
| 连接 1 | 前沿实验室需求与专家数据定位 [n1] | 成立 14 个月内突破 $100M ARR 运行率 [n2] | |
| 连接 2 | 成立 14 个月内突破 $100M ARR 运行率 [n2] | 据报道 $3.2B 估值(32.0x 倍数) [n3] | |
| 连接 3 | 据报道 $3.2B 估值(32.0x 倍数) [n3] | 估值悬顶与下调轮次脆弱性 [n5] | |
| 连接 4 | 实验室集中度过高与高专家成本结构 [n4] | 估值悬顶与下调轮次脆弱性 [n5] | |
| 连接 5 | 估值悬顶与下调轮次脆弱性 [n5] | 投资建议:继续研究 / 估值判断:偏高 [n6] |
推导逻辑模拟了分析维度的把关要素与定性风险权衡。
[CV002, CV003, CV004, CV006, CV007, CV008, CV013]8.5 附图与附表
免责声明
本报告仅供研究与尽调参考,不构成投资建议。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | AfterQuery was founded in 2024 in San Francisco by Carlos Georgescu, Spencer Mateega, and Danny Tang as an applied AI training data research lab. | 高 | SO001, SO004, SO005 |
| CO002 | AfterQuery participated in the Y Combinator Winter 2025 accelerator cohort and secured $500,000 in pre-seed funding. | 高 | SO004, SO005, SO008 |
| CO003 | Latka estimates AfterQuery generated early commercial traction with frontier labs reaching a $6.5M revenue run rate in September 2025. | 中 | SO003 |
| CO004 | On April 9, 2026, AfterQuery announced a $30 million Series A funding round at a $300 million post-money valuation led by Altos Ventures. | 高 | SO002, SO006 |
| CO005 | AfterQuery reached an annualized revenue run rate exceeding $100 million in April 2026 approximately 14 months after its founding. | 高 | SO002, SO006 |
| CO006 | AfterQuery curates datasets using a network of nearly 100,000 verified practicing professionals spanning software, law, finance, and medicine. | 高 | SO002, SO006 |
| CO007 | AfterQuery's technical offering encompasses supervised fine-tuning data, reinforcement learning reward rubrics, and simulated agent environments including MCP server integrations. | 高 | SO001, SO002, SO008 |
| CO008 | In September 2026, TechCrunch and Forbes reported that AfterQuery raised a new funding round valuing the company at $3.2 billion as Y Combinator's fastest-ever reported unicorn. | 中 | SO007 |
| CO009 | AfterQuery's core founding executive team comprises Chief Executive Officer Spencer Mateega, Carlos Georgescu, and Danny Tang. | 高 | SO004, SO005 |
| CO010 | The Series A investor syndicate includes institutional investors Altos Ventures, The Raine Group, Y Combinator, BoxGroup, and Latitude Capital alongside angel investors from frontier AI labs. | 高 | SO004, SO006 |
| CO011 | AfterQuery reported a full-time core headcount of 30 employees in mid-2026, reflecting high capital efficiency and reliance on contractor expert networks. | 高 | SO003, SO005 |
| CO012 | Total disclosed cumulative venture funding for AfterQuery across its pre-seed and Series A rounds stands at approximately $30.5 million. | 中 | SO004, SO008 |
| CO013 | AfterQuery faces severe customer concentration risk as its initial commercial revenue is heavily concentrated among a small number of frontier foundation AI labs. | 中 | SO008 |
| CO014 | Higher cost structures from custom expert curation and in-house tooling constrain gross margin expansion relative to marketplace platforms like Turing or Mercor unless off-the-shelf datasets scale. | 中 | SO008 |
| CM001 | AfterQuery reached an annualized revenue run rate of $100 million in April 2026. | 高 | SM004, SM005, SM006 |
| CM002 | AfterQuery provides datasets and reinforcement learning environments to major AI labs including Nvidia. | 高 | SM004, SM005 |
| CM003 | Grand View Research estimated the global AI training dataset market at approximately $3.9 billion in 2026. | 中 | SM001 |
| CM004 | Fortune Business Insights and Mordor Intelligence estimated the AI training data and related market size between $4.4 billion and $11.9 billion in 2026. | 中 | SM001 |
| CM005 | Aggregated private AI data provider revenues including Scale AI, Mercor, and Surge AI demonstrate multi-billion-dollar commercial spend exceeding conservative market research figures. | 中 | SM001 |
| CM006 | AfterQuery offers a three-layer post-training product stack comprising supervised fine-tuning data, reinforcement learning environments, and evaluation infrastructure. | 中 | SM007 |
| CM007 | AfterQuery maintains a network of nearly 100,000 verified professionals across medical, legal, financial, and engineering domains. | 中 | SM002, SM004 |
| CM008 | Enterprise AI adoption shows roughly 62% of organizations experimenting with AI agents while only 23% report scaling them in production, per McKinsey data cited by Sacra. | 中 | SM007 |
| CM009 | Target buyers for expert training data are primarily foundation model post-training teams, evaluation leads, and specialized enterprise AI groups. | 中 | SM007 |
| CM010 | AfterQuery faces customer concentration risk due to reliance on a small cluster of frontier model labs with lumpy, large-contract procurement patterns. | 中 | SM001, SM007 |
| CM011 | Chinese AI laboratories represent an estimated $50 million of recurring revenue for AfterQuery, introducing potential geopolitical and regulatory trade exposure. | 中 | SM001 |
| CM012 | Frontier AI lab budgets in 2026 increasingly shifted toward reinforcement learning environments and domain-specific reasoning benchmarks. | 中 | SM001, SM007 |
| CM013 | AfterQuery recruits cross-functional personnel to design dataset specifications that influence foundation model evaluations and performance. | 中 | SM003, SM008 |
| CM014 | AfterQuery's reported gross revenue figures lack public disclosure regarding contractor pass-through rates and net gross margins. | 中 | SM001 |
| CP001 | Direct peer Mercor and Surge AI compete directly with AfterQuery in human expert datasets and RL environments, with Mercor reporting penetration into top AI labs. | 中 | SP001 |
| CP002 | Scale AI represents the incumbent data platform benchmark, combining subject-matter experts, RLHF pipelines, and dedicated RL Environments launched in early 2026. | 高 | SP001, SP002 |
| CP003 | Labelbox operates an expert network via its Alignerr platform with knowledge workers and PhDs, providing RL data and private benchmarks to frontier labs. | 中 | SP001, SP008 |
| CP004 | AfterQuery differentiates from traditional labelers by focusing on complex professional workflows and RL environments rather than simple static label procurement. | 中 | SP001, SP005 |
| CP005 | TechCrunch reported that AfterQuery's customer list includes frontier AI participants such as Nvidia, Legora, and Korean AI lab Motif Technologies. | 高 | SP002, SP004 |
| CP006 | AfterQuery relies on in-house tooling and direct management of nearly 100,000 verified practicing professionals across medicine, law, finance, and software engineering. | 中 | SP003 |
| CP007 | Turing and marketplace competitors leverage large talent pools and synthetic pipelines to compete on throughput and price for coding and domain reasoning data. | 中 | SP001 |
| CP008 | Evaluation and harness platforms including Braintrust, LangSmith, and Patronus enable enterprises to manage test sets and human review internally. | 中 | SP001 |
| CP009 | Commercial pricing for expert AI training data remains programmatic and project-based, scoped by domain complexity, expert seniority, and sandbox environment requirements. | 中 | SP001 |
| CP010 | Tracxn tracks 240 active competitors in the data and dataset category, with 78 funded entities competing across labeling, curation, and intelligence. | 中 | SP008 |
| CP011 | Disintermediation and insourcing by major foundation model labs represent a material commercial threat if frontier labs build internal expert annotation operations. | 中 | SP001, SP004 |
| CP012 | High custom engineering costs for bespoke reinforcement learning sandboxes create higher operational expense compared to pure labor-arbitrage data marketplaces. | 中 | SP001 |
| CP013 | PromptLoop notes that traditional data providers and benchmarking organizations compete with AfterQuery for specialized datasets used in advanced model evaluation. | 中 | SP005 |
| CP014 | Y Combinator company disclosures emphasize that foundation models fail on up to 60% of realistic daily finance tasks without domain-specific professional training data. | 中 | SP007 |
| CI001 | AfterQuery announced a $30 million Series A financing round at a $300 million valuation on April 9, 2026, led by Altos Ventures with participation from The Raine Group, Y Combinator, and BoxGroup. | 高 | SI002, SI003, SI008 |
| CI002 | AfterQuery reported surpassing a $100 million annual revenue run rate in April 2026, approximately fourteen months after its founding. | 高 | SI002, SI003, SI005 |
| CI003 | In September 2026, media sources reported that AfterQuery reached a $3.2 billion valuation in a new funding round, which would represent more than a tenfold valuation expansion within five months. | 中 | SI001, SI005 |
| CI004 | Sacra estimates that AfterQuery reached $100 million in annualized revenue in April 2026, derived across custom datasets, bespoke reinforcement learning environments, off-the-shelf packages, and enterprise consulting. | 中 | SI004 |
| CI005 | AfterQuery's monetization model is project-based and programmatic rather than pure seat-license SaaS, with pricing scoped to domain difficulty, expert mix, sandbox complexity, and dataset volume. | 中 | SI004 |
| CI006 | Commercial delivery includes reinforcement learning sandboxes built on APIs, developer environments, and terminal containers where model test results serve as reward signals. | 中 | SI004 |
| CI007 | The company maintains an active network of nearly 100,000 verified practicing domain professionals across software engineering, medicine, law, and finance to author and validate training data. | 中 | SI002, SI003 |
| CI008 | AfterQuery relies on internal researchers to validate training data quality and benchmark performance improvements before delivering datasets to frontier AI labs. | 中 | SI001, SI004 |
| CI009 | Co-founder Spencer Mateega claimed in July 2026 that the company's recurring revenue had expanded into the hundreds of millions, while external reports indicate the company is operating profitably. | 中 | SI001 |
| CI010 | Full-time employee headcount was reported at approximately 30 employees in mid-2026, reflecting high revenue-per-employee leverage driven by external contractor utilization. | 中 | SI006 |
| CI011 | Sacra reports that AfterQuery's reliance on verified subject-matter experts and bespoke tooling creates a higher cost structure than marketplace competitors, constraining gross margin expansion. | 中 | SI004 |
| CI012 | Revenue is heavily concentrated among a small cohort of frontier model labs that possess both the budget to fund large-scale data programs and the potential to internalize data curation. | 中 | SI004, SI005 |
| CI013 | Total disclosed primary capital raised through Series A equals approximately $30.5 million across a $500,000 pre-seed round and the $30 million Series A. | 中 | SI004, SI008 |
| CI014 | Public benchmark evaluations such as FinanceQA serve as demand-generation assets by demonstrating that off-the-shelf models fail complex domain workflows prior to specialized fine-tuning. | 中 | SI004, SI007 |
| CE001 | AfterQuery develops a specialized data layer for foundation model training focused on procedural expert behavior rather than static question answering. | 高 | SE002, SE004 |
| CE002 | AfterQuery's core offering comprises supervised fine-tuning data, interactive reinforcement learning environments, and domain-specific evaluation benchmarks. | 高 | SE002, SE008 |
| CE003 | Supervised fine-tuning packages provide prompt-response pairs and reasoning traces to establish baseline professional behavior across finance, law, coding, and medicine. | 中 | SE002 |
| CE004 | AfterQuery constructs reinforcement learning environments inside isolated sandboxes incorporating real APIs, developer tools, and Model Context Protocol servers. | 中 | SE002 |
| CE005 | AfterQuery uses containerized environments where fresh Docker instances execute terminal commands and test suites to generate deterministic reward signals. | 中 | SE002 |
| CE006 | The company builds custom benchmarks including FinanceQA, IDE-Bench, and Market-Bench to diagnose model capability gaps and validate fine-tuning gains. | 高 | SE002, SE008 |
| CE007 | AfterQuery claims an expert contributor network of nearly 100,000 verified practicing professionals across engineering, medicine, law, and finance. | 中 | SE004 |
| CE008 | AfterQuery operates an expert contributor portal at experts.afterquery.com for recruiting and onboarding domain specialists. | 中 | SE001 |
| CE009 | The core engineering stack relies on Python, Go, TypeScript, containerized cloud infrastructure, and custom internal human-in-the-loop tooling. | 中 | SE007 |
| CE010 | AfterQuery maintains SOC 2 Type II and ISO 27001 compliance to meet security procurement requirements for enterprise codebases and proprietary workflows. | 中 | SE002 |
| CE011 | Unlike labor marketplaces that rely on offshore crowd workers, AfterQuery develops custom in-house software and directly manages expert curation teams. | 高 | SE002, SE004 |
| CE012 | Engineering recruitment mandates building programmatic security monitoring, access control, and incident response systems for multi-tenant AI training data flows. | 中 | SE003, SE007 |
| CE013 | Leading frontier model labs represent AfterQuery's primary customer base while also developing internal data curation capabilities that pose potential competitive substitution risks. | 高 | SE002, SE005 |
| CE014 | Latka reports that AfterQuery operated with approximately 30 employees in 2026, highlighting a lean operational core relative to its reported commercial scale. | 中 | SE006 |
| CU001 | AfterQuery announced that it surpassed a $100 million annual revenue run rate within 14 months of inception following its April 2026 Series A round. | 高 | SU003, SU001 |
| CU002 | AfterQuery reportedly achieved a $3.2 billion post-money valuation in September 2026, becoming the fastest startup in Y Combinator history to reach unicorn status from launch. | 高 | SU001, SU004, SU005 |
| CU003 | AfterQuery's primary customer base is concentrated among frontier AI foundation model laboratories and major enterprise technology companies. | 高 | SU002, SU003 |
| CU004 | AfterQuery claims that every leading AI lab uses its datasets and reinforcement learning environments to encode and scale domain expertise across benchmarks. | 中 | SU003 |
| CU005 | Independent tech reporting names Nvidia, Legora, and Korean foundation model lab Motif Technologies as confirmed commercial customers of AfterQuery. | 中 | SU001 |
| CU006 | AfterQuery's commercial agreements are predominantly structured as high-touch consultative project contracts and bespoke RL environment builds rather than seat-based SaaS subscriptions. | 中 | SU002 |
| CU007 | Contract pricing at AfterQuery is dynamically scoped based on professional domain difficulty, expert mix, environment complexity, and delivered dataset volume. | 中 | SU002 |
| CU008 | AfterQuery utilizes an in-house expert network of nearly 100,000 verified practicing professionals across engineering, medicine, law, and finance to generate training data. | 中 | SU003, SU008 |
| CU009 | Industry analysts flag acute revenue concentration risk, noting that AfterQuery relies heavily on a small cohort of frontier labs that represent both its largest clients and potential insourcing threats. | 中 | SU002 |
| CU010 | Critical tech industry analysis questions the operational durability of AfterQuery's valuation, noting that long-term profitability and scaling efficiency beyond initial high-profile accounts remain unproven. | 中 | SU004 |
| CU011 | Publicly accessible market research and commercial databases record zero audited figures for AfterQuery's active customer account count, net revenue retention, or logo churn rates as of late 2026. | 中 | SU007 |
| CU012 | AfterQuery leverages public research benchmarks such as FinanceQA and Terminal-Bench to expose model reasoning deficiencies and drive customer inbound demand. | 中 | SU002, SU008 |
| CU013 | AfterQuery faces direct competitive pressure from incumbent data engines like Scale AI and talent marketplace challengers such as Mercor, Surge AI, and Labelbox. | 中 | SU002 |
| CU014 | Commercial enterprise expansion is underway through strategic advisory partnerships, such as initiatives with The Raine Group to diagnose and automate internal financial workflows. | 中 | SU002, SU003 |
| CR001 | AfterQuery recruits practicing attorneys across corporate, litigation, IP, and regulatory specialties to establish rigorous grading standards and citation integrity rubrics for legal AI datasets. | 中 | SR003 |
| CR002 | AfterQuery utilizes ISO 27001 and SOC 2 Type II compliance certifications as essential procurement deal enablers for enterprise clients sharing proprietary workflows and codebases. | 中 | SR004 |
| CR003 | AfterQuery maintains a dedicated expert contractor infrastructure and privacy policy portal to manage external contributor data and intellectual property obligations. | 高 | SR002, SR005 |
| CR004 | AfterQuery positions its offering as original expert reasoning traces and custom simulated environments rather than web-scraped data to avoid intellectual property disputes. | 高 | SR004, SR008 |
| CR005 | Domain-specific AI datasets in specialized fields such as law and finance face the operational risk that subtle reasoning errors and incorrect citations may be missed without strict multi-step verification rubrics. | 中 | SR003, SR004 |
| CR006 | Physical robotics teleoperation and dataset collection are vulnerable to hardware failures including daisy-chain wiring disconnects, motor overload, and power supply voltage mismatches. | 中 | SR001 |
| CR007 | AfterQuery operates with an internal core team of approximately 30 employees as of 2026, creating operational leverage risks while overseeing large-scale enterprise engagements. | 高 | SR007, SR008 |
| CR008 | Training large vision-language-action robotics policies requires significant GPU memory, ranging from 15 GB to 24 GB or more, which introduces compute constraints for complex multimodal data collection. | 中 | SR001 |
| CR009 | AfterQuery experiences customer concentration risk due to its revenue being heavily weighted toward a small cohort of frontier AI labs that are also capable of developing internal data operations. | 中 | SR004 |
| CR010 | AfterQuery's reliance on verified professional domain experts and custom in-house software creates a higher unit cost structure than marketplace competitors that utilize crowdsourced labor arbitrage. | 中 | SR004 |
| CR011 | AfterQuery reports working with nearly 100,000 verified practicing professionals across engineering, medicine, law, and finance to generate training data that cannot be produced synthetically. | 中 | SR005 |
| CR012 | AfterQuery was reportedly valued at $3.2 billion in September 2026, representing a more than tenfold valuation increase from its $300 million Series A announced five months earlier. | 高 | SR005, SR006 |
| CR013 | A strategic decision by major frontier AI labs to internalize post-training data curation or shift spend to bundled platform incumbents poses a primary threat to AfterQuery's customer durability. | 中 | SR004 |
| CR014 | Sustaining a $3.2 billion valuation requires AfterQuery to successfully convert bespoke lab engagements into scalable, reusable off-the-shelf datasets and broad enterprise agent deployments. | 中 | SR004, SR006 |
| CR015 | AfterQuery announced surpassing a $100 million revenue run rate in 2026, creating high operational expectations against its relatively lean 30-person organizational foundation. | 高 | SR005, SR007 |
| CV001 | AfterQuery raised a $30 million Series A financing at a $300 million valuation in April 2026 led by Altos Ventures with participation from The Raine Group, Y Combinator, BoxGroup, and Latitude Capital. | 高 | SV003, SV004, SV007 |
| CV002 | AfterQuery reported surpassing $100 million in annual revenue run rate within 14 months of inception following its Series A round. | 高 | SV004, SV005, SV007 |
| CV003 | In September 2026, AfterQuery reportedly raised a Series B round valuing the startup at $3.2 billion, representing a greater than 10x valuation step-up over five months. | 高 | SV001, SV002, SV005 |
| CV004 | At a $3.2 billion valuation on a reported $100 million revenue base, AfterQuery trades at an implied 32.0x revenue multiple. | 高 | SV001, SV007 |
| CV005 | The investment thesis centers on proprietary reinforcement learning environments and a network of nearly 100,000 verified professionals delivering high-quality training data to leading AI labs. | 高 | SV003, SV004, SV007 |
| CV006 | The anti-thesis highlights severe revenue concentration among a small number of frontier AI labs that are both AfterQuery's primary customers and potential internal competitors. | 中 | SV002, SV007 |
| CV007 | AfterQuery's reliance on paid domain professionals and custom tooling creates a higher cost structure that constrains gross margins relative to marketplace labor aggregators. | 中 | SV007 |
| CV008 | Industry analysts warn that the reported $3.2 billion valuation reflects price assigned in the round rather than balance sheet cash, creating heightened operational pressure to sustain growth and governance standards. | 中 | SV002 |
| CV009 | Compared to private peers like Motive at 6.1x revenue and public comps like Affle 3i at 6.9x and Ncino at 4.0x, AfterQuery's 32.0x multiple commands an extreme growth premium comparable to generative AI peer Black Forest Labs at 33.9x. | 中 | SV001 |
| CV010 | Market blog commentary reported speculative figures of $120 million ARR and 30x to 50x prevailing revenue multiples, underscoring intense private-market hype and multiple inflation surrounding the transaction. | 低 | SV008 |
| CV011 | A base-case scenario projecting $220 million ARR in 2027 at a normalized 18x multiple yields a $3.96 billion valuation, providing a modest 24% return over the entry price. | 中 | SV001, SV007 |
| CV012 | A bear-case multiple contraction to 10x on $150 million ARR would imply a $1.50 billion valuation, exposing late-stage investors to a greater than 50% downside loss. | 中 | SV001, SV002 |
| CV013 | Based on stretched valuation multiples, high lab concentration, and unverified primary balance sheet cash, the supported investment stance is stretched with a research-more recommendation. | 中 | SV001, SV002, SV007 |
| CV014 | Key thesis-break triggers include frontier labs insourcing post-training pipelines, enterprise pilot conversion falling below expectations, or a down-round financing event. | 中 | SV002, SV007 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | AfterQuery | Expert Data for Frontier AI - AfterQuery | AfterQuery is an applied research lab curating data solutions for frontier foundation model development. Models trained on outputs plateau. Models trained on reasoning improve. We build datasets that reflect how experts actually solve problems — step by step, decision by decision. |
| SO002 | SiliconANGLE | AI training data startup AfterQuery nabs $30M investment | San Francisco-based AfterQuery provides datasets that AI developers use to train their models. The 14-month-old company says its customer base includes “every leading AI lab.” Its annual recurring revenue topped $100 million shortly before the funding round. |
| SO003 | Latka | AfterQuery Revenue 2026: $100M ARR, $300M Valuation | AfterQuery employs approximately 30 people as of 2026, down from 59 in 2025. |
| SO004 | Tracxn | AfterQuery | Platform that specializes datasets for artificial intelligences. It helps AI companies access human-curated data that cannot be found online or synthetically generated. |
| SO005 | Y Combinator | AfterQuery: Applied research lab curating data solutions for foundation model… | Y Combinator | Carlos and Spencer first met in high school at a summer program at Google and then interned together at Meta. Danny and Spencer met in high school, where they built a startup. They then both got into Wharton, sold that startup, and became roommates. |
| SO006 | AfterQuery | Human expertise, reimagined - AfterQuery | Today, AfterQuery is excited to announce our $30 million Series A at a $300 million valuation, led by Altos Ventures, with participation from The Raine Group and our existing investors, Y Combinator, BoxGroup, and Latitude Capital. |
| SO007 | TechCrunch | AfterQuery reportedly becomes Y Combinator's fastest-ever unicorn, now valued at $3.2B | TechCrunch | AI training-data startup AfterQuery has reportedly raised a round that valued it at $3.2 billion. This just five months after announcing its $30 million Series A at a $300 million valuation in April. |
| SO008 | Sacra | AfterQuery revenue, valuation & funding | Lab concentration: AfterQuery's revenue is concentrated in a small number of frontier AI labs that are both its largest customers and potential competitors, so a decision by even one or two labs to internalize post-training data operations or shift to a bundled platform vendor could create a material revenue gap that the enterprise channel would take time to fill. |
| SM001 | Signals Inbox | Is AfterQuery really worth $3.2B today? | Grand View Research puts the global AI training-dataset market around $3.9 billion in 2026. Fortune Business Insights comes out around $4.4 billion. Mordor Intelligence, using a broader definition, is closer to $11.9 billion. |
| SM002 | Tech Funding News | AfterQuery raises $30M at $300M valuation from Altos Ventures as AI labs race for expert training data — TFN | AfterQuery offers two main products: high-quality datasets that capture expert reasoning in specific fields and training environments where AI models can practice decision-making in realistic professional situations. The company says that every major lab is now a customer. |
| SM003 | Simplify | Strategic Projects Lead @ AfterQuery | Simplify | Work with leading artificial intelligence organizations to build next-generation data solutions that push the limits of model capability while directly driving revenue. |
| SM004 | AfterQuery | Human expertise, reimagined - AfterQuery | Today, every leading AI lab uses AfterQuery’s datasets and reinforcement learning environments to encode, evaluate, and scale expertise across dozens of domains. |
| SM005 | TechCrunch | AfterQuery reportedly becomes Y Combinator's fastest-ever unicorn, now valued at $3.2B | TechCrunch | In April, the San Francisco startup said it had reached an annualized revenue run rate of $100 million and that it was working with many of the biggest labs. It has named companies such as Nvidia, Legora, and the Korean AI lab Motif Technologies as customers. |
| SM006 | Latka | AfterQuery Revenue 2026: $100M ARR, $300M Valuation | In 2026, AfterQuery's revenue reached $100M. The company previously reported $100M in 2026. |
| SM007 | Sacra | AfterQuery revenue, valuation & funding | The offering has three layers. The first is supervised fine-tuning data: prompt-response pairs and reasoning traces that set baseline professional behavior before reinforcement learning begins. The second is RL environments: custom sandboxes built on top of real APIs, MCP servers, developer tools, browsers, and desktop interfaces where models learn to chain actions, call tools, and recover from failures inside realistic workflows. |
| SM008 | Lensa | Marketing Lead job in San Francisco at AfterQuery | Lensa | Design, launch, and optimize multi-channel marketing campaigns to drive awareness, product adoption, and customer engagement across frontier AI labs |
| SP001 | Sacra | AfterQuery revenue, valuation & funding | Scale AI is the clearest competitive benchmark. Its GenAI Data Engine combines vetted subject-matter experts, RLHF pipelines, and model evaluation tooling, and Scale launched dedicated RL Environments in early 2026, with nearly half of new data-training projects now involving reinforcement-learning environments. |
| SP002 | TechCrunch | AfterQuery reportedly becomes Y Combinator's fastest-ever unicorn, now valued at $3.2B | AfterQuery is among the new crop of startups following in the footsteps of Mercor and Scale that employ knowledge professionals like doctors, lawyers, and other specialists to do model training. |
| SP003 | AfterQuery | Human expertise, reimagined | Rather than outsourcing data collection, we custom-build the tools and workflows to manage creation meticulously in-house for every project. The result is higher-quality data, superior operational oversight for our team, and a significantly better experience for the verified expert contributors who power our platform. |
| SP004 | Simplify | Business Operations Generalist @ AfterQuery | Major labs can internalize expert-data pipelines, crushing margins within 2027. |
| SP005 | PromptLoop | What Does AfterQuery Do? - Company Overview | Competitors may include traditional data providers, AI research labs, and benchmarking organizations, particularly those supplying specialized or hard-to-find datasets. However, AfterQuery differentiates itself by focusing on datasets that are both human-generated and tailored for advanced AI evaluation and development use cases. |
| SP006 | Latka | AfterQuery Revenue 2026: $100M ARR, $300M Valuation | AfterQuery has raised $30M in total funding across 1 round, most recently a $30M Series A round in 2026. |
| SP007 | Y Combinator | AfterQuery: Applied research lab curating data solutions for foundation model… | Current AI models are trained to be smart but are not trained on the work of actual professionals in many fields. |
| SP008 | Tracxn | AfterQuery - 2026 Company Profile, Team, Funding | The company has 240 active competitors, including 78 funded and 9 that have exited |
| SI001 | Forbes | AfterQuery Becomes YC's Fastest Unicorn At $3.2 Billion | AfterQuery is profitable and has lined up a lead investor, one of the people said. |
| SI002 | Business Wire | AfterQuery Raises $30 Million Series A Round at $300 Million Valuation | AfterQuery has surpassed $100 million in annual revenue run rate, a milestone that reflects both the urgency of the problem we are solving and the trust the world's leading AI labs have placed in us. |
| SI003 | AfterQuery | Human expertise, reimagined - AfterQuery | Since closing our Series A a few months ago, AfterQuery has surpassed $100 million in revenue run rate and is backed by angels from Google DeepMind, OpenAI, Anthropic, Meta Superintelligence Labs, and Microsoft AI. |
| SI004 | Sacra | AfterQuery revenue, valuation & funding - Sacra | AfterQuery's quality differentiation depends on verified domain experts, custom-built tooling, and in-house workflow management rather than marketplace-style labor arbitrage, so its cost structure is higher than competitors like Turing or Mercor that can flex supply more cheaply, which constrains gross margin expansion unless the company converts bespoke engagement work into reusable off-the-shelf inventory at scale. |
| SI005 | TechCrunch | AfterQuery reportedly becomes Y Combinator's fastest-ever unicorn, now valued at $3.2B | TechCrunch | In April, the San Francisco startup said it had reached an annualized revenue run rate of $100 million and that it was working with many of the biggest labs. |
| SI006 | GetLatka | AfterQuery Revenue 2026: $100M ARR, $300M Valuation | AfterQuery employs approximately 30 people as of 2026, down from 59 in 2025. |
| SI007 | Y Combinator | AfterQuery: Applied research lab curating data solutions for foundation model… | Y Combinator | We are building a platform to onboard experts to work on custom datasets. Think working professionals at top firms and the smartest students in various subjects (PhDs). |
| SI008 | Tracxn | AfterQuery - 2026 Company Profile, Team, Funding ... - Tracxn | AfterQuery has raised a total funding of$30.5M over 2 round s.Its first funding round was on 2025.Its latest funding round was a Series A round on Apr 09, 2026 for $30M. |
| SE001 | AfterQuery | Apply to Jobs - AfterQuery Experts | AfterQuery Experts - Transform the Future of Technology |
| SE002 | Sacra | AfterQuery revenue, valuation & funding | The offering has three layers. The first is supervised fine-tuning data: prompt-response pairs and reasoning traces that set baseline professional behavior before reinforcement learning begins. The second is RL environments: custom sandboxes built on top of real APIs, MCP servers, developer tools, browsers, and desktop interfaces where models learn to chain actions, call tools, and recover from failures inside realistic workflows. |
| SE003 | AfterQuery via Ashby | Senior Software Engineer - Security/Infrastructure @ AfterQuery | Senior Software Engineer - Security/Infrastructure @ AfterQuery |
| SE004 | AfterQuery | Human expertise, reimagined - AfterQuery | Today, every leading AI lab uses AfterQuery’s datasets and reinforcement learning environments to encode, evaluate, and scale expertise across dozens of domains. |
| SE005 | TechCrunch | AfterQuery reportedly becomes Y Combinator's fastest-ever unicorn, now valued at $3.2B | TechCrunch | AI training-data startup AfterQuery has reportedly raised a round that valued it at $3.2 billion. This just five months after announcing its $30 million Series A at a $300 million valuation in April. |
| SE006 | Latka | AfterQuery Revenue 2026: $100M ARR, $300M Valuation | AfterQuery generated $100M in revenue in 2026. |
| SE007 | AfterQuery via Vaia | Software Engineer - Security/Infrastructure at AfterQuery | Apply now! – Vaia – Talents | You’ll be responsible for protecting a system that handles highly sensitive workflows across AI training, expert networks, and enterprise integrations with leading model labs. |
| SE008 | Y Combinator / AfterQuery | AfterQuery: Applied research lab curating data solutions for foundation model… | Y Combinator | We are building a platform to onboard experts to work on custom datasets. Think working professionals at top firms and the smartest students in various subjects (PhDs). |
| SU001 | TechCrunch | AfterQuery reportedly becomes Y Combinator's fastest-ever unicorn, now valued at $3.2B | It has named companies such as Nvidia, Legora, and the Korean AI lab Motif Technologies as customers. |
| SU002 | Sacra | AfterQuery revenue, valuation & funding | AfterQuery's revenue is concentrated in a small number of frontier AI labs that are both its largest customers and potential competitors, so a decision by even one or two labs to internalize post-training data operations or shift to a bundled platform vendor could create a material revenue gap that the enterprise channel would take time to fill. |
| SU003 | AfterQuery | Human expertise, reimagined - AfterQuery | Today, every leading AI lab uses AfterQuery’s datasets and reinforcement learning environments to encode, evaluate, and scale expertise across dozens of domains. |
| SU004 | Uncle Space Station | AfterQuery Reaches $3.2B Valuation in Record Time for Y Combinator Startup - Uncle Space Station | The reported $3.2 billion valuation is highly dependent on an *announced* revenue run rate and investor confidence in a specific, high-touch AI training methodology. The actual, sustained profitability and operational efficiency of "encoding patterns, decisions, and reasoning" at scale remain to be demonstrated. |
| SU005 | The Entrepreneur Story | AfterQuery Hit $3.2B in Five Months — YC's Fastest Unicorn Ever | In a nascent but rapidly expanding market, identifying and securing early adopters or lighthouse customers can be transformative. AfterQuery might have focused on a specific niche within the AI model training space, targeting companies with immediate, severe pain points. |
| SU006 | Seedtable | AfterQuery — Funding, Investors & Team | Seedtable | AfterQuery is a San Francisco-based machine learning and artificial intelligence company, founded in 2025. |
| SU007 | Latka | AfterQuery Revenue 2026: $100M ARR, $300M Valuation | We do not have customer count information for AfterQuery yet. |
| SU008 | Y Combinator | AfterQuery: Applied research lab curating data solutions for foundation model… | Y Combinator | We are building a platform to onboard experts to work on custom datasets. Think working professionals at top firms and the smartest students in various subjects (PhDs). |
| SR001 | GitHub | lerobot/AGENT_GUIDE.md at main · AfterQuery-Research/lerobot | Using a 12 V PSU on a 5 V / 7.4 V arm (or vice-versa) will trip this error; confirm your motor variant before powering up. |
| SR002 | AfterQuery | AfterQuery Experts - Transform the Future of Technology | AfterQuery Experts - Transform the Future of Technology |
| SR003 | AfterQuery | Strategic Projects Lead - Legal job in San Francisco at AfterQuery | Lensa | Build the grading standards and evaluation criteria that let a lab distinguish real legal reasoning from confident-sounding error, including citation integrity and jurisdictional accuracy. |
| SR004 | Sacra | AfterQuery revenue, valuation & funding | Lab concentration: AfterQuery's revenue is concentrated in a small number of frontier AI labs that are both its largest customers and potential competitors, so a decision by even one or two labs to internalize post-training data operations or shift to a bundled platform vendor could create a material revenue gap that the enterprise channel would take time to fill. |
| SR005 | AfterQuery | Human expertise, reimagined - AfterQuery | Today, every leading AI lab uses AfterQuery’s datasets and reinforcement learning environments to encode, evaluate, and scale expertise across dozens of domains. |
| SR006 | TechCrunch | AfterQuery reportedly becomes Y Combinator's fastest-ever unicorn, now valued at $3.2B | TechCrunch | AI training-data startup AfterQuery has reportedly raised a round that valued it at $3.2 billion. This just five months after announcing its $30 million Series A at a $300 million valuation in April. |
| SR007 | Latka | AfterQuery Revenue 2026: $100M ARR, $300M Valuation | AfterQuery employs approximately 30 people as of 2026, down from 59 in 2025. |
| SR008 | Y Combinator | AfterQuery: Applied research lab curating data solutions for foundation model… | Y Combinator | AI models need high-quality training data to effectively automate complex professional tasks - we provide that data. |
| SV001 | Multiples | AfterQuery Revenue, Valuation, Funding & Investors | Multiples | AfterQuery is currently valued at $3.2B, based on its Series B in September 2026. |
| SV002 | TMCnet | AfterQuery Secures Funding at Reported $3.2 Billion Valuation | The $3.2 billion valuation describes the price investors assigned to AfterQuery in the round, not the amount of cash placed on the company's balance sheet. Financing terms, investor participation, revenue, and customer figures also were not disclosed in the supplied information. |
| SV003 | The Top Voices | AfterQuery Raises $30M to Scale Expert Data for AI Models | AfterQuery, a data infrastructure company providing high-quality datasets and reinforcement learning environments for AI models, announced a $30 million Series A at a $300 million valuation led by Altos Ventures. The round included participation from The Raine Group, Y Combinator, BoxGroup, and Latitude Capital. |
| SV004 | AfterQuery | Human expertise, reimagined - AfterQuery | Today, AfterQuery is excited to announce our $30 million Series A at a $300 million valuation, led by Altos Ventures, with participation from The Raine Group and our existing investors, Y Combinator, BoxGroup, and Latitude Capital. |
| SV005 | TechCrunch | AfterQuery reportedly becomes Y Combinator's fastest-ever unicorn, now valued at $3.2B | TechCrunch | AI training-data startup AfterQuery has reportedly raised a round that valued it at $3.2 billion. This just five months after announcing its $30 million Series A at a $300 million valuation in April. |
| SV006 | GetLatka | AfterQuery Revenue 2026: $100M ARR, $300M Valuation | AfterQuery reached a $300M valuation in 2026, set during its Valuation round. AfterQuery has raised $30M in total funding across 1 round, most recently a $30M Series A round in 2026. |
| SV007 | Sacra | AfterQuery revenue, valuation & funding | Sacra estimates that AfterQuery hit $100M in annualized revenue in April 2026. The revenue mix spans custom dataset engagements, bespoke RL environment development, off-the-shelf data packages, and enterprise consulting work. |
| SV008 | Mauveverse | Fastest Unicorn Startup 2026: AfterQuery Hits $3.2B in 5 Months | AfterQuery, an AI model training startup, has reportedly become Y Combinator’s fastest-ever unicorn, reaching a $3.2 billion valuation in just five months. This shatters the previous record held by Brex, which took two years to hit a $1 billion valuation. |