初创公司尽调
尽调报告 AI / data platform / cloud infrastructure Private growth-stage company with strategic investment disclosed in 2026 2026-08-12

Thinking Machines Data Science, Inc.

东南亚应用 AI 服务龙头——战略可信度成立,高估值叙事尚未证实

Thinking Machines 是一家战略可信度扎实的东南亚应用 AI 公司,合作伙伴和客户验证都是真实的;但公开证据还撑不起溢价独角兽叙事,也不足以给出高置信估值判断。在收入质量、融资条款和客户集中度看清之前,应继续研究。

封面要素

成立时间 01
2015 [CO003]
最新资本事件 02
Temus strategic investment (Jul 2026) [CV002]
已披露公开估值 03
[CV004]
已培训专业人士 05
10,000+ [CO005]
具名客户证明 06
EastWest Bank [CU005, CU006]
投资建议 07
research-more [CV010]

公司概况

Thinking Machines Data Science, Inc. 是 Stephanie Sy 于 2015 年在 Manila 创立的应用 AI 和数据工程公司。公开材料把它定位为服务东南亚受监管、数据密集型组织的专家,能力覆盖企业数据平台、 客户智能、文档智能、位置智能和生成式 AI 部署。公司看起来更像服务主导,而不是自助 SaaS 主导:它销售战略、 实施、部署和赋能工作,同时越来越多地打包可复用解决方案模板和开源地理空间资产。公开客户证明最强的部分来自 EastWest Bank 和 UNICEF 相关工作;OpenAI 合作伙伴身份以及 Temus 在 July 2026 的战略投资, 进一步增强了外部可信度。即便如此,收入、利润率、留存和融资条款披露仍然稀少。

官网
thinkingmachin.es
创始人
Stephanie Sy
创立地点
Manila, Philippines
总部
Manila, Philippines
产品
Thinking Machines 销售企业数据平台、客户智能系统、文档智能工作流、地理空间 / 位置智能产品,以及生成式 AI 部署服务。公开产品更强调安全的公有云数据基础设施、AI/ML 实施、分析工作流,以及变革管理或培训支持,而不是一个自助软件产品。
客户
菲律宾和东南亚的大型企业与机构,覆盖金融服务、零售、企业集团、公民组织,以及发展或公共利益相关场景。
商业模式
主要靠面向企业 AI 和数据平台项目的战略、实施、部署和采用工作变现;有一些产品化解决方案模板迹象, 但没有公开自助定价,也没有披露经常性软件收入。
阶段
Private growth stage with regional expansion ambitions and disclosed strategic capital from Temus in July 2026.
融资情况
Temus 于 July 2026 宣布对公司进行战略投资。公开来源未披露交易规模、估值、累计融资或融资条款。
[CO003, CO004, CO009, CO010, CO011, CE001, CE008, CE010]

执行摘要

主要优势

  • OpenAI 合作伙伴身份和 Temus 2026 年 7 月的战略投资,为这家以菲律宾为根基的 AI 服务公司提供了少见的外部背书。
  • EastWest Bank 项目和 UNICEF 相关工作已经公开验证,公司交付过可识别的生产级或机构级场景,而不是只停留在叙事里的初创公司。
  • 产品面覆盖数据平台、地理空间 AI、文档智能、客户智能和生成式 AI 部署,Thinking Machines 因此能从多个入口切入企业现代化预算。
  • 东南亚企业 AI 市场仍在结构性增长,本地落地能力和治理经验都稀缺,公司看起来进入得较早。
  • 公司做的是有治理要求、复杂度更高的企业 AI 和数据问题,因此相对普通外包有理由拿到一定溢价。

主要风险

  • 收入、利润率、留存、现金和股权结构都没有公开披露,而这些正是支撑入场估值必须拿到的信息。
  • 公开记录能支撑战略可信度,却支撑不了题目暗示的 $10B+ 估值叙事,叙事与证据错位的风险很高。
  • 业务看起来仍由服务牵引;除非公司证明产品具备经常性经济性,否则估值倍数会受限。
  • 商业杠杆明显绑在合作伙伴和少数可见标杆客户上,客户集中和依赖风险因此抬高。
  • 第 7 章还识别出监管、招聘、基础设施和交付执行风险,可能拖慢市场兴趣转化为可持续经济性的速度。

未决问题

  • 分部收入、经常性收入占比和头部客户集中度仍未披露;没有这些数据,就无法对估值做压力测试。
  • 毛利率、交付利用率、定价纪律和服务 / 软件结构都未公开,估值倍数选择因此高度主观。
  • 最近融资条款、累计融资额、股份类别和清算优先权都没有公开信息。
  • 公开客户数量信号互相冲突(110+ 与 150+),也看不出续约深度、合同期限或扩张行为。
  • 公司缺少达到公开市场标准的投资人材料或审计披露包,退出准备度不足,也限制了任何 IPO 式终值假设的置信度。

目录

Chapter 01

01公司概况

1.1 身份、定位和区域足迹

Thinking Machines Data Science, Inc. 现在把自己呈现为 AI 和数据转型公司,而不是狭义项目工作室。 其当前公开材料持续强调东南亚范围内的企业 AI 部署、云数据平台、文档智能、客户智能和生成式 AI 赋能。 公司最稳定的地理事实也相当清楚:多份 2026 年来源称其创立于 Manila,目前在 Manila、Singapore 和 Bangkok 运营;OpenAI 合作伙伴页面则列出服务国家为 Philippines、Singapore 和 Thailand。 官网还给出公开规模信号——培训超过 10,000 名专业人士、服务超过 150 个全球客户、净推荐值(Net Promoter Score)为 85——但这些数字需要谨慎使用,因为合作伙伴相关的 July 2026 披露只引用了超过 110 个已服务客户。 综合来看,Thinking Machines 最适合理解为一家区域性企业 AI 咨询与构建伙伴,根在菲律宾, 但销售的是跨境能力故事。官网强调人机协作和工作流采用,外部报道则把这套叙事连接到银行、公民技术和气候导向地理空间分析中的具体客户工作。 身份是连贯的;指标方向上亮眼;这些指标背后的精确分母仍有一部分不透明。[CO001, CO002, CO003, CO004, CO005, CO006]

KPI 快照表
指标数值 / 状态日期置信度缺口
成立时间2015(现有来源共识)2026-07-30UNICEF Venture Fund 页面列为 2016
总部 / 运营基地菲律宾马尼拉2026-07-30检索来源未披露具体街道级总部地址
现有办公室马尼拉 | 新加坡 | 曼谷2026-08-12
商业模式企业 AI 与数据转型服务2026-08-12
OpenAI 状态首个亚太服务合作伙伴;后为高级合作伙伴2026-08-12高级合作伙伴状态的具体授予日期未披露
服务客户数110+ 至 150+,视来源而定2026-07-30计数方法未披露
培训专业人士数10,000+2026-08-12
NPS852026-08-12官网自报
员工数外部区间为 51-200;确切人数未披露2026-08-12仅找到相互冲突的目录式估计,以及较早的 80+ 团队表述
已披露外部融资UNICEF Venture Fund US$449,598;2026 年 Temus 金额未披露2026-08-12无公开投后估值或所有权条款

当前规模指标混合了公司说法和合作伙伴相关报道。确切员工数、收入、董事会构成和 2026 年交易经济性仍未披露。

[CO003, CO004, CO005, CO006, CO007, CO008]
FO002: 公司快照逻辑

当前公司逻辑把 Manila 根基、能力支柱、客户验证、合作伙伴杠杆和区域规模化串在一起。

[CO001, CO002, CO003, CO004, CO005, CO014]
FO003: 快照 KPI

公开 KPI 视角,把硬运营锚点与软性、冲突或未披露指标分开。

KPI 卡片在公开记录互相冲突或缺少精确数字时,刻意用区间或“未披露”呈现。

[CO003, CO004, CO005, CO006, CO007, CO008]

1.2 创始人、领导层和治理姿态

在每一份关于 Thinking Machines 的公开描述里,Stephanie Sy 都是毫无疑问的重心。官方与第三方资料在基本事实上一致: 她创立公司,2026 年仍担任 CEO,曾就读 Stanford、任职 Google,并从 Silicon Valley 回到菲律宾创业。 Harvard Business Publishing、Hustleshare 和 Ignition 都为她的创始人故事补上类似纹理:她回国,是因为同时看到了家庭牵引和市场缺口; 随后围绕严谨解题、客户教育和叙事能力搭起公司。这段背景重要,因为公司看起来仍然高度创始人品牌化。即便 Temus 交易之后, Stephanie Sy 仍是 Thinking Machines 使命、公共声音和运营连续性中最主要的具名高管。 治理透明度弱于创始人可见度。公开检索到的材料没有列出董事会、更完整的高管梯队,也没有说明 Temus 投资后的精确控制权。 外部读者主要看到 Sy,以及 Temus 高管 Sng Ren Yeong 和 Sutowo Wong,后两者出现在合并后的 Applied AI & Data 团队语境中。 这足以理解战略方向,但不足以完整尽调接班人深度或交易后的决策权。创始人依赖仍是当前公开记录里的真实特征。[CO009, CO010, CO011, CO019, CO020, CO042]

管理层与创始人表
人物报告日期职务状态背景 / 覆盖关键人物依赖
Stephanie Sy创始人、CEO;亦任 Temus Applied AI & Data 董事总经理创始人(在任)Stanford 校友;前 Google;TM 带创始人品牌的公开面孔,也是 Temus 交易后连续性的核心信号关键
Sng Ren YeongTemus CEO外部战略支持方Temus CEO,阐述投资后的规模化理由与整合投资逻辑
Sutowo WongTemus Applied AI & Data 董事总经理外部合并团队联合负责人交易后与 Sy 共同领导合并后的 Applied AI & Data 团队

公开检索材料清楚点名 Sy,但未披露完整 Thinking Machines 高管梯队或董事会。纳入合作伙伴侧领导人,是因为他们会塑造交易后的运营模式。

[CO009, CO010, CO011, CO019, CO020, CO042]

1.3 资本基础、伙伴地图和利益相关方重要性

2026 年最重要的公司事件,是 Temus 于 30 July 2026 宣布的战略投资。尽管财务透明度仍不足,这一公告在运营层面高度重要。 Temus 本身由 Temasek 成立,它把这笔交易定位为:把 Thinking Machines 十年的交付经验,与一个规模更大的新加坡转型平台结合起来。 公告也明确,Thinking Machines 将继续使用自有品牌,客户项目不会中断;Stephanie Sy 同时会在 Temus 内担任董事总经理。 公开记录仍未给出经济条款:没有披露投资金额、估值,也没有披露所有权或治理条款。 更早的外部支持似乎来自 UNICEF Venture Fund 和更广泛的 UNICEF 创新生态。这些来源锚定了公司早期的地理空间与开源阶段, 也提供了本轮检索唯一直接取得的公开融资数字。OpenAI 的战略重要性在于获客和可信度伙伴,而不是已披露投资方。EastWest Bank 则是有意义的客户证明点,因为它显示了受监管行业内部的生产级 AI 工作。因此,利益相关方地图比不完整的股权结构表更重要。[CO014, CO015, CO016, CO017, CO018, CO019]

利益相关方或投资人地图
利益相关方角色切入点 / 证据经济或战略重要性尽调问题
Temus战略投资人和运营平台2026-07-30 战略投资公告2026 年首要规模事件;扩大交付能力和区域覆盖披露投资金额、所有权、治理权利和整合里程碑
OpenAI商业化与可信度合作伙伴APAC 服务合作伙伴公告和合作伙伴页面传递前沿模型访问和企业 AI 商业相关性信号厘清收入对 OpenAI 驱动项目的依赖,以及合作伙伴经济性
UNICEF Venture Fund早期外部支持方 / 生态伙伴UNICEF Venture Fund 毕业项目页面本轮检索到的最早公开融资数字;锚定地理空间和开源阶段厘清支持属于赠款、股权还是混合资本
EastWest Bank旗舰客户证明点官网客户证言和 2022 年 MBC 活动稿展示受监管金融服务环境中的生产级 AI 工作验证当前范围、扩张收入和案例研究时效性
公共与公民部门伙伴长期采用渠道Forbes、Ignition、UNICEF、UNDP 引用支撑“本地打造、服务本地”的可信度叙事和社会影响护城河拆分声誉收益与经常性商业贡献

利益相关方地图比股权结构表更清晰。2026 年 Temus 交易具有战略重要性,但经济性仍未披露。

[CO014, CO015, CO016, CO017, CO018, CO021]

1.4 里程碑、定位变化、矛盾和明确尽调缺口

大约十年里,Thinking Machines 看起来经历了三个可见阶段。第一阶段是菲律宾数据科学咨询公司,2018 年已经公开露面, 服务企业、政府、NGO 和创业公司。第二阶段是地理空间与社会影响,UNICEF 相关材料突出财富地图、卫星影像分析、开源工具和气候导向应用。 第三阶段是当下的企业转型,公开信息聚焦生产级 AI 系统、治理、高管培训、OpenAI 赋能的采用,以及通过 Manila、Singapore 和 Bangkok 进行区域推出。 这条轨迹可信,但公开记录比打磨后的叙事更嘈杂。不同来源对公司创立于 2015 年还是 2016 年说法不一。当前来源也不一致: 相关客户数究竟是超过 110 还是超过 150;外部名录与官方材料在员工规模区间上不一致,甚至对公司本土地域标签也不一致。 这些矛盾并不否定业务,但意味着本章应对精确员工数、董事会构成、收入和估值等未获支持的私有指标明确保持谨慎。这是扎实的身份章节, 不是完整的所有权或财务披露包。[CO012, CO013, CO023, CO024, CO025, CO026]

里程碑表
日期事件类型金额 / 状态参与方含义
2015Stephanie Sy 在马尼拉创立公司创立Stephanie Sy奠定 2026 年材料使用的菲律宾起源叙事
2017MMDA / Waze 交通分析展示早期公共部门数据工作产品Thinking Machines;MMDA;Waze 数据早期证明公司能把数据转化为运营决策
2018Stephanie Sy 入选 Forbes Asia 30 Under 30;地理空间分析成为最强新业务线规模化Stephanie Sy、Forbes 与 Thinking Machines在公司围绕地理空间 AI 专业化时,抬高创始人可见度
2019-05-26Harvard 人物稿发布 Stephanie Sy 创始人故事治理Harvard Business Publishing第三方验证创始人-市场匹配叙事
2019UNICEF Venture Fund 毕业项目页面列出已投资 $449,598,并重点介绍 GeoMancer / Tiffany融资449598UNICEF Venture Fund 与 Thinking Machines本轮唯一直接检索到的公开融资数字
2021UNICEF Innovation 人物稿称 Sustainability Team 已启动,团队超过 80 人规模化UNICEF Innovation;Thinking Machines标志公司从创业咨询公司演进为更大的主题平台
2022-03-28EastWest Bank 项目在 Makati Business Club 活动上公开讨论合作EastWest Bank、Makati Business Club 与 Thinking Machines展示受监管客户环境中的生产级 AI
2026-07-30Temus 战略投资公布;Sy 在继续担任 CEO 的同时加入 Temus 领导层融资Temus、Stephanie Sy 与 Thinking Machines改写规模叙事,但估值和控制条款仍未披露

时间线强调身份、融资、客户证明和规模的公开脉络。若干公司新闻室专题链接无法完整读取,因此只纳入有证据支持的里程碑。

[CO003, CO011, CO012, CO013, CO016, CO022]
FO001: 公司里程碑时间线

公开时间线从 Manila 创立开始,串起地理空间 R&D、客户证明,以及 2026 年 Temus 规模化事件。

部分里程碑只标年份,因为检索到的来源能确定时间点,但没有更完整的发布时间表。

[CO003, CO011, CO012, CO013, CO016, CO017]

1.5 图表

Chapter 02

02市场分析

2.1 市场边界和规模测算视角

Thinking Machines 的相关市场不是「全部 AI」,而是组织内部规划、治理、集成和运营 AI 与数据系统相关企业支出的更窄切片。 纳入的支出包括数据平台现代化、工作流集成、负责任 AI 设计、模型赋能、变革管理,以及受监管或数据密集型职能中的部署支持。 排除的支出包括前沿模型训练、芯片制造、超大规模云厂商资本开支和大众消费级 AI 应用。这个区分重要,因为菲律宾标题级 AI 市场看起来可能很大,但其中只有一小部分能被一家重实施的咨询公司变现。 公开规模来源仍能提供有用的自上而下锚点。Trade.gov 引用菲律宾 AI 市场将从 2024 年约 US$772 million 增长到 2030 年约 US$3.49 billion,区域来源则显示东南亚 AI 行业在 2024 年已经超过 US$4 billion,并且快速增长。这些数字确立了有意义的 总可用市场(TAM)。但公开证据没有给出治理密集型企业 AI 服务在菲律宾的精确可服务市场(SAM)数字,因此更可靠的视角是采用成熟度: 实验普遍、规模化不均、强烈需要能打通概念验证与生产之间缺口的伙伴。 2026 年实务尽调结论是,市场规模必须和执行深度一起读。在许多买家仍把实验转化为受治理工作流的市场里,像 Thinking Machines 这样的公司能拿到的收入池,更多由交付复杂度和组织变革决定,而不是由 AI 用户数量简单相加决定。[CM001, CM002, CM024, CM025, CM026, CM031]

市场定义表
细分 / 类别纳入支出排除支出买方 / 付款方与 Thinking Machines 的相关性
企业 AI 战略与治理用例发现、政策设计、数据治理、风险框架独立法律抗辩、仅审计项目CEO / CIO / COO,合规团队支持
数据平台现代化云数据平台、管道、仪表盘、存储现代化商品化托管和超大规模云厂商资本开支CIO / CDO / 运营
工作流集成与变革管理将 AI 嵌入服务、运营和分析工作流无实施的现成席位授权业务单元,IT / 安全把关
模型启用与应用构建ChatGPT Enterprise 启用、智能体应用设计、文档 AI、面向客户的 AI前沿模型训练或基础模型 R&DCIO / CTO / 产品负责人
基础设施扩张本地云采用、边缘就绪、数据中心邻近性芯片制造、电厂资本开支、超大规模云厂商园区建设基础设施 / 平台团队间接
消费 AI 和无关外包不适用于 TM 核心服务大众市场应用、通用劳动力套利、无关 BPO 席位消费者或商品化采购团队

市场边界聚焦实施牵引的企业 AI 和数据转型预算,而不是完整 AI 技术栈。

[CM024, CM025, CM026]
TAM / SAM / SOM 或规模测算视角表
视角发布方 / 依据年份地理范围数值方法 / 含义置信度局限
AI 市场总盘Trade.gov / UNESCO 简介2024菲律宾US$772M美国出口指南使用的全国 AI 市场基准不特指仅服务支出
AI 市场总盘预测Trade.gov / UNESCO 简介2030菲律宾US$3.49B前瞻市场预测,隐含 28.6% CAGR预测,不是实际支出
企业使用发生率Swarm 调查2026菲律宾92% 组织使用 AI受访组织中使用任何 AI 的占比使用不等于支出
超越试点的成熟度下限由 Swarm 调查推导2026菲律宾35% 超过 POC100% 减去仍处概念验证阶段的 65%推导采用指标,不是支出
区域 AI 板块锚点Source of Asia 公司2024东南亚>US$4B区域 AI 板块价值锚区域范围宽,且为二手方法
区域部署成熟度基准EDB2026东南亚46% 超过试点走出试点的企业复合加权占比区域复合口径,不限菲律宾
TM 可服务推断作者综合部署导向来源2026菲律宾 / 东南亚未披露可服务市场是需要治理、集成、培训和平台工作的预算子集没有直接公开 SAM / SOM 数字

本章有意使用多个视角,因为公开来源没有披露干净的仅服务口径 SAM 或公司特定 SOM。

[CM001, CM006, CM007, CM031, CM032, CM039]
FM001: 市场规模测算视角

由东南亚 AI 机会总量,一层层收窄到 Thinking Machines 以实施牵引的可服务切入口。

底层采用定性口径,因为没有公开来源直接给出仅服务类 SAM 或 TM 专属 SOM。

[CM001, CM006, CM007, CM024, CM025, CM031]
FM002: 市场估算区间

采用成熟度区间:下限取菲律宾越过试点的水平,上限参照东南亚与新加坡基准。

该区间比较的是采用成熟度,不是市场支出规模;纳入它是因为公开 SAM/SOM 估算不可得。

[CM007, CM032, CM033]

2.2 买家分层、工作流优先级和采用路径

买家地图由几类细分市场带头:这些市场里,数据质量、合规和工作流重设计与模型选择同样重要。在菲律宾,最适合 Thinking Machines 式工作的垂直行业是 IT-BPM/BPO、金融服务、零售与企业集团、公共部门、电信 / 物流和医疗健康。Trade.gov、Swarm 和公司证据都指向这些行业已经在主动采用。其内部,经济买家通常是 CEO、COO、CIO、CTO 或业务单元负责人;日常用户是运营、 服务、分析和一线团队;真正的守门人则是数据、安全和合规职能。 Swarm 的调研尤其有用,因为它展示的是运营层面的采用,而不是口号。组织已经在用 AI 做内部自动化、内容生成和数据分析, 几乎一半表示自己处于 AI 应用开发模式。但多数组织仍在消费供应商工具,而不是自建专有技术栈。这种模式更利好实施伙伴,而不是纯模型厂商: 难点不是发明模型,而是选择工具、把工具集成进系统、安全治理,并帮助组织围绕这些工具重设计工作流。 这也解释了为什么买家地图看起来比即时收入兑现更宽。许多客户可能从培训、政策或狭窄工作流自动化开始;当信任、数据就绪度和合规舒适度提升后, 才扩展到更广的平台或托管交付范围。[CM004, CM005, CM006, CM007, CM008, CM009]

细分 / 买方地图
细分买方用户付款方 / 预算所有者工作流 / 预算采用触发因素
IT-BPM / BPOCOO / 转型负责人运营经理、坐席、QA、分析运营 / 转型预算自动化、分析、客户支持成本压力和质量提升
BFSICIO / COO / 数字化负责人欺诈、风险、服务、网点运营技术和业务单元预算欺诈检测、对账、服务 AI符合治理要求的生产率提升
零售 / 综合集团业务单元负责人 / CIO营销、商品管理、供应链数字化 / 分析预算客户智能、预测、文档流利润率改善和个性化
公共部门机构负责人 / 项目负责人政策、规划、一线员工机构现代化预算公民服务、规划、韧性分析服务质量和政策要求
电信 / 物流COO / 网络或服务负责人客户运营、路由、规划团队运营 / 网络预算需求预测、路由、分群规模复杂度和数据密度
医疗健康医院管理者 / 数字化负责人临床运营、行政人员转型 / IT 预算文档 AI、分诊、分析效率与合规压力

预算归属因行业而异,但实施成功几乎总是取决于 IT、安全、数据和业务运营协同。

[CM004, CM005, CM008, CM009, CM027, CM028]
FM003: 细分市场吸引力与治理强度图

矩阵比较菲律宾主要买方细分市场的实施强度、治理负担和供应商替代风险。

定性评级综合了采用、治理和可服务性证据,并不代表第三方打分数据集。

[CM027, CM028, CM039, CM042]
FM004: 采用漏斗 / 价值链图

菲律宾企业旅程从广泛实验走向更窄、受治理约束的生产部署。

最后阶段是作者设定的代理指标,用于可视化治理和集成关卡后的收窄;公开来源对瓶颈的量化比最终生产占比更清楚。

[CM006, CM007, CM010, CM012, CM019, CM022]

2.3 增长驱动、监管和基础设施约束

三股力量正在同时推动市场增长。第一,宏观需求上升:菲律宾组织想获得生产率提升,尤其是在 BPO、银行、客户运营和决策支持中。 第二,NAISR 2.0、CAIR 和更广泛的数字基础设施推动,让国家政策更明确。第三,区域背景让 AI 现代化更难推迟: Singapore、Thailand 及邻近 ASEAN 市场正在快速增加云、数据中心和 AI 治理能力,足以重设竞争预期。 约束同样真实。Trade.gov、OECD、NPC、BSP、Swarm 和 Nexdigm 都指出技术栈不同位置存在摩擦:人才短缺、隐私和安全担忧、 法律不确定性、薄弱的内部数据战略能力、电力成本、供电可靠性,以及热情与制度化之间挥之不去的差距。Thinking Machines 会从中受益,但逻辑并不简单。如果市场已经具备完美的自助 AI 成熟度,对实施伙伴的需求会下降。菲律宾不是这样的市场。拖慢采用的同一组条件——治理负担、 工作流重设计和基础设施脆弱性——也在为深度介入的交付模式创造付费需求。 东南亚基准进一步强化这一点。Singapore 已经是更高成熟度部署的区域试验场;菲律宾作为服务主导的现代化市场,比作为独立基础设施枢纽更有吸引力。 这种不对称对 Thinking Machines 有用,因为它的产品最强之处,正是在买家需要把可用模型转化为受治理企业工作流时提供帮助。[CM003, CM010, CM011, CM013, CM014, CM015]

增长驱动因素与约束表
驱动因素 / 约束方向时点影响尽调问题
NAISR 2.0 与 CAIR正向2024-2026政府明确推动 AI 采用、R&D 和治理这项政策会有多少转成采购或补助?
IT-BPM 自动化压力正向当前大型出口行业持续带来工作流 AI 需求哪些细分垂直领域已经在买,哪些还在探索?
云与数据中心扩张正向当前 / 中期提高生产部署和平台工程落地的可行性哪些云 / 托管机房新增资源真正触达中端市场买家?
广泛试验正向当前92% 使用率说明认知已不是核心瓶颈试点能多快转成有治理的生产系统?
人才短缺负向当前57% 的阻碍占比让内部团队难以独自规模化TM 能否稳定把赋能和培训变现?
安全与隐私顾虑负向当前受监管买家需要治理投入重的实施伙伴买家实际要求哪些隐私内建设计产物?
法律与监管不确定性负向当前 / 中期待审 AI 相关法案和不断演进的指引拖慢采购信心行业专项规则会多快定型?
电力成本与可靠性负向中期抬高基础设施成本,也可能拖慢超大规模云周边采用本地基础设施成熟度会不会限制 Manila 走廊以外的 AI 工作负载?

拖慢菲律宾 AI 规模化的因素,也给能降低部署风险的高触达实施厂商留下空间。

[CM003, CM010, CM013, CM015, CM017, CM019]

2.4 图表

Chapter 03

03竞争格局

3.1 格局:既有巨头、本地公司、平台和替代方案

Thinking Machines 周围的竞争格局,应按待完成任务来分组,而不是贴上简单的「AI 公司」标签。第一类是 Accenture 和 IBM 等全球既有巨头,它们能把 AI 打包进更大的云、数据和企业转型项目。第二类是 AWS Professional Services 和 Google Cloud Consulting 等平台原生服务部门,它们直接争夺高价值部署,同时也赋能伙伴生态。第三类是 NCS 以及现在的 Temus 等区域集成商, 它们带来新加坡规模、信任和东南亚运营覆盖。第四类是 Stratpoint、Exist 和 Senti 等菲律宾本地重叠者,它们各自与 Thinking Machines 的产品在较窄切片上重合。最后,对有能力的企业团队而言,内部自建仍是替代方案。 这套分层重要,因为没有单一竞争者主导所有层级。买家可以从既有巨头采购战略,从超大规模云厂商采购云执行,从小众专家采购语言工具, 再由内部团队或本地伙伴完成内部工作流集成。因此,Thinking Machines 获胜不是靠消灭替代项,而是靠占住一个有用的中间位置: 比全球系统集成商更专注、更有本地根基;又比狭窄 AI 精品店更宽、更适合企业。 尽调时的关键错误,是只拿 Thinking Machines 和纸面上看起来相似的公司比较。实践中,公司真正面对的是更大的转型项目、 云相邻服务、点解决方案供应商,以及既有企业团队的惯性。[CP001, CP004, CP005, CP007, CP009, CP010]

竞争对手画像表
竞争对手类别规模 / 融资目标客群差异化局限
Thinking Machines区域精品 AI / 数据集成商Temus 支持的战略投资;110+ 至 150+ 客户;培训 10k+ 人东南亚企业、BFSI、零售、集团企业、公民社会部门从数据平台到 AI 的连续性;OpenAI 伙伴信号;菲律宾根基规模小于 GSI 和超大规模云厂商
Accenture全球既有集成商799k 员工;120+ 个国家 9k+ 客户大型企业和转型买家规模、采购准入、广泛转型范围本地针对性和精品服务感较弱
IBM Consulting全球既有集成商170+ 个国家 300k+ 员工受监管和混合云企业软件 + 咨询 + 基础设施与联盟可能显得平台 / 伙伴导向过重,而不是本地精品服务
AWS Professional Services平台原生服务AWS 平台和交付中心支持AWS 对齐企业深度云邻近和 AI 框架强烈拉向 AWS 技术栈
Google Cloud Consulting平台原生服务Google Cloud 和伙伴生态支持Google Cloud 对齐企业Google 工程能力 + 带伙伴一起交付强烈拉向 Google 技术栈
NCS区域集成商按厂商收入计的东南亚服务市场份额 #1(IDC 2025H1,公司引用)东南亚公私部门企业区域信任、托管服务、新加坡可信度菲律宾根基不够鲜明
Stratpoint本地数字化转型同业25+ 年;菲律宾本土 AWS 服务提供商菲律宾企业应用、云和数据项目以 AWS 为中心的本地交付,已有客户证明数字化范围更宽,AI 专家定位较弱
Exist本地工程主导同业获奖菲律宾软件公司;数据与 AI 是一大支柱企业软件与数据项目合规 / 韧性工程,以及数据 / AI 支持AI 原生市场定位较弱
Senti AI本地垂直 AI 专家2024 年被 Kollab 收购客服、对话式 AI、NLP 买家Tagalog / Taglish 对话式 AI 和产品化方案范围窄于 TM 的完整数据平台 + AI 供给
内部自建现状替代方案取决于企业人手和内部预算能力较强的大型企业控制权和定制贴合度人才、治理和规模化负担

最相关的比较是能力与问题的匹配度,而不是每个玩家是否自称 AI 公司。

[CP003, CP005, CP007, CP009, CP010, CP011]
FP001: 竞争定位图

按本地 / 区域客户贴近度与交付规模广度,对关键竞争者做序数定位。

轴分数是分析师基于本轮公开证据给出的序数估计,不是实测量化指数。

[CP020, CP021, CP029, CP037, CP038, CP040]

3.2 能力宽度、包装和分销力量

Thinking Machines 最站得住脚的公开定位,是从数据基础到受治理 AI 部署之间的连续性。其官方材料、OpenAI 资料和 Temus 公告都强调在真实工作流中设计、部署和采用 AI。这比 Senti 明确产品化的对话式 AI 赛道更宽,也比 Stratpoint 或 Exist 更明确地围绕 AI;后两者是在更宽的软件或工程组合里呈现数据 / AI。但 Thinking Machines 的广度仍远不及 Accenture、 IBM、AWS、Google 或 NCS 的全球覆盖和多服务宽度。 包装方式反映同一模式。多数面向企业的竞争者披露的是能力,而不是透明价格,暗示销售以报价制项目为主。Senti 通过具名解决方案更明显地产品化; 超大规模云厂商服务通常围绕自家平台作为转型加速器出售。分销同样不对称。全球既有巨头能触达更大预算、更广泛联盟和成熟采购渠道。 Thinking Machines 与 Temus 结合缩小了差距,但没有抹平差距。实务结果是,Thinking Machines 可能在买家需要动手集成、 治理和上下文理解,而不是最低成本人力或最广全球系统集成足迹时最有竞争力。[CP002, CP003, CP008, CP012, CP014, CP017]

功能 / 能力矩阵
购买标准Thinking Machines全球 GSI(Accenture / IBM)平台原生服务(AWS / GCP)区域集成商(NCS / Temus)本地同业(Stratpoint / Exist)Senti内部自建
数据平台现代化全面全面自有云上部分至全面全面全面定制
有治理的企业 AI 部署全面全面自有云上部分至全面全面部分部分定制
变革管理 / 培训全面全面部分部分至全面部分部分定制
OpenAI 专项服务信号全面未知 / 不一未知 / 不一Unknown
本地语言对话式 AI部分部分部分部分Unknown全面定制
企业转型覆盖范围部分全面部分全面部分部分
公有云邻近性全面全面全面全面全面(Stratpoint 偏 AWS)部分定制
菲律宾本地精品服务关注度全面

未知或不一的单元格反映缺少已获取的公开证据,而不是负面评价。

[CP001, CP008, CP009, CP010, CP017, CP018]
定价 / 打包对比
竞争对手合同模式覆盖能力折扣 / 未知项含义
Thinking Machines定制项目 / 咨询 / 培训战略、数据平台、部署、变革管理、AI 赋能无公开标价只有交付持续差异化,才能支撑高端定位
Accenture定制企业项目转型、数据、AI、运营模式变革实际定价和折扣未披露可把 AI 打包进更大预算
IBM Consulting定制企业项目咨询、混合云、联盟、数据 / AI 交付实际定价未披露联盟导向打法可能拿下大型受监管客户
AWS Professional Services围绕 AWS 的定制服务迁移、现代化、AI 框架、专项方案定价可能经谈判,并与 AWS 用量挂钩可压价或用打包压过独立云中立项目
Google Cloud Consulting围绕 Google Cloud 的定制服务战略、工程、伙伴、实施支持定价未披露买家标准化到 GCP 时更强
NCS定制咨询 / 托管服务AI 驱动转型、托管 IT、CX 现代化公开定价未披露区域托管服务深度是采购优势
Stratpoint / Exist定制项目 / 托管交付云、软件、数据,以及部分 AI 项目公开定价未披露可凭本地关系和更广工程能力积极竞争
Senti更产品化 + 定制服务对话式 AI 产品、NLP、服务商业条款未公开可能比综合咨询公司更快赢下点状方案
内部自建薪酬 / 供应商 / 资本开支预算定制建设、治理、工具、集成真实成本通常不透明前期看似更便宜,但常把成本转成执行风险

公开定价不透明本身就是一个发现:这个市场卖的是结果和可信度,不是透明费率卡。

[CP026, CP027, CP028, CP031, CP035]
FP002: 功能广度 / 能力图

汇总展示菲律宾和东南亚企业 AI 买方最关键的七项能力维度。

强弱评级由检索到的公开材料综合而来;覆盖未知或不稳定的地方,应直接向供应商验证。

[CP018, CP019, CP020, CP021, CP027, CP032]

3.3 切换成本、护城河持久性和反向证据

反向证据真实存在,不应低估。Thinking Machines 似乎没有硬技术垄断,也没有受益于明显有利于溢价捕获的公开定价不透明。 超大规模云厂商越来越能把 AI 服务与核心云支出绑定,全球咨询公司也能把 AI 打包进规模大得多的转型项目。OpenAI 合作伙伴身份有帮助, 但随着伙伴网络扩张,仅凭徽章不太可能形成持久排他性。内部自建也是真实替代方案,尤其适用于工程能力强的企业。 缓和这一点的是,东南亚企业采用 AI 最难的部分很少只是模型访问。真正难的是在混乱条件下编排:碎片化数据、治理检查、复杂工作流和组织变革。 Temus 明确称 Thinking Machines 擅长这一层;系统一旦嵌入客户管线和决策流程,切换成本也正是在这一层上升。因此,护城河看起来是执行型和关系型, 不是平台垄断型。持久优势取决于公司能否在大型集成商和云生态进一步压缩品类之前,把执行声誉转化为可复用的区域分销。 因此,客户推荐和上线后的扩张,比口号式定位更重要。如果 Thinking Machines 反复成为企业在试点卡住后信任的团队, 其竞争姿态会强于简单员工数比较所暗示的水平;如果不能,市场可能被压向更大的集成商或更窄的工具。[CP022, CP023, CP024, CP025, CP031, CP033]

护城河耐久性 / 竞争风险登记表
护城河主张威胁严重程度缓解措施 / 尽调问题
OpenAI 伙伴徽章伙伴网络扩张会稀释独占性向管理层询问徽章带来的赢单贡献和留存效果
菲律宾根基的企业可信度GSI 招本地团队或收购本地能力验证本地根基是否真的缩短销售周期或提高赢率
数据平台 + AI 连续性云厂商和集成商打包类似端到端供给要求案例级证据,证明投产速度快于同业
高触达变革管理培训和咨询很快商品化把工作坊收入与嵌入式部署收入拆开
Temus 支持的区域扩张合并后整合复杂度或品牌稀释审查 Temus 后的管线转化和交叉销售证据
嵌入式工作流集成客户可能在建设阶段后转为内包评估托管服务续约率和上线后扩张历史
精品服务关注度企业采购可能偏好更大厂商梳理平均客单价上限与采购门槛
本地语言 / 语境贴合度Senti 等垂直专家可能赢下更明确的产品化用例厘清 TM 是合作、竞争,还是避开这些赛道

护城河有合理性,但公开证据更支持执行护城河,而不是硬技术垄断。

[CP021, CP022, CP024, CP028, CP029, CP033]
FP003: 护城河 / 准备度 KPI

用紧凑视图展示 Thinking Machines 的竞争韧性,以及本章识别的市场压力。

分数是分析师基于公开证据给出的序数判断,不是统计模型输出。

[CP020, CP021, CP033, CP034, CP035, CP036]

3.4 图表

Chapter 04

04财务情况

4.1 收入模式和市场打法

Thinking Machines 的公开信息指向一个围绕企业服务而非产品主导软件搭建的收入模式。公司销售数据平台、工作流 AI、生成式 AI 采用,以及客户智能、文档智能、位置智能等领域化解决方案包周边的咨询与交付工作。联系流程、解决方案描述和面向部署的语言,都指向咨询式范围界定和报价制合同。 公开证据不支持经典自助 SaaS 销售动作,也不支持规模可观的按用量变现。 市场打法也看起来原生属于企业销售。公司靠可引用部署、思想领导力和伙伴信号来营销可信度。官网免费课程最适合理解为需求生成, 而不是直接收入证明。OpenAI 合作伙伴身份和 Temus 关系可能扩大线索流和信任,而 EastWest Bank 等客户引用说明公司能拿下受监管客户; 成功实施后,扩张收入可能随之而来。这类业务大概率通过项目、扩展、赋能,以及可能存在的持续支持入账,而不是通过透明席位定价。 这也意味着收入确认大概率更多按里程碑或项目发生,而不是默认经常性。财务质量因此取决于初始部署能否可靠扩展为更广的平台、支持或后续实施范围。[CI002, CI003, CI004, CI005, CI006, CI007]

收入来源表
收入流机制单位当前价值 / 状态质量尽调问题
企业 AI 战略与采用咨询项目 / 顾问费项目 / 冲刺有公开证据,未量化仅发现阶段与实施阶段各占多少?
数据平台现代化设计 / 建设 / 迁移项目项目 / 阶段官网有公开证据中-高各阶段平均合同额和毛利率是多少?
定制 AI 工作流实施企业范围部署项目 / 上线有公开证据,未量化中-高首次部署后带来多少扩张收入?
领域方案(客户 / 文档 / 地点)方案模板 + 服务项目 / 模块有公开证据,打包方式不清这是可复用加速器,还是定制开发?
培训 / 赋能工作坊、课程、领导层赋能队列 / 场次由 10k+ 受训人数和免费课程漏斗佐证低-中收入中教育与交付各占多少?
经常性支持 / 托管服务部署后支持或优化固定顾问费 / 托管范围可能存在,但公开资料未量化是否存在合同化经常性收入基础?

检索到的证据支持混合服务模式,但无法清晰拆分项目收入与经常性收入。

[CI002, CI003, CI004, CI005, CI028]
定价 / 变现表
价格 / 单位 / 合同标价与实际成交价折扣 / 未知项来源含义
企业咨询范围无公开标价实际成交价未知官方 TM 网站企业销售以报价为准
数据平台项目无公开标价折扣未知官方 TM 网站可能按复杂度和时间线定范围
解决方案包无公开标价打包复用程度不清官方 TM 解决方案页面若可重复交付,可能改善毛利
培训 / 赋能公开展示免费课程;付费培训未标价变现不清主页 / OpenAI 资料页漏斗可能先于企业销售
伙伴主导工作商业分成未披露渠道经济性未知OpenAI / Temus 来源伙伴打法可能比标价更重要
托管支持无公开商业条款续约结构未知无直接公开证据经常性收入基础未验证

未公开定价本身就是一个发现,也提高了尽调中核实实际费率和折扣的重要性。

[CI005, CI006, CI007, CI008, CI009, CI030]
FI001: 收入模式桥接图

Thinking Machines 似乎靠边界清晰的企业 AI 项目变现,并可把项目延展到更广的平台建设与采用推进。

[CI002, CI003, CI004, CI005, CI009, CI028]

4.2 成本结构、单位经济和资本强度

由于公开产品服务占比高,经济引擎很可能是劳动力。高级顾问、数据工程师、ML 从业者和面向客户的变革专家,可能是主要成本中心; 云 / 工具转嫁成本、售前、差旅和内部赋能是次要成本。这种结构通常让利用率、人员结构杠杆和变更单纪律成为最重要的毛利率变量。 也就是说,客户数或培训覆盖等公开牵引代理指标,不如积压订单质量、平均交易规模和转为重复业务的工作占比重要。 该模式的正面,是相对基础设施重型 AI 企业固定资本开支较低。Thinking Machines 似乎不拥有数据中心或算力集群, 因此财务风险应更多来自人员成本承诺和营运资本节奏,而不是资产融资。负面在于,项目型业务外部看起来可能很强, 但如果回款慢、范围漂移或扩张工作未能复现,底层会隐藏利润率波动。公开证据没有揭示公司的实际利用率、DSO 或毛利率画像, 因此单位经济分析仍主要是一张驱动因素地图加尽调问题清单。 换句话说,业务可能有吸引人的智力杠杆,但没有软件式财务杠杆。可复用加速器能帮忙;但除非它们实质压缩交付工时或提高扩张率, 利润率仍主要由人员经济决定。[CI017, CI018, CI019, CI020, CI021, CI022]

单位经济性表
指标数值 / null置信度为什么重要尽调要求
平均合同价值null需要它把客户数代理指标换算成收入按服务线和行业拆分 ACV
按服务线划分的毛利率null检验数据平台或 AI 业务是否结构性更赚钱提供过去 12 个月的毛利瀑布
顾问利用率null服务经济性的核心驱动按角色提供可计费利用率区间
销售周期长度null影响营运资本和招聘节奏提供合格线索到签约的中位天数
扩张 / 后续项目率null衡量收入质量和账户深度提供首单收入与后续收入的队列数据
DSO / 回款null营运资本风险的关键输入提供应收账款账龄和合同付款条款
伙伴来源管线占比null检验 OpenAI / Temus 渠道价值按渠道提供签约订单来源结构
培训收入占比null区分漏斗活动和已变现教育提供已签培训收入占总额比例

本章有意将公开证据不足的单位经济性字段留为 null。

[CI019, CI020, CI030, CI034]
FI002: 单位经济模型桥接图

服务业务经济性可能不太靠许可证杠杆,更多取决于人员结构、利用率、范围控制和复购。

[CI017, CI018, CI019, CI020, CI030]
FI004: 资本强度 / 现金流图

现金使用可能流向招聘、售前、交付人才储备和区域扩张,而非自有基础设施。

[CI021, CI022, CI023, CI025, CI035, CI037]

4.3 资本充足性、战略融资和披露限制

Temus 交易是公开资本分析的重心。多方来源称其为一笔战略性后期投资,意在扩大区域足迹、提高生产级交付能力,同时保留领导层和运营。 这在方向上有利:它暗示外部资本可得,公司是在相关性基础上扩张,而不是明显陷入困境。与此同时,DealStreetAsia 明确指出财务细节未披露; 没有公开证据揭示投资规模、手头现金、烧钱速度或现金跑道。 因此,本章主要结论关乎披露质量,而不是一个隐藏数字答案。注册来源显示公司存在并已注册成立,但开放网络能访问的私营公司详细文件有限。 相比之下,IBM 和 Accenture 的上市公司门户展示了成熟可比公司拥有多少更多财务背景。Thinking Machines 缺少收入、 利润率、债务和现金数据,资本充足性无法证实。公平结论是业务看起来商业可信、战略融资成立; 但没有管理层数据室资料,就无法严格承销财务。 承销后果很简单:即便战略叙事有利,也不能替代管理账目。没有管理账目,投资者实际上是在承销创始人声誉、客户证明和战略伙伴验证, 而不是一套可见财务报表故事。[CI001, CI011, CI012, CI013, CI014, CI015]

资本充足性表
指标公开状态为什么重要当前判断尽调要求
账面现金未披露现金跑道核心输入Unknown提供最新非受限现金余额
月度烧钱未披露显示融资依赖度Unknown提供平均月度净烧钱
现金跑道(月)未披露检验下一轮融资紧迫性Unknown提供基准与增长计划下的现金跑道
计划资金用途部分披露显示资本用途偏防守还是进攻区域扩张和交付扩容提供招聘、GTM 和运营的具体分配
下一轮触发条件未披露显示未来融资风险Unknown解释会触发新增资本需求的里程碑
债务 / 项目融资义务公开资料未见可能形成隐藏下行风险未发现公开证据确认债务、担保和表外承诺

公开资料透露了 Temus 交易的战略意图,但没有给出做投资判断所需的资产负债表事实。

[CI013, CI014, CI015, CI016, CI023, CI024]
公开财务缺口表
缺失的私有指标影响具体尽调路径
按收入流划分的收入卡住收入质量评估索取过去 24 个月按战略、建设、支持和培训拆分的收入
按收入流划分的毛利率卡住毛利路径分析索取带成本的项目损益表,以及按服务线划分的混合毛利率
现金 / 烧钱 / 现金跑道卡住资本充足性判断索取最新管理账和月度现金桥
积压订单和管线质量卡住未来收入可信度索取已签积压订单、加权管线和伙伴来源管线明细
回款 / DSO卡住营运资本评估索取应收账款账龄和标准开票里程碑
扩张与留存卡住耐久性评估索取从首单到扩张 / 续约收入的队列视图

公司可能财务健康,但公开证据太薄,无法证明。

[CI012, CI026, CI033, CI034, CI038, CI040]
FI003: 财务估算区间

公开估值输入项的可见度不均:法律存续和业务牵引代理指标可观察,但收入、现金和利润率基本不透明。

分数是分析师基于公开证据给出的可见度评级;用于诊断估值输入项,不代表财务表现。

[CI001, CI011, CI012, CI026, CI032, CI033]

4.4 图表

Chapter 05

05产品与技术

5.1 组合、模块和客户工作流匹配

Thinking Machines 的产品呈现最适合理解为分层的服务与解决方案组合。底层是数据基础和云数据平台;其上是客户智能、文档智能、 位置智能等应用模块;再往上,较新的生成式 AI 层围绕基础模型生态做采用、评估和工作流集成。这不是一个用简单席位模式销售的单一横向平台, 而是一组通过企业实施交付的可复用解决方案模式。 放到客户工作流里,这些模块对应具体任务:统一并细分客户数据,从大型文档集里搜索和抽取,把卫星或位置数据转化为决策, 并把 GenAI 部署进安全业务工作流。这种广度重要,因为它让 Thinking Machines 能在模型选择前后都接触企业——通常是在数据准备、 集成和采用这些更有价值的阶段。取舍在于,公开页面更能证明用例多样性,而不是标准化产品经济。 这种运营形态对尽调重要,因为它解释了为什么客户价值和产品经济可能分叉。可复用工作流并不自动等于一个自洽的软件 SKU; 在 Thinking Machines 这里,复用性看起来存在于交付打法、模型库和领域模板中。[CE001, CE004, CE006, CE008, CE010, CE016]

产品模块 / 资产矩阵
模块 / 资产用户状态 / 成熟度差异化尽调缺口
数据平台数据 / 分析团队成熟云原生企业数据底座需要具名架构参考和支持指标
客户智能营销、分析、CRM 团队成熟身份匹配 + 分群 + AI 模型库需要基准准确率和部署数量
文档智能知识工作者、运营、合规成熟跨数百万份文档的搜索 + 抽取需要精确率 / 召回率和延迟指标
位置智能战略、规划、地理空间用户差异化 / 先进卫星影像 + 地理空间 AI + 数据合作需要收入贡献和模型性能基准
工作场景生成式 AI高管和企业团队新兴但已打包生产部署框架 + 培训 + 治理需要路线图和可重复性指标
开源地理空间工具研究人员 / 工程师真实存在但支持较轻开发者信号和技术深度需要厘清商业用途与社区用途

模块地图显示,Thinking Machines 在数据工程与应用 AI 交汇处覆盖最广。

[CE001, CE004, CE006, CE008, CE010, CE020]
工作流 / 用例表
用户任务当前工作流公司方案可衡量收益限制
统一碎片化客户数据系统间记录割裂客户智能黄金客户集和分群无公开转化 / 提升指标
从散乱文档中挖价值人工阅读和搜索文档智能结构化抽取和可扩展搜索无公开精确率基准
把位置数据转成决策临时 GIS 或外包分析位置智能地理空间 AI 和遥感洞察需要更强的真实产品证据
为 AI 准备数据混乱、割裂的数据管线数据平台安全摄取 / 转换 / 分析底座无公开部署时间基准
从 AI 试点走向生产零散试验生成式 AI 框架 + 变革管理面向生产的推广路径层级较新,公开案例细节较少
提升企业团队技能AI 熟练度低培训和高管赋能采用准备度和治理提升培训经济性未公开

客户工作流契合度清楚;公开指标体系不够清楚。

[CE003, CE004, CE006, CE008, CE010, CE013]
FE002: 客户工作流 / 运营流程

Thinking Machines 通常从业务问题切入,再搭建生产采用所需的数据和 AI 工作流。

[CE010, CE013, CE014, CE016, CE036]

5.2 架构、依赖和成熟度

官方材料暗示的架构是云原生、集成密集且有意保持供应商灵活。Thinking Machines 反复声称可部署在主要公有云、安全存储库、 公有云文档平台,并可从结构化和非结构化来源摄取数据。即便没有私有架构图,依赖地图也很清楚:公司依赖公有云基础设施、 OpenAI 等外部基础模型生态、客户数据访问,以及特定领域的数据伙伴。模式不是拥有最深基础设施,而是把多层编排成生产系统。 成熟度不均,但可信。数据平台和经典分析工作流看起来是最老、最稳定的一层。对这类公司而言,地理空间 AI 显得异常深: 有长期 UNICEF 相关工作、开源工具和可持续应用。GenAI 产品较新,但比一份泛泛咨询材料更明确:教育、实验、执行、实施框架和银行用例引用。 因此,公开证据支持存在真实部署栈,但不构成标准化可靠性指标或公开更新日志纪律的正式证明。 因此,架构看起来务实,而不是垂直一体化。这样可以加速部署,并让公司贴近客户需求;但也意味着技术栈继承了云、伙伴模型和数据访问依赖, 其中一部分不在公司直接控制之内。[CE002, CE003, CE007, CE009, CE017, CE018]

技术 / 运营架构表
层级 / 流程 / 组件角色依赖风险
公有云数据基础设施计算、存储、分析底座主要公有云供应商云成本 / 供应商依赖
数据摄取与转换规范化结构化和非结构化输入客户数据访问 + 摄取工具数据质量瓶颈
模型层(传统 ML / 生成式 AI)预测、抽取、生成、计算机视觉外部模型 + 内部模型库模型漂移或供应商变化
应用和工作流层搜索、仪表盘、智能助手、定制应用软件设计和客户系统集成复杂度
人机协同落地层培训、治理、推广、变革管理客户流程负责人和操作人员工作流不匹配时采用率低
地理空间数据合作遥感和增强空间数据集第三方数据访问和权利数据权利 / 可用性风险

架构模块化、优先做集成,但也高度依赖外部环节。

[CE002, CE003, CE017, CE018, CE019, CE029]
路线图 / 发布 / 开发阶段表
日期 / 阶段功能 / 里程碑状态含义来源
2018-2021地理空间业务线成为主要业务,并组建 Sustainability Team历史 / 已佐证显示其 R&D 脉络比普通生成式 AI 新进入者更深UNICEF 来源
2024-2026OpenAI 服务合作伙伴及 Advanced Partner 进阶当前显示生成式 AI 方案打包更成型、伙伴入口更紧密主页 / OpenAI / TechEDT
2026Temus 整合当前扩大工程和部署班底Temus / TNGlobal / Context
2026曼谷办公室和更广泛的东南亚支持当前提升区域交付覆盖公司案例 / Temus
当前面向银行和零售的生产级 GenAI 框架当前说法显示产品化较新,但推进有意图GenAI 页面
当前开源地理空间包,仅按尽力而为维护当前信号有用,但不能保证企业级支持GitHub 代码库

公开可见的路线图主要由组织扩张和产品包装牵引,不是传统软件发布日历。

[CE021, CE022, CE023, CE025, CE033, CE034]
FE001: 产品架构图

从云 / 数据基础到应用模块与 GenAI 部署的分层视图。

[CE001, CE002, CE003, CE017, CE018]
FE003: 关键依赖图

公司的技术栈依赖云基础设施、伙伴模型、客户数据和专业地理空间输入。

[CE018, CE019, CE020, CE021, CE022, CE030]

5.3 知识产权、信任控制和工程文化

Thinking Machines 最好的公开差异化不是专利墙,而是领域专用应用模型、开源地理空间工具,以及把研究采用和企业实施混在一起的交付文化。 UNICEF 来源把公司与 GeoMancer、Tiffany、GeoWrangler、财富估算模型、PM2.5 估算和 Sustainability Team 连接起来。 GitHub 又提供了开发者信号证据,说明这些产物是真实包或工具,而不只是幻灯片。这比许多咨询式 AI 公司公开提供的技术证据更强。 信任与安全图景体面但不完整。官方页面强调安全存储库、加密、云无关部署、治理、以人为中心的推出和变革管理;OpenAI 与 Temus 材料强化了公司在受监管、混乱的企业环境中运营。但公开证据没有确认许多大型软件买家期待的正式安全和可靠性产物, 例如公开认证、正常运行时间报告或独立基准性能。因此,产品技术结论对应用广度和工程文化为正面,但对硬护城河和外部验证控制只能给中等。 尽调的关键技术问题,不是 Thinking Machines 能不能构建有用的 AI 系统——公开证据显示它能。更难的问题是, 这套能力有多少会复利成可防御、可复用的产品优势,而不是停留在高度依赖人的交付经验。[CE021, CE022, CE023, CE024, CE025, CE026]

信任 / 质量 / 合规表
控制 / 认证 / 质量指标状态范围缺口
安全仓库 / 加密声明公司宣称数据平台和文档工作流无公开独立审计证据
云无关部署声明公司宣称客户、文档、位置、数据平台模块需要参考架构
治理和以人为本的推广公司宣称 / 伙伴佐证生成式 AI 和企业部署需要正式政策文件
隐私法适用性外部影响重大客户和文档用例需要实际控制映射
正式安全认证公开资料未确认全公司认证状态未知
公开可靠性 / 状态报告公开资料未确认全公司无公开状态页或 SLA 证据
模型性能基准部分宣称提到部分测试,但无广泛基准需要分模块指标

信任态势看起来合理,但买方仍会要求安全和可靠性证据;这些证据在网站上看不到。

[CE030, CE031, CE032]
FE004: 产品成熟度 / 能力图

主要产品族和技术维度的能力成熟度图。

[CE020, CE023, CE027, CE030, CE033, CE040]

5.4 图表

Chapter 06

06客户情况

6.1 客户基础、分层和买家地图

现有证据指向的客户基础,是大型组织,而不是小额交易型买家。OpenAI、Temus 及相关新闻来源把 Thinking Machines 放在金融服务、 零售、企业集团和公民组织中;更广的市场证据也表明,这些正是工作流集成、治理和数据清理成本足够高、足以支撑专业实施伙伴的账户。 实务上,可能的经济买家是资深转型、运营、技术或业务线负责人;日常用户是分析师、一线员工、网点员工和知识工作者;付款方则是企业或项目预算, 而不是团队级信用卡开支。 从地理看,客户故事仍然先是菲律宾,再逐步扩大到东南亚。这既符合公司的办公室足迹,也符合 Temus 组合。更重要的不是单纯地理, 而是账户质量:TM 在客户拥有复杂工作流、受监管数据或大规模运营的地方最相关。这提高了买家质量,即便公开客户标识密度仍然有限。 服务主导的 AI 公司会喜欢这种客户画像,因为成熟买家往往预算更大;一旦建立信任,后续用例也更多。 这也是更难打入的细分市场,因此每一个公开引用都比原始客户标识数更有价值。[CU001, CU002, CU003, CU004, CU012, CU013]

客户分层表
客群买方 / 用户 / 付费方使用场景规模收入 / 战略价值缺口
金融服务转型 / 运营 / 网点用户 / 企业预算数据平台、AI 工作流、CX AI战略价值高,可作为受监管行业标杆收入占比未知
零售 / 企业集团业务单元负责人 / 分析师 / 数字化预算客户智能和工作流 AI中高集团内有交叉销售潜力具名 logo 偏少
电信规划 / 分析 / 基础设施团队 / 企业预算地理空间分析和规划拉开 BFSI 之外的客户结构未披露具名客户
公共 / 市政部门项目负责人 / 分析师 / 机构预算地理空间、发展和公共服务 AI提供战略可信度和政策贴合度商业经济性不清楚
生成式 AI 赋能客户高管 / 知识工作者 / 转型预算培训、工作流部署、治理可撬动更广的企业项目具名账户名单偏薄
借助合作伙伴拓展的区域企业账户Temus / 合作伙伴关联发起人 / 企业预算生产级 AI 部署起步可能提高平均账户规模实际转化未公开

客群清晰,但公开披露在 logo 密度和收入拆分上仍然滞后。

[CU003, CU004, CU014, CU017, CU026, CU027]
客户增长 / 采用轨迹表
指标数值日期来源置信度含义缺失分母
服务客户数110+2026Temus / DealStreet / TNGlobal已有一定存量客户基础目前有多少仍活跃?
服务客户数150+2026OpenAI 合作伙伴资料基数可能比其他媒体报道显示的更大客户定义未知
已培训专业人员10,000+2026OpenAI / Temus / 相关新闻为销售管线和赋能铺开广泛触点多少转化为付费账户?
定制 GenAI 应用数十个2026案例目录显示 GenAI 活动已有一定规模每个应用对应客户数未知
EastWest 数字交易占总交易 51%2025EastWest 官方公告显示客户侧数字化成熟度不是 TM 专属指标
EastWest NPS 变化+8 points2025-2026EastWest 官方公告显示客户侧满意度提升不能单独归因于 TM

采用轨迹有碎片化证据可见,但分母质量偏弱。

[CU001, CU002, CU008, CU009, CU028]
FU001: 客户旅程图

TM 通常从一个工作流痛点切入,再靠交付信任和相邻用例加深关系。

旅程根据公开客户案例、合作伙伴叙述和 TM 的咨询式销售触点推断。

[CU003, CU013, CU014, CU023, CU034]

6.2 具名客户证明和采用深度

EastWest Bank 是客户证明中最清楚的公开锚点。Thinking Machines 自家官网包含 EastWest 关于采用 AI 并将其生产化的直接推荐语; EastWest 官方 2025-2026 发布也显示,该行继续投资数字开户、客服 AI 和咨询工作流。这些客户侧发布不能证明 Thinking Machines 导致了每一个披露结果,但它们确实把 TM 的工作放进一个真实环境:数字使用、NPS、投诉减少和 AI 赋能客户体验都是董事会相关话题, 而不是创新剧场。因此,即使归因必须谨慎,EastWest 仍是高质量引用。 EastWest 之外,证据从具名客户标识转向可信但更薄的证明。stories 名录提到数十个定制 GenAI 应用;UNICEF 相关材料披露了一个东南亚大型电信合同, 以及公共 / 公民领域对地理空间工作的需求;伙伴来源强调跨行业部署了数百个系统。这个模式与真实采用广度一致,但与整体客户数主张相比, 公开记录仍然缺少足够多可引用案例。 EastWest 证据也说明了本章该如何阅读客户成果。客户侧银行指标可以验证环境真实且在扩张,即便它们不能干净拆成可归因于 TM 的 ROI。 尽调仍然用得上这些信息,因为它把真实企业项目与幻灯片区分开来。[CU005, CU006, CU007, CU008, CU009, CU010]

具名客户佐证表
客户客群部署 / 使用场景生产部署 / 试点结果局限
EastWest BankBFSIAI 采用、生产化、数据平台和 CX 相关工作流偏生产部署公开背书,加上客户侧 AI 和数字化结果披露无法把该行所有结果都归因于 TM
UNICEF 相关气候 / 发展项目公共 / 市政 / 发展贫困、气候和公共数据的地理空间应用生产 / 应用研究外部引用的模型和可持续发展应用客户 / 合作伙伴经济性未公开
东南亚大型电信客户(未具名)电信卫星影像分析和地理空间规划偏生产部署UNICEF 称其为 TM 迄今最大合同之一客户名称未披露,结果指标有限

本表保留了具名公开佐证与未具名但可信的部署证据之间的差异。

[CU005, CU006, CU007, CU025, CU026, CU027]
留存 / 重复使用 / 满意度表
指标数值 / 空值客群置信度尽调问题
NRRnull全部提供按队列划分的美元口径净留存
GRRnull全部提供按账户队列划分的总留存
流失率null全部提供过去 24 个月的 logo 流失和收入流失
合同期null企业按服务线提供初始合同期限中位数
TM 客户 NPSnull全部提供系统化客户满意度指标
客户侧代理指标:EastWest NPS+8 pointsBFSI 标杆账户澄清 TM 是否实质推动了这一改善

公开证据无法直接支撑,多数留存指标仍为空值。

[CU009, CU019, CU020, CU032]
FU002: 采用 / 部署漏斗

公开证据从宽泛客户数说法收窄到具名留存证明时,口径明显变窄。

只有顶部和底部阶段直接来自公开披露;中间阶段是分析师代理值,用来显示证据收窄,而不是硬性数量。

[CU001, CU002, CU005, CU019, CU029]
FU003: 客户证明矩阵

不同账户和细分市场的客户证据质量差异明显。

[CU005, CU011, CU025, CU027, CU029, CU036]

6.3 留存、扩张和集中度风险

这是客户故事里披露最不足的部分。没有找到公开 NRR、GRR、流失、续约、合同期限或头部客户集中度数据。 也就是说,无法严谨量化 Thinking Machines 究竟是黏性多年的伙伴、后续工作偶发的项目制服务商,还是两者之间。 最好的间接证据来自生产型引用、客户侧持续数字化投资,以及伙伴对 TM 能随更大区域客户扩张的信心。 上行情景可信:通过数据或 AI 采用工作切入,在混乱企业工作流中赢得信任,再扩展到相邻用例或地区。下行风险同样可信: 少数强引用可能掩盖集中度,或掩盖公司依赖少数旗舰账户的事实。因此,客户结论对买家质量和具名证明质量为正面, 但对留存持久性和集中度暴露明确不完整。 换句话说,证据支持一组高质量引用,但不支持一个完全可承销的客户模型。下一步尽调不是看更多营销文案, 而是拿到账户级数据:活跃客户、扩张历史、集中度和续约行为。[CU014, CU015, CU016, CU017, CU018, CU019]

扩张与集中度风险表
扩张驱动因素集中度风险影响尽调路径
先落地数据平台,再扩展到 AI 工作流头部客户依赖度未知可能提高账户耐久性,也可能掩盖集中度索要前 10 大客户收入占比和扩张历史
借 Temus 伙伴渠道触达区域客户依赖伙伴来源账户可能加速增长,但削弱控制力索要按渠道拆分的来源管线结构和利润率
银行标杆质量过度依赖 EastWest 作为旗舰公开佐证公开叙事可能过度押注单一账户索要按行业划分的更多具名客户背书
培训和赋能项目从培训转化为企业收入的路径未知可能是有用漏斗,也可能是低质量管线索要从赋能到付费交付的转化指标
电信和公共部门多元化未具名账户掩盖真实规模和续约质量多元化可能被高估在 NDA 下索要具名客户名单和合同状态
区域办公室布局缺少账户级可见证据的扩张成本如果需求跟不上,可能稀释聚焦度索要按国家拆分的客户和收入组合

集中度大概率不是灾难性问题,但公开证据太薄,无法排除。

[CU014, CU017, CU018, CU033, CU034, CU035]
客户披露限制表
缺失披露重要性最佳尽调路径
前 10 大客户占比验证集中度风险要求提供按客户和行业拆分的收入集中度
续约 / 流失指标验证耐久性要求提供按队列拆分的留存收入和流失
合同条款验证收入可见度和采购风险审阅代表性 MSA / SOW 文件
按行业划分的具名客户背书验证 EastWest 之外的可复制性索要覆盖 BFSI、电信、零售、市政的客户访谈名单
国家收入组合验证东南亚扩张真实性索要已签收入和管线的地域拆分
用户 / 席位 / 工作流渗透验证真实采用深度索要标杆部署的账户级推广指标

客户质量看起来真实,但支撑投资判断的关键数据仍大多未公开。

[CU015, CU018, CU019, CU023, CU035, CU036]
FU004: 留存 / 重复合作队列

缺少直接留存指标,耐久性只能从生产案例、客户持续投入和伙伴信心来推断。

[CU019, CU020, CU030, CU031, CU033, CU036]

6.4 图表

Chapter 07

07风险

7.1 监管、法律和市场形成风险

监管风险具有实质性,因为 Thinking Machines 处理客户数据、文档,以及与金融服务相邻的工作流。菲律宾隐私制度已经适用于处理个人数据的 AI 系统, BSP 也开始阐明银行 AI 的自愿监管预期。这些框架支持负责任采用,但也提高了治理、文档、人类监督和偏见控制的门槛。 公司最佳客户常常是受监管企业,风险不是「AI 被禁止」,而是「部署变慢、文档更多、成本更高」。 市场形成风险仍然真实。Swarm 和 Trade.gov 展示的是一个兴趣高但制度化不均的市场:实验普遍、全公司使用有限, 大量企业卡在概念验证阶段。从一个角度看,这对实施伙伴商业有利;但这也意味着销售周期、采购摩擦和预算转化,可能比乐观 AI 叙事暗示的更久保持波动。 因此,监管节奏风险对投资者尤其重要。一家公司可能因为能处理更难的治理环境而胜出,但如果文档负担、审批关口或客户谨慎上升得比交易规模更快, 仍会受损。银行和政府买家在治理标准变化后也往往行动缓慢,所以即便位置不错的供应商,也可能看到销售管线转化在几个季度里明显拉长。[CR001, CR002, CR003, CR004, CR005, CR006]

监管 / 法律风险登记表
规则 / 案件司法辖区状态发生概率严重性缓释措施剩余风险暴露尽调路径
NPC AI / 隐私指引菲律宾已生效指引治理前置的部署模型可能有帮助索要隐私保护内建控制和事件历史
BSP AI 治理预期菲律宾 / BFSI自愿但带监管监督属性中高BFSI 经验和治理定位中高索要银行部署控制材料
NAISR 2.0 / 政策演进菲律宾支持性强但仍在演进政策契合度和本地根基跟踪 AI 法案、DTI 指引和行业规则
诉讼 / 执法可见度菲律宾 / 东南亚未发现公开事件低中已抓取材料中无已知公开行动Unknown开展超出网络来源的法律和监管核查

行按严重性排序,但开放网络上的部分法律可见度有限,因此仍不完整。

[CR006, CR007, CR008, CR009, CR034, CR035]
FR001: 风险热力图

剩余严重度最高的风险是伙伴依赖、人才 / 交付规模,以及披露不足带来的客户不透明。

[CR014, CR016, CR027, CR032, CR035, CR042]

7.2 运营、依赖和交付风险

从运营看,Thinking Machines 架在几类外部依赖上。公有云基础设施、OpenAI 相关模型生态、客户数据访问和地理空间数据来源, 都会影响交付。其中一些对现代 AI 服务公司来说很普通,但仍会形成实际故障模式:云成本上升、供应商路线图变化、API 依赖、 数据权利争议或客户数据可用性延迟。地理空间工具链又多一层复杂性,因为一些公开开发者产物明显是社区型、尽力而为; 这对技术文化很好,但不等同于企业支持保证。 菲律宾运营条件又加上一层执行风险。基础设施来源强调电力成本、供电可靠性,以及菲律宾相对 Singapore 或 Thailand 在数据中心规模上的不成熟。 这些约束与人才稀缺叠加,会抬高交付成本、拖慢模型迭代,并让区域支持更困难,除非 Temus 组合实质扩展团队深度和跨境执行能力。 Temus 关系部分抵消了这些风险,因为它扩大了交付能力,但没有消除风险。甚至可以说,更大的区域雄心提高了依赖故障的成本, 因为更多客户、地区和交付团队可能一次性受影响。这一点重要,因为 Thinking Machines 销售的是复杂数据和 AI 成果; 当支持、人员或集成假设破裂时,补救会吃掉高价值专家时间,而不是创造增量收入。[CR010, CR011, CR012, CR013, CR022, CR023]

运营 / 质量 / 安全风险登记表
失效模式发生概率严重性缓释成熟度剩余风险暴露未解决缺口
电力成本 / 可靠性拖慢 AI 交付低中需要实际基础设施和故障切换架构
云 / API 依赖改变经济性或路线图需要多供应商备用方案和利润率敏感性
安全或隐私控制对受监管买方不足需要独立认证和审计
试点到生产的摩擦拖慢转化中高需要管线阶段数据和转化率
开源地理空间工具缺少企业级支持需要说明 OSS 如何映射到受支持的内部工具

运营风险不太像硬件故障,更像依赖承压时的交付可靠性问题。

[CR002, CR003, CR010, CR011, CR022, CR023]
合作伙伴 / 依赖风险登记表
依赖交易对手角色集中度失败情景严重性缓释措施剩余风险暴露
基础模型 / AI 生态OpenAI 及同业生成式 AI 赋能与合作伙伴背书合作伙伴地位削弱,或供应商经济条款变化保持更宽的应用 AI 技术栈与云灵活性
战略规模化合作伙伴Temus区域触达与交付梯队厚度中高整合摩擦,或战略优先级漂移保持品牌、领导层和客户连续性中高
公有云提供商主要云厂商托管与分析底座成本飙升、架构约束、平台集中云无关交付主张中高
客户数据访问企业客户数据与文档输入数据质量问题或访问延迟拖住部署咨询式范围界定与治理
地理空间 / 卫星数据合作伙伴外部数据供应商位置智能输入数据权利或可用性变化分散数据来源和合同

上述均属 AI 服务公司的常规依赖,但在年轻的区域平台里仍可能迅速叠加。

[CR012, CR013, CR024, CR032, CR033]
FR002: 风险传导图

多项独立风险最终会传导到收入质量、利润率、客户耐久性和估值。

[CR003, CR011, CR016, CR017, CR027, CR037]
FR003: 依赖图

TM 的企业交付能否成功,取决于少数伙伴、数据源和监管方。

[CR012, CR013, CR024, CR032, CR033]

7.3 人员、财务模型和投资逻辑破裂风险

在服务占比高的业务里,人员风险和披露风险绑得很紧。Thinking Machines 明显带有创始人烙印,公开规模信号又不够一致, 外部很难判断更宽的管理层梯队到底有多深。这一点关键,因为服务型公司很少死于模型失效;更常见的是利用率裂开、 招聘卡住、客户集中度反噬,或者创始人的关系资本难以复制。 因此,财务模型风险本质上是信息不透明风险。外界看不到收入质量、毛利率、留存、集中度或现金跑道。投资人只能在看不到底层经营仪表盘的情况下, 信任可靠伙伴、严肃客户和一套有吸引力的战略叙事。正确反应不是直接否定公司,而是把否决条件划清楚: 合作伙伴身份弱化、客户集中度意外暴露、BFSI 监管摩擦,或无法以可接受毛利率配齐区域交付团队。 实际投资教训是,不能只靠叙事给这家公司做投资判断。投资测算必须围绕集中度、合作伙伴经济性、人员韧性和监管运营证据划出明确红线。 披露不足时,下行情景应假设招聘放慢、回款不均、跨境扩张延后,而不是近期势头顺滑延续。[CR014, CR015, CR016, CR017, CR018, CR019]

人才 / 执行风险登记表
角色 / 职能依赖或缺口发生可能性严重度缓释措施尽调路径
创始人 / 外部信任Stephanie Sy 中心化Temus 梯队与更广泛领导团队梳理已委派的客户归属和第二梯队负责人
高级交付人才菲律宾 AI 人才稀缺培训文化与区域招聘要求提供离职率、未填岗位和利用率
区域执行领导力东南亚扩张,覆盖 Manila / Singapore / Bangkok中高Temus 运营基础设施审阅组织架构图和区域 P&L 归属
规模可见度员工数信号相互冲突多个公开资料指向增长要求按职能和地点提供实际员工数
招聘流程质量靠文化驱动招聘,规模化后可能不够学习导向文化审阅招聘漏斗和高级人才成功率

创始人主导的服务公司里,一旦客户负载跑得比交付梯队更快,执行风险会迅速上升。

[CR001, CR027, CR028, CR029, CR030]
缓释措施与否决标准表
风险可监测触发项阈值 / 事件行动含义
伙伴依赖OpenAI / Temus 状态变化关系丧失、降级,或经济条款实质恶化重新评估 GTM 和差异化
客户集中度头部客户占比意外偏高一两个客户贡献超额收入下修估值,并要求需求多元化计划
监管收紧BFSI 或隐私执法阻断用例主要客户部署放缓或冻结除非控制措施得到验证,否则降低确信度
人才 / 交付压力利用率飙升、离职率上升、招聘滞后交付梯队厚度跟不上销售管线假设增长和利润率更低
运营依赖云或数据源依赖实质扰乱交付交付周期延误,或利润率被压缩要求提供架构兜底方案证据
财务不透明管理层无法拿出队列、利润率、现金和集中度数据核心指标没有干净的数据室不按溢价倍数承销

否决标准应持续监测,不只在融资节点检查。

[CR037, CR038, CR039, CR040, CR041, CR042]

7.4 图表

Chapter 08

08估值

8.1 估值投资逻辑、反向逻辑与价格纪律

在战略价值层面,Thinking Machines 比较像有价值的资产;但到了投资测算层面,仍不够清晰。第 1 至第 7 章显示, 这是一家真实的东南亚 AI / 数据服务公司,有可信创始人、OpenAI 合作伙伴身份、具名企业和发展部门客户背书, 以及新的 Temus 战略资本信号。这些都是重要正面因素,因为它们降低了公司只是 PPT 式 AI 故事的概率。 但它们没有解决核心估值问题:公开证据仍看不到收入、毛利率、留存、定价、现金或股权结构条款。 估值缺口重要,因为在用户提示里,Thinking Machines 不是被当作一家沉闷的外包店估值;它被框成一家带有前沿融资叙事的独角兽。 本轮检索没有任何材料能支撑 Thinking Machines Data Science, Inc. 的这种定位。可支撑的解读更窄: 一家植根菲律宾、受尊重的应用 AI 公司,带有合作伙伴可选性,但经营披露远低于投资人证明数十亿美元价格所需的水平。 因此,反向逻辑不是“公司不好”,而是“公司不错,经济性不清,叙事可能被拉得过满”。价格纪律应把 TM 当作服务占比较高、 具备部分产品上行空间的应用 AI 平台候选,而不是已经被证明的软件复利机器或稀缺模型实验室资产。按这种框架,上行仍有空间, 但建议也高度依赖价格和证据。[CV001, CV002, CV003, CV004, CV005, CV006]

投资建议摘要表
维度评估置信度决策含义
整体建议继续研究在收入、留存和条款清晰前,不要按溢价入场估值承销
风险评级尽管公司质地看起来真实,但不透明和执行风险占主导
估值立场公司可能有吸引力;但当前溢价价格缺乏支撑战略可选性存在,但高于保守私募市场区间的价格支撑较弱
手头最佳证据Temus 战略投资、OpenAI 合作伙伴身份、具名客户验证这些证据能验证相关性,不能验证具体估值锚
主要阻碍收入 / 利润率 / 股权结构表均缺乏公开透明度缺少这些输入,就必须用情景纪律替代确信定价

建议表总结按证据校正后的判断,不是市场报价。

[CV002, CV003, CV005, CV010, CV018, CV041]
投资逻辑 / 反向逻辑表
论点改变判断的证据
投资逻辑:Thinking Machines 是真实的区域应用 AI 业务,不是纸面公司多个公开客户或经审计的增长数据会显著强化这一判断
投资逻辑:OpenAI 和 Temus 关系显示生态可信度合作伙伴身份丧失,或关系并非商业性质,会迅速削弱该信号
投资逻辑:AI / 数据平台专长可以支撑相对普通外包的溢价若证据显示工作主要是定制服务、复用很少,溢价会被压缩
反向逻辑:公开披露太薄,无法形成承销确信管理层披露收入、利润率和留存,可直接回应这一点
反向逻辑:服务收入占比限制软件式倍数潜力若能证明平台收入可经常性、可附加销售,或可按使用量变现,投资论证会改善
反向逻辑:公开记录不支持独角兽叙事需要披露定价轮或独立估值标记,才足以推翻这一点

论点由各章证据综合而来,并说明哪些证据可以证伪任一方向。

[CV001, CV004, CV006, CV007, CV009, CV017]
FV001: 建议逻辑

真实的战略质量撞上薄弱的价格支撑和不完整估值输入,形成这条建议。

仅为逻辑流;它总结投资判断链条,不是数值模型。

[CV001, CV002, CV003, CV004, CV005, CV010]

8.2 可比公司组与融资背景

最有用的公开可比视角是混合口径:AI / 数据实施公司,以及股权价值仍高度依赖服务收入质量的数字化转型平台。 2026 年 8 月的公开市场读数相当冷。按抓取到的市值和 TTM 收入页面,Endava 约为 0.15x 收入,Globant 约为 0.65x, EPAM 约为 0.90x,Genpact 约为 1.08x,Accenture 约为 1.47x。换句话说,即便是披露远好得多的可信全球运营商, 2026 年也不会自动拿到软件式倍数。 Thinking Machines 理应比通用 IT 服务拿到一定溢价,因为类别更窄、增长更快:企业 AI 采用、受治理的数据平台、地理空间分析都比标准人员外包更有差异化。 但它也应低于软件或前沿 AI 叙事,因为检索记录仍显示业务由服务牵引、锚定合作伙伴,且经常性收入不透明。 Temus 交易强化了上述结论,而不是消除它。这是强质量信号,也证明一家严肃区域运营商看到了战略价值。但公开证据仍把它描述为战略投资, 而不是披露经济条款、价格透明的基准轮融资。因此,投资人可以把这笔交易作为相关性验证,却不能把它当作任何具体私募估值标记合理的证明。[CV011, CV012, CV013, CV014, CV015, CV016]

可比估值表
可比公司2026 年 8 月市值TTM 收入隐含市值 / 收入相关性局限
Endava$0.15B$0.98B~0.15x显示服务驱动的数字工程在增长放缓时,估值会多快被下修数字工程业务更宽;不是 AI 专家型公司
Globant$1.60B$2.45B~0.65x较高端转型与产品工程可比公司,带有一定 AI 敞口比 TM 更大、更全球化,也更分散
EPAM Systems$5.02B$5.55B~0.90x有规模的技术服务标杆,工程声誉强披露成熟度和交付厚度高得多
Genpact$5.59B$5.16B~1.08x可作为分析 / 运营转型和企业工作流定位参照BPO / 运营出身不同于 TM 的 AI 叙事
Accenture$107.42B$73.10B~1.47x受监管行业可信企业执行的大型市值上限可比公司规模过大且业务过分散,不能作为干净的私营初创可比公司

上市公司数值来自已抓取的 2026 年 8 月市值和收入页面;它们是可比参照,不是直接同业。

[CV011, CV012, CV013, CV014, CV015, CV016]
FV002: 估值敏感性

示意性估值结果会随假设收入规模、以及投资者愿意为 AI 服务业务支付的倍数大幅摆动。

由于公开收入未披露,收入代理值仅作示意。目的在于说明:高额标称估值站得住之前,需要补多少证据。

[CV016, CV017, CV023, CV027, CV028, CV029]

8.3 乐观、基准与悲观情景

由于披露的估值输入很弱,正确做法不是追求虚假精确,而是守住有边界的情景纪律。悲观情景假设 TM 仍主要是专家服务和实施业务, 区域转化更慢,并按公开市场服务倍数估值。基准情景假设公司能以优于普通咨询的经济性,证明 AI 交付可以重复, 并因区域稀缺性获得适度溢价。乐观情景假设公司把 Temus 和 OpenAI 光环加上公开客户证明,转化成更产品化、 更可重复的区域平台故事,附着率更强、资本效率更好。 这些情景最好被看作入场纪律区间,而不是断言管理层今天会接受这些精确标记。公开证据无法提供完整 DCF 或软件式队列模型所需的输入。 但仍能得出一个重要判断:保守假设下,TM 可支撑估值大概率以数千万至低数亿美元计;只有在高度有利的战略溢价情景下, 才会延伸到低数十亿美元。 估值区间很宽,但结论仍清晰。公司也许足够好,未来能长成大得多的结果;但公开证据还不足以支撑当下的 $10B 式叙事。 正确的投资姿态是,在支付任何前沿风格价格前,先要求公司按里程碑拿出证据。[CV024, CV025, CV026, CV027, CV028, CV029]

乐观 / 基准 / 悲观情景表
情景核心假设估值逻辑概率信号关键风险
悲观交付仍以服务为主;区域扩张较慢;经常性软件证据有限按上市服务公司式纪律,以约 1x-2x 收入等价经济性推算,可支撑约 $40m-$150m如果客户集中度、定价透明度或伙伴杠杆不及预期,该情景就有意义利润率压缩、复购疲软,或招聘停滞
基准TM 证明 AI 交付可重复,并在 SEA 企业 AI 中获得适度战略溢价如果商业质量优于普通咨询,按混合 3x-4x 高端服务视角,可支撑约 $180m-$500m从当前证据看最可能:质地可见,经济性不可见披露始终追不上;溢价向上市可比公司水平收缩
乐观Temus + OpenAI + 客户证明叠加,形成区域平台杠杆和强可重复变现如果经常性经济性和区域规模得到验证,高溢价战略情景可支撑约 $600m-$1.3b需要里程碑兑现,而不只是叙事延续除非经济性证明远好于上市可比公司,否则仍远低于缺乏支撑的 $10B 式叙事

情景区间是示意性承销带,因为公开财务输入不完整。

[CV024, CV025, CV026, CV027, CV028, CV029]
FV003: 估值 / 回报区间

情景区间刻意放宽;公司质量看起来有分量,但估值输入质量偏低。

区间是分析师指定的投资判断范围,不是市场报价。未来轮次稀释未建模,因为股权结构表条款未披露。

[CV027, CV028, CV029, CV030]

8.4 退出准备度、尽调要求与最终结论

基于本轮可得证据,Thinking Machines 还没到公开市场就绪状态。没有公开经审计投资人材料,没有披露收入桥,也看不到股权结构, 更无法由外部确认利用率、集中度或经常性软件占比。价值并非不存在;只是还不能清晰转化成可投资的公开市场式测算。 若 TM 证明自己已搭起有防御力的区域数据 / AI 业务版图,战略买家或后期私募投资人仍可能支付明显高于公开可比公司的价格。 今天最现实的正向退出更可能是战略退出或财务投资人支持的退出,而不是 IPO:区域 IT 服务整合者、希望获得东南亚企业 AI 产能的咨询 / 云合作伙伴, 或想要本地市场入口的数据平台收购方,都比短期上市更合理。对外部资本而言,门槛问题不是公司是否亮眼,而是下一美元相对于不透明和服务式利润率压缩风险, 能否买到足够的证据调整后上行。 结论:继续研究。置信度为中,因为公司质量证据真实存在,但估值支撑不足。风险评级为高,因为缺失信息正好集中在入场价格纪律最关键的位置: 收入质量、留存、利润率和融资条款。只有这些缺口被披露并经得起审视,建议才可能改变。[CV034, CV035, CV036, CV037, CV038, CV039]

投资逻辑破裂与否决触发项表
触发项阈值对投资逻辑的传导行动含义
收入质量不及预期尽调后仍没有经常性或可重复收入证据溢价高于服务公司的论点坍塌按保守服务公司区间重新定价,或停止
客户集中度高头部客户主导收入,但没有持久续约让战略叙事变脆弱要求集中度折价,或停止
伙伴光环消退OpenAI / Temus 关联缺少商业证明时变弱降低可信度和分销可选性下调护城河假设
交付经济性不及预期利用率、利润率或回款呈现项目型风险特征基准估值区间明显压缩按项目业务处理,而不是平台候选
受监管行业采用放缓BFSI / 政府 AI 部署停滞或周期大幅拉长拉长规模化时间,并提高现金转化风险只保留观察清单 / 跟踪姿态

触发项是后续尽调或重新定价可监测的红旗,不是预测。

[CV031, CV038, CV039, CV040, CV041]
最终尽调要求表
主题缺失证据重要性负责人 / 尽调路径
收入质量分部收入、重复收入和前 10 大客户结构需要判断 TM 是否配得上战略溢价管理层材料包和客户队列审阅
利润率 / 利用率毛利率、交付利用率和服务 / 软件组合区分可扩展平台经济性与咨询经济性管理账目和交付仪表盘
留存 / 集中度续约率、扩张历史和合同期限服务主导业务的核心下行变量客户队列分析和合同样本
股权结构表 / 条款股份类别、优先权、稀释和 Temus 条款没有条款结构,无法评估入场回报账法律尽调和融资数据室
伙伴经济性OpenAI、Temus 和云关系的商业机制品牌光环不等于变现杠杆伙伴协议摘要和来源管线分析

每项尽调要求都是提升估值置信度的阻断项,不是锦上添花。

[CV003, CV018, CV019, CV034, CV037, CV042]
FV004: 投资 KPI

IC 式评分显示,公司有真实战略吸引力,但估值支撑不完整。

KPI 评分只综合章节证据,不能替代管理层尽调。

[CV001, CV003, CV008, CV016, CV018, CV034]

免责声明

本报告是自动生成的尽调摘要,仅基于截至 12 August 2026 检索到的公开信息。它不构成投资建议,也不构成买入或卖出任何证券的要约。Thinking Machines Data Science, Inc. 是私营公司,并未公开披露高置信度估值所需的收入、利润率、现金流、留存、股权结构表或融资条款信息。各章节中的任何估值区间都是经证据调整的情景区间,不是市场报价或公平性意见;作出任何投资决定前,应直接用公司一手材料验证。

证据索引

结论
编号陈述可信度来源
CO001 Thinking Machines describes itself as an AI and data transformation company operating across Southeast Asia. SO001, SO002, SO012
CO002 The company says it helps enterprises design, deploy, and adopt AI systems that improve how work gets done. SO001, SO003, SO011, SO012
CO003 Temus, Context, and TNGlobal each describe Thinking Machines as founded in Manila in 2015 by Stephanie Sy. SO011, SO013, SO014
CO004 Current public materials say the company operates offices in Manila, Singapore, and Bangkok. SO001, SO011, SO012, SO013
CO005 The homepage and OpenAI partner page say Thinking Machines has trained more than 10,000 professionals in AI. SO001, SO011, SO012
CO006 The homepage and OpenAI partner page say the company has worked with more than 150 global clients. SO001, SO012
CO007 Temus-linked July 2026 coverage says Thinking Machines has served more than 110 clients, which conflicts with the 150-plus claim on the company website. SO011, SO013, SO014, SO015, SO016, SO017
CO008 The homepage publishes an 85 Net Promoter Score as a current company metric. SO001
CO009 Stephanie Sy remains the publicly named founder and chief executive officer of Thinking Machines in 2026. SO001, SO011, SO013, SO021
CO010 Public profiles consistently describe Stephanie Sy as a Stanford alumna and former Google employee. SO001, SO020, SO021, SO023
CO011 Harvard Business Publishing and Hustleshare say Sy returned from Silicon Valley to build in the Philippines and closer to family. SO018, SO020
CO012 By 2018, independent profiles already positioned Thinking Machines as one of the few Philippine data-science consultancies serving business and public-sector use cases. SO019, SO021
CO013 Forbes in 2018 said Thinking Machines then had offices in Manila and San Francisco. SO019
CO014 OpenAI and TechEDT identify Thinking Machines as OpenAI’s first official Services Partner in Asia Pacific. SO012, SO026
CO015 The homepage and Temus materials say Thinking Machines was later recognized as an OpenAI Advanced Partner. SO001, SO011, SO012
CO016 Temus announced a strategic investment in Thinking Machines on 2026-07-30. SO011, SO013, SO014, SO015
CO017 None of the Temus, TNGlobal, or Context disclosures retrieved in this run published the investment amount or a valuation for the 2026 deal. SO011, SO013, SO014
CO018 After the deal, the company said it would operate as “Thinking Machines Data Science, a Temus entity.” SO011, SO014, SO015
CO019 The Temus deal made Stephanie Sy a Managing Director for Applied AI & Data at Temus while she continued as CEO of Thinking Machines. SO011, SO013, SO014, SO015
CO020 Temus said the Thinking Machines brand, leadership, and client engagements would remain unchanged after the investment. SO011, SO015, SO016
CO021 Temus positioned the combination as a way to move enterprises from AI pilots into production-grade deployment across Southeast Asia. SO011, SO013, SO015, SO016, SO017
CO022 UNICEF Venture Fund’s graduate page lists $449,598 invested into Thinking Machines and a funding status of active growth period. SO022
CO023 UNICEF Venture Fund’s graduate page says Thinking Machines was founded in 2016, conflicting with multiple 2026 sources that say 2015. SO022
CO024 UNICEF-backed profiles show Thinking Machines initially focused on satellite-imagery and geospatial AI for wealth mapping, infrastructure planning, and climate-development use cases. SO022, SO023, SO024
CO025 The UNICEF Venture Fund graduate page says the team open-sourced GeoMancer and Tiffany during the programme. SO022
CO026 Current official product pages emphasize enterprise AI transformation, data platforms, document intelligence, customer intelligence, and generative AI rather than a pure geospatial niche. SO001, SO003, SO004, SO005, SO006, SO007, SO008
CO027 The data-platforms page says Thinking Machines builds scalable cloud data and analytics platforms and real-time dashboards on a single source of truth. SO004
CO028 The location-intelligence page says Thinking Machines maintains an exhaustive geospatial data catalog and data partnerships. SO007
CO029 The document-intelligence page says Thinking Machines structures documents into public-cloud data platforms using open-standard technologies. SO006
CO030 The customer-intelligence page says Thinking Machines builds real-time customer profiles and segments from transactional behavior. SO005
CO031 The about page says the company works alongside client teams to build lasting skills and deliver early wins designed to scale. SO002, SO008
CO032 The generative-AI page says AI projects often fail when leadership lacks a clear roadmap, so Thinking Machines begins engagements with training and strategic alignment. SO008
CO033 Trade.gov says only 14.9 percent of Philippine firms currently use AI technologies, making adoption uneven in the company’s home market. SO025
CO034 Trade.gov says the Philippine policy environment is supportive but evolving, with no standalone comprehensive AI law and several AI-related bills still pending in Congress. SO025
CO035 It is reasonable to infer that Thinking Machines’ emphasis on governance, training, and workflow integration is partly a response to uneven adoption, compliance complexity, and data-quality constraints in its home market. SO002, SO008, SO025
CO036 The OpenAI partner page lists the countries served as the Philippines, Singapore, and Thailand. SO012
CO037 Forbes’ 2018 profile said the firm’s clients already included corporates, government agencies, NGOs, and startups. SO019
CO038 Makati Business Club’s 2022 event writeup says Thinking Machines helped EastWest Bank automate reconciliation across 400-plus ATMs and 2 million monthly transactions. SO027
CO039 The homepage includes an EastWest Bank testimonial describing Thinking Machines as a key partner in adopting and productionalizing AI. SO001
CO040 UNICEF Innovation’s profile said Sy’s team had grown to over 80 people and launched a Sustainability Team in 2021. SO023
CO041 LeadIQ classifies the company in the 51-200 employee band and as Singapore-based, which conflicts with official Manila-founded and Philippines-headquartered descriptions. SO011, SO014, SO028
CO042 Current retrieved public sources still do not disclose exact 2026 headcount, board composition, revenue, or a post-money valuation. SO011, SO013, SO014, SO028
CO043 Context framed the Temus transaction as reinforcing investor confidence in Philippine-built AI capability. SO014
CO044 Business Daily Media and Media OutReach say the firm has co-developed hundreds of AI systems over its first decade. SO011, SO015, SO016, SO017
CO045 Harvard, Ignition, and Hustleshare all present storytelling, curiosity, and willingness to learn as central parts of Stephanie Sy’s leadership style. SO018, SO020, SO021
CO046 The contact and homepage materials present the company as reachable through a Manila-led regional operation rather than through a separate standalone Singapore headquarters site. SO001, SO010
CM001 The Philippine AI market was approximately US$772 million in 2024 and is projected to reach about US$3.49 billion by 2030, implying roughly 28.6 percent CAGR. SM001, SM002
CM002 Trade.gov characterizes the Philippine AI opportunity as early-stage and meaningful rather than already saturated. SM001
CM003 Only 14.9 percent of Philippine firms currently use AI technologies according to the U.S. Commercial Service market note. SM001
CM004 The Philippine IT-BPM sector ended 2025 at about US$40 billion in export revenues with a workforce of roughly 1.9 million. SM001
CM005 An industry survey cited by Trade.gov says 67 percent of respondent IT-BPM firms had already incorporated AI tools into operations. SM001, SM002
CM006 Swarm’s 2026 survey found that 92 percent of Philippine organizations had used AI in some capacity. SM006
CM007 Swarm’s same survey found that 65 percent of organizations remained at the proof-of-concept stage. SM006
CM008 Swarm reports that 61 percent of organizations have CEOs, CTOs, or equivalent senior leaders directly leading AI initiatives. SM006
CM009 Swarm says internal automation, content creation, and data analysis are the most deployed AI workflow categories at 65 percent, 64 percent, and 60 percent respectively. SM006
CM010 Swarm identifies talent scarcity at 57 percent and security/privacy concerns at 40 percent as the top two barriers to scaling AI. SM006
CM011 Swarm says 83 percent of organizations report employees using ChatGPT, creating shadow-AI exposure when controls lag adoption. SM006
CM012 Nearly half of organizations in Swarm’s sample, 47 percent, report they are already in the AI application development stage. SM006
CM013 NAISR 2.0 launched in July 2024 and is organized around two pillars: Innovation and Implementation. SM003, SM007
CM014 OECD’s summary says NAISR 2.0 operationalizes four strategic dimensions: research and development, digitisation and infrastructure, workforce development, and AI governance and ethics. SM003
CM015 The DTI-linked CAIR initiative is intended to become the country’s first AI hub for socio-economic R&D and practical applications. SM007
CM016 OECD’s NAISR profile says the Philippines has more than 800,000 college graduates annually, over 200,000 from STEM fields, and more than 1,300 IT-BPM companies generating US$35.5 billion in annual revenue. SM003
CM017 OECD says NAISR 2.0 explicitly names limited local use cases for SMEs, scarce computational and human resources, insufficient enterprise data-strategy capacity, and legal uncertainty as adoption barriers. SM003
CM018 Elegal’s summary says DTI wants Philippine gross R&D expenditure to rise from about 0.30 percent of GDP toward UNESCO’s 1 percent benchmark. SM007
CM019 The NPC’s AI advisory states that the Data Privacy Act and its implementing rules apply to AI-system development, training, testing, and deployment whenever personal data is processed. SM004
CM020 The NPC advisory requires transparency, accountability, lawful basis, data minimization, governance mechanisms, and meaningful human intervention for AI systems processing personal data. SM004
CM021 The NPC advisory explicitly warns against AI washing and instructs organizations to monitor and mitigate systemic, human, and statistical bias. SM004
CM022 BSP Memorandum M-2026-031 provides voluntary supervisory expectations for AI governance in financial institutions using a proportionality principle. SM005, SM013
CM023 The BSP framework groups its guidance under the STARS principles: Sustainability, Transparency, Accountability, Responsibility, and Security. SM013
CM024 For Thinking Machines, the relevant market is enterprise AI and data-transformation services that combine governance, training, deployment, integration, and data-platform work rather than frontier-model R&D or hardware manufacturing. SM016, SM017, SM018, SM019, SM020
CM025 Included spend in that market should cover enterprise AI strategy, data-platform modernization, workflow integration, governance, change management, and deployment support in regulated and data-rich sectors. SM001, SM004, SM016, SM017, SM018
CM026 Excluded spend should include foundation-model training, hyperscaler capex, custom semiconductor manufacturing, generic consumer AI apps, and unrelated labor-arbitrage outsourcing. SM001, SM008, SM009, SM011
CM027 The most relevant Philippine buyer segments for implementation-led AI services are IT-BPM/BPO, BFSI, retail and conglomerates, public sector, telecommunications/logistics, and healthcare. SM001, SM002, SM006, SM010, SM016, SM020
CM028 Within those segments, the economic buyer is usually a C-suite or business-unit sponsor, while IT, security, data, and operations teams are the core users and gatekeepers. SM004, SM006, SM013, SM018
CM029 Demand drivers include BPO cost pressure, productivity gains, cloud and data-center expansion, national AI-policy support, and competition to move from pilots into production. SM001, SM006, SM009, SM010, SM013
CM030 Digital in Asia says Southeast Asia could see about US$30 billion of data-center investment by 2030 and about 20 percent annual demand growth through 2028. SM009
CM031 Source of Asia says Southeast Asia’s AI sector was worth more than US$4 billion in 2024 and could grow more than fourfold by 2033. SM008
CM032 EDB’s regional report says nearly half of Southeast Asian companies have moved beyond AI pilots, placing the region slightly ahead of the global average. SM011
CM033 EDB also reports that Singapore is a regional leader in scaled adoption while much of the region is still turning experimentation into durable impact. SM011
CM034 Digital in Asia and Nexdigm both place the Philippines earlier in the data-center development cycle than Singapore or Thailand. SM009, SM010
CM035 Singapore functions as the regional command node while Thailand is rapidly scaling hyperscale capacity, making the Philippines comparatively more services-led than infrastructure-led in the near term. SM009, SM011, SM014
CM036 Nexdigm says Philippine AI-infrastructure demand is rising in banking, e-commerce, healthcare, and BPO and increasingly requires cloud, storage, high-performance compute, and connectivity upgrades. SM010
CM037 Nexdigm identifies high electricity cost and power-reliability constraints as real limits on hyperscale expansion in the Philippines. SM010
CM038 Ajentik cites e-Conomy SEA 2025 for more than US$2.3 billion invested into over 680 ASEAN AI startups in the prior year, with Singapore as the funding center. SM014
CM039 The serviceable market for Thinking Machines is narrower than the headline AI-market figures because the company sells deployment-intensive services rather than generalized model access. SM006, SM016, SM017, SM018, SM019, SM020
CM040 No retrieved public source directly publishes a numeric SAM or SOM for implementation-led AI services in the Philippines. SM001, SM002, SM006, SM010
CM041 Swarm reports that most Philippine organizations are consuming AI tools rather than building proprietary model stacks, with only 12 percent using ML frameworks and 10 percent using NVIDIA CUDA. SM006
CM042 That tooling profile favors implementation partners who can govern, integrate, and productionize vendor models faster than pure model builders can sell bespoke infrastructure. SM006, SM018, SM019, SM020
CM043 Trade.gov’s strategic-technologies guide identifies AI priority sectors including healthcare, education, agriculture, logistics, and digital services, broadening the addressable vertical map beyond BPO alone. SM002
CM044 Gorriceta says NAISR 2.0 gives extra weight to labor reskilling, ethics, and legal adaptation as the Philippines absorbs AI into BPO and IT-intensive sectors. SM012
CP001 Thinking Machines positions itself across the full path from data foundations to advanced generative AI rather than as a single-product AI tool vendor. SP001, SP002, SP025
CP002 Thinking Machines is presented by OpenAI and its own website as the first APAC Services Partner and an OpenAI Advanced Partner, making partner status a real but non-exclusive trust signal. SP001, SP003, SP004
CP003 Public sources place Thinking Machines at over 110 to over 150 clients served and over 10,000 professionals trained, indicating meaningful proof of delivery but some source-to-source variation. SP003, SP004, SP005, SP006
CP004 Temus expands Thinking Machines’ regional reach and delivery bench, especially for production-grade deployments across Southeast Asia. SP004, SP005, SP006
CP005 Accenture is a global-scale incumbent with about 799,000 employees and more than 9,000 clients in over 120 countries. SP008
CP006 Accenture’s AI/Data positioning emphasizes enterprise-scale reinvention, agentic AI, AI-ready data, and ROI rather than local niche specialization. SP007
CP007 IBM combines software, consulting, and infrastructure at global scale and reports more than 300,000 employees across 170+ countries. SP010
CP008 IBM Consulting’s data-and-AI posture is alliance-heavy and hybrid-cloud oriented, with explicit links to Adobe, AWS, Microsoft, and Snowflake ecosystems. SP009
CP009 AWS Professional Services competes as a platform-native services arm offering AI-enhanced delivery, specialized solutions, and enterprise-scale data and generative-AI frameworks. SP011
CP010 Google Cloud Consulting competes with a partner-inclusive model and claims customers are materially more likely to implement cloud quickly when strategy is customized from the start. SP012
CP011 NCS is a regional incumbent rather than a pure Philippine boutique and cites No. 1 Southeast Asia services market share by vendor revenue in IDC’s 2025H1 tracker. SP013, SP014
CP012 NCS also highlights trust and managed-services credibility, including a Data Protection Trustmark and recognition in customer-experience and managed IT categories. SP013
CP013 Stratpoint presents itself as a Philippines-rooted digital-transformation firm with 25+ years of experience across software, cloud, data, and AI. SP015, SP016
CP014 Stratpoint claims to be the first homegrown AWS services provider in the Philippines and showcases customer proofs including Globe and UnionBank. SP015, SP016
CP015 Exist positions itself as one of the Philippines’ most awarded software-development firms and frames Data & AI as one pillar inside a broader enterprise-engineering offer. SP017, SP018
CP016 Senti AI positions itself as an AI pioneer in the Philippines and states it was acquired by Kollab in December 2024. SP019
CP017 Senti’s strongest overlap with Thinking Machines is in conversational AI and customer-service use cases, especially English, Tagalog, and Taglish contact-center workflows. SP020, SP021
CP018 Thinking Machines differentiates from Senti by spanning data platforms, deployment, training, and change-management work beyond conversational AI alone. SP001, SP002, SP020, SP021, SP025
CP019 Thinking Machines differentiates from Stratpoint and Exist by presenting a more explicit AI-and-data identity and by holding an OpenAI services badge neither local peer publicly foregrounds. SP001, SP003, SP015, SP017, SP018
CP020 Global incumbents outperform Thinking Machines on delivery scale, partner ecosystems, and breadth of enterprise transformation budget access. SP005, SP007, SP008, SP009, SP010, SP011, SP012, SP013, SP014
CP021 Thinking Machines likely outperforms global incumbents on boutique attention, Philippine-rooted credibility, and a sharper narrative around Southeast Asian enterprise AI deployment. SP001, SP004, SP005, SP006, SP025
CP022 AWS Professional Services and Google Cloud Consulting are both suppliers and substitutes for Thinking Machines because they can enable partners while also contracting directly with enterprise buyers. SP003, SP011, SP012
CP023 Switching costs appear low at the model or cloud-vendor layer because most competitors emphasize ecosystems, partnerships, and integration rather than closed proprietary stacks. SP009, SP011, SP012, SP015, SP017
CP024 Switching costs become materially higher once a deployment is embedded into data pipelines, workflows, governance, and change-management processes. SP001, SP002, SP004, SP021, SP023
CP025 Multi-homing is likely normal for enterprise buyers because system integrators, cloud platforms, and niche specialists each occupy different layers of the AI stack. SP009, SP011, SP012, SP017, SP020
CP026 Public price transparency is weak across the competitive set; most vendors market capabilities and outcomes while leaving contract pricing quote-based or undisclosed. SP007, SP009, SP011, SP012, SP015, SP017, SP020
CP027 Senti is the most visibly productized local peer in the fetched set because it names packaged solutions such as contact-center and chat-assistant offerings. SP020
CP028 Accenture, IBM, and NCS can bundle AI work into much larger transformation or managed-services budgets than Thinking Machines can on a standalone basis. SP007, SP008, SP009, SP010, SP013, SP014
CP029 Temus partly offsets Thinking Machines’ scale disadvantage by increasing bench depth, adjacent consulting capability, and regional distribution. SP004, SP005, SP006
CP030 Internal build remains a real substitute for large enterprises with strong engineering and data teams. SP022, SP023, SP024
CP031 Internal build is still constrained by talent scarcity, pilot-to-production friction, and governance burdens in the Philippine market. SP022, SP023
CP032 Feature breadth is distributed rather than concentrated: global incumbents dominate scale and alliances, cloud-native services dominate platform adjacency, and local specialists differentiate on context or product shape. SP007, SP009, SP011, SP012, SP015, SP017, SP020
CP033 Thinking Machines’ moat is execution-oriented — engineering, governance, and integration under messy enterprise conditions — not proprietary frontier-model ownership. SP004, SP005, SP025
CP034 That execution moat is more durable in embedded platform and workflow programs than in generic AI awareness or training offerings. SP001, SP002, SP003, SP023
CP035 Generic AI education, workshops, and advisory are exposed to commoditization as hyperscalers, partner programs, and large integrators expand enablement content. SP001, SP011, SP012, SP024
CP036 OpenAI partner status strengthens trust and signal quality but is not a hard lock-in mechanism because it can be diluted as broader partner networks expand. SP001, SP003
CP037 Regional integrators such as NCS and Temus show that Singapore-based firms are moving aggressively into production-grade AI delivery across Southeast Asia. SP004, SP013, SP014, SP024
CP038 The most credible local competitive threats to Thinking Machines are not identical copies of its full model but partial overlaps: Senti in conversational AI, Stratpoint in AWS-led transformation, and Exist in broader engineering-led data projects. SP015, SP017, SP020, SP021
CP039 Thinking Machines’ EastWest Bank proof and sector mix give it stronger regulated-enterprise case evidence than many small local AI boutiques. SP001, SP003
CP040 No single fetched competitor combines Thinking Machines’ Philippine roots, OpenAI service badge, data-platform depth, and recent Temus-backed regional scale-up in one package. SP003, SP004, SP005, SP015, SP017, SP020
CP041 Likely entrants over the next two years include additional cloud partners, regional GSIs, and better-capitalized internal enterprise AI teams rather than only local startups. SP011, SP012, SP023, SP024
CP042 The competitive set is fragmented enough that enterprises may intentionally mix vendors by use case, which reduces the odds of one-firm winner-take-most dynamics in the near term. SP009, SP011, SP012, SP020, SP023
CI001 Companies House PH lists Thinking Machines Data Science Inc. as a Philippine stock corporation registered under SEC number CS201522940 and established in 2015. SI016
CI002 Thinking Machines’ public site presents a services-led business rather than a self-serve SaaS product, emphasizing design, deployment, adoption, and implementation. SI001, SI002, SI003
CI003 The company’s likely core revenue streams include enterprise AI strategy, data-platform modernization, custom AI workflow implementation, and change-management support. SI001, SI003, SI005
CI004 Productized solution pages for customer, document, and location intelligence imply repeatable solution templates, but still appear to be delivered as enterprise services rather than self-serve subscriptions. SI003, SI006, SI007, SI008
CI005 Training and enablement are part of the commercial offer: the company says it has trained more than 10,000 professionals and advertises a free course as a top-of-funnel asset. SI001, SI009
CI006 The free online course is better interpreted as lead generation and category education than as direct evidence of material training revenue. SI001
CI007 The contact and solution pages imply a consultative sales process with bespoke scoping rather than published usage-based pricing. SI003, SI004
CI008 No public list pricing was found on the fetched Thinking Machines surfaces for implementation, platform, or training work. SI001, SI003, SI004, SI005
CI009 OpenAI partner status and the Temus combination likely contribute channel-driven lead flow in addition to direct founder-led and reference-led enterprise selling. SI009, SI010, SI012
CI010 Thinking Machines’ GTM appears concentrated on large-enterprise adoption problems rather than SMB self-serve conversion. SI001, SI004, SI021, SI022
CI011 Public traction proxies include 110+ to 150+ organizations served, 10,000+ professionals trained, and offices in Manila, Singapore, and Bangkok. SI009, SI010, SI012, SI013, SI014, SI015
CI012 No public revenue, ARR, gross margin, net income, or cash-balance figures were found for Thinking Machines. SI001, SI010, SI014, SI015, SI016, SI017
CI013 DealStreetAsia explicitly reports that Thinking Machines did not disclose the financial details of the Temus strategic investment. SI015
CI014 Dealroom describes the Temus financing as a strategic late-stage investment. SI014
CI015 Public reporting says the Temus investment is intended to expand Thinking Machines’ regional footprint while preserving brand, leadership, and operations. SI014, SI015
CI016 Temus frames the combination as expanding production-grade AI deployment capability and delivery infrastructure across Southeast Asia. SI010, SI023
CI017 A services-led AI integrator like Thinking Machines is likely labor-heavy, with consultant and engineer compensation as the dominant cost line. SI001, SI002, SI003, SI010
CI018 Likely secondary cost drivers include cloud/tool pass-through, pre-sales solutioning, travel, hiring, training, and partner enablement. SI001, SI003, SI009, SI010
CI019 Gross-margin outcomes likely depend on staff seniority mix, utilization, scope discipline, reuse of accelerators, and whether support becomes recurring. SI003, SI005, SI010, SI021
CI020 Working-capital risk likely arises from long enterprise sales cycles, milestone-based delivery, and collections timing rather than inventory or hardware exposure. SI004, SI021, SI022
CI021 Thinking Machines appears structurally less capex-intensive than infrastructure-heavy AI businesses because its public offer centers on services, software, and data workflows rather than owned compute assets. SI001, SI003, SI005
CI022 That lower capex profile does not eliminate financing need because regional expansion and senior talent acquisition can still consume cash quickly. SI010, SI014, SI021
CI023 No public evidence of debt facilities, project-finance obligations, or heavy balance-sheet leverage was found in the fetched materials. SI014, SI015, SI016, SI017
CI024 Public evidence does not suggest acute distress; instead it suggests a commercially credible but under-disclosed private company taking strategic capital to scale. SI010, SI014, SI015, SI023
CI025 The strategic nature of the Temus transaction implies external capital was still useful for expansion even if it does not prove cash scarcity. SI014, SI015
CI026 Official open-web access to detailed filings for this private Philippine company is limited: the SEC Express portal was blocked during retrieval and Companies House PH exposed only basic registry fields. SI016, SI017
CI027 Founder-profile and ecosystem-credibility sources improve confidence in company legitimacy and ecosystem position, but they do not resolve core underwriting metrics such as revenue or burn. SI018, SI019, SI020
CI028 The company’s business model likely mixes one-off project revenue with some repeat expansion work and possibly managed support, but the recurring share is not publicly quantified. SI001, SI003, SI010, SI012
CI029 There is no convincing public evidence that Thinking Machines operates a usage-based software revenue model at material scale. SI001, SI003, SI005, SI006, SI007, SI008
CI030 Because the commercial surface is quote-based and services-led, average deal size, delivery utilization, and attach-rate on follow-on work matter more than web-published sticker prices. SI003, SI004, SI021
CI031 EastWest Bank’s public customer proof suggests Thinking Machines can land regulated-enterprise work that may expand after initial adoption use cases succeed. SI001
CI032 Public-company investor-relations and SEC-filing portals for IBM and Accenture illustrate the level of disclosure available for mature public comparables and highlight how opaque Thinking Machines remains by contrast. SI024, SI025
CI033 The lack of disclosed revenue, margin, cash, and debt data means no responsible point estimate for TM revenue or runway can be made from public evidence alone. SI012, SI015, SI016, SI017
CI034 A responsible underwriting model would need backlog, average contract value, utilization, gross margin by service line, collection days, and partner-sourced pipeline contribution. SI009, SI010, SI021, SI022
CI035 The late-stage strategic investment likely supports expansion of delivery capacity and go-to-market reach more than balance-sheet-heavy asset purchases. SI010, SI014, SI015, SI023
CI036 A services company can often bootstrap longer than deep-tech infrastructure startups because it does not need to finance owned compute or manufacturing assets. SI001, SI003, SI021
CI037 However, expansion into Singapore and Bangkok plus higher-end enterprise delivery still implies meaningful bench-building and leadership-cost commitments. SI010, SI011, SI012
CI038 Financial opacity, not obvious commercial invalidity, is the primary blocker in this chapter. SI010, SI015, SI016, SI017
CI039 The combination of traction proxies, strategic investors, and enterprise reference points supports a view that revenue quality could be solid, but the evidence is not granular enough to test margin durability. SI009, SI010, SI014, SI001
CI040 Valuation work in Chapter 8 will therefore have to rely on service-business analogs, traction proxies, and wide scenario ranges rather than direct company financial disclosures. SI012, SI015, SI024, SI025
CE001 Thinking Machines presents a portfolio that spans data foundations, cloud data platforms, classic AI/ML solutions, and generative-AI deployment. SE001, SE002, SE007
CE002 Its Data Platforms offer focuses on secure, enterprise-grade public-cloud data infrastructure rather than on-prem proprietary appliances. SE002
CE003 The Data Platforms materials claim petabyte-scale analysis, encrypted central repositories, automated ingestion, and serverless warehouse-style processing. SE002
CE004 Customer Intelligence is positioned around customer-data unification, identity matching, micro-segmentation, and predictive modeling on cloud infrastructure. SE003
CE005 Customer Intelligence claims use of a pre-trained AI model for entity matching plus an AI model library and Python modeling frameworks. SE003
CE006 Document Intelligence is framed as a custom AI knowledge solution for document ingestion, extraction, search, and structured-data conversion at large scale. SE004
CE007 Document Intelligence explicitly says it is public-cloud agnostic and designed around open-standard technologies. SE004
CE008 Location Intelligence claims geospatial AI capabilities including satellite-image analysis, wealth prediction, land-use change extraction, and infrastructure detection. SE005
CE009 Location Intelligence claims access to a broad geospatial data catalog, data partnerships, and pre-trained AI/ML models developed by the company. SE005
CE010 The generative-AI offer is staged around educate, experiment, and execute, signaling a deployment pathway rather than a single model or tool. SE006
CE011 The GenAI page claims a proven production framework tested in banking and retail. SE006
CE012 TechEDT reports that the OpenAI collaboration includes ChatGPT Enterprise enablement, custom agentic-AI app design, hands-on training, and implementation frameworks. SE009
CE013 OpenAI’s partner page says Thinking Machines co-creates with client teams and embeds change management from day one. SE008
CE014 The homepage says forward-deployed engineers work alongside client teams to identify workflows, integrate systems, and move use cases into production. SE001
CE015 Temus describes Thinking Machines as strong in the engineering, governance, and integration layer needed to run AI systems under constrained data, regulatory requirements, and complex workflows. SE010, SE016
CE016 The overall product stack looks more like an applied-AI systems integrator with reusable accelerators than a standalone packaged-software vendor. SE001, SE003, SE004, SE005, SE006, SE007
CE017 The company repeatedly claims deployment across all major public cloud platforms, supporting a cloud-agnostic rather than single-vendor architecture. SE002, SE003, SE004, SE005
CE018 Public materials show dependence on external model and cloud ecosystems — especially OpenAI and public-cloud providers — rather than ownership of foundational model infrastructure. SE002, SE006, SE008, SE009
CE019 No public evidence suggests Thinking Machines owns a foundation model or custom semiconductor stack. SE001, SE006, SE008, SE009
CE020 The strongest public evidence for productized technical assets sits in geospatial tooling and applied models rather than in general-purpose enterprise software. SE005, SE012, SE013, SE021, SE022, SE023
CE021 UNICEF Venture Fund says Thinking Machines created GeoMancer and Tiffany as open-source geospatial tools and used AI to estimate household wealth from satellite and spatial data. SE012
CE022 UNICEF’s innovation profile says Thinking Machines later developed GeoWrangler, maintained a Sustainability Team from 2021, and built models such as PM2.5 estimation for Thailand. SE013
CE023 The GitHub repositories provide direct developer-signal evidence that GeoMancer, GeoWrangler, and Tiffany were published as reusable engineering artifacts under Thinking Machines branding. SE015, SE021, SE022, SE023
CE024 GeoMancer supports feature engineering across vector data and multiple data-warehouse backends, implying practical internal tool-building capability around geospatial ML workflows. SE021
CE025 GeoWrangler is described as a community-maintained geodata-wrangling package supported on a volunteer best-effort basis, which is useful as developer-signal but not enterprise support proof. SE022
CE026 Tiffany shows tooling for labeled geospatial image generation and demonstrates practical computer-vision workflow support rather than a broad commercial platform. SE023
CE027 Across multiple official pages, Thinking Machines claims publication activity in top journals and use of models such as BERT and T5 in client work. SE002, SE003, SE004, SE005
CE028 UNICEF sources strengthen that engineering-culture story by showing externally referenced geospatial research and tooling rather than pure marketing assertions. SE012, SE013, SE014
CE029 The company’s technical culture appears multidisciplinary, blending data engineering, ML, system architecture, domain operations, and change management. SE001, SE006, SE008, SE010, SE015, SE025
CE030 Trust and safety posture is visible mainly through design language — secure repositories, encryption, cloud controls, governance frameworks, and human-centered deployment — rather than through public certifications or uptime guarantees. SE001, SE002, SE004, SE008, SE011
CE031 The NPC AI advisory is relevant because many TM use cases involve personal data, documents, and customer records, making lawful basis, minimization, transparency, and human intervention real operating requirements. SE003, SE004, SE011
CE032 Public sources do not confirm formal certifications such as SOC 2, ISO 27001, or a public status-page reliability program. SE001, SE020, SE025
CE033 Product maturity varies by module: data-platform and classical data/AI services appear seasoned, while the GenAI layer looks newer but already packaged with explicit deployment methodology. SE001, SE002, SE006, SE010
CE034 The Bangkok office and broader Southeast Asian footprint suggest a support and deployment model that is extending beyond the Philippines. SE010, SE019
CE035 The product roadmap that is visible publicly is organizational and packaging oriented — Advanced Partner status, Temus integration, Bangkok expansion, and expanded OpenAI programs — more than feature-changelog oriented. SE001, SE009, SE010, SE019
CE036 Practical enterprise orientation is reinforced by third-party commentary around banking workflows, executive enablement, and business AI application rather than consumer experimentation. SE006, SE009
CE037 The best public production-readiness evidence is still implementation narrative rather than benchmark data: companies across banking and retail, EastWest Bank proof on the homepage, and Temus statements about production-grade AI. SE001, SE006, SE010
CE038 The company appears to rely on co-creation and client-specific architecture tailoring, which is good for fit but means supportability and repeatability are harder to verify from public pages alone. SE008, SE014, SE025
CE039 No public patent portfolio or exclusive IP estate was found in the fetched materials. SE012, SE013
CE040 Overall, Thinking Machines shows broad applied-AI capability, unusually strong public geospatial R&D/developer signal, and a credible deployment playbook, but weaker evidence of a hard standalone software moat. SE001, SE010, SE012, SE013, SE021, SE022, SE023
CU001 Public sources place Thinking Machines’ customer base at over 110 to over 150 organizations, depending on the source and date. SU003, SU004, SU005, SU020, SU021
CU002 The lower bound visible in partner and press sources is about 110 customers, while OpenAI’s partner profile gives a higher figure of over 150 clients. SU003, SU004
CU003 The customer base is concentrated in enterprise and institution-like segments rather than self-serve SMBs, especially financial services, retail, conglomerates, and civic organizations. SU003, SU004, SU015
CU004 Thinking Machines’ public customer narrative remains Philippines-rooted but increasingly Southeast Asia-oriented through offices and the Temus relationship. SU004, SU005, SU006, SU015
CU005 The strongest named public customer proof in the fetched set is EastWest Bank. SU001, SU010, SU011, SU012
CU006 Thinking Machines’ homepage quotes EastWest Bank saying TM was a key partner in adopting and productionalizing AI and that bank staff gained more time for higher-value tasks. SU001
CU007 EastWest’s own official communications show continued investment in AI-enabled customer experience, digital onboarding, dispute resolution, and advisory workflows in 2025-2026. SU011, SU012, SU013
CU008 EastWest disclosed that digital transactions reached 51 percent of total transactions in 2025. SU011
CU009 EastWest reported an eight-point improvement in Net Promoter Score and a lower complaints ratio, but those are customer-side bank metrics rather than direct Thinking Machines retention metrics. SU011
CU010 EastWest’s ESTA platform won three 2026 Digital CX Awards and functions as both an acquisition and service channel for the bank. SU012
CU011 EastWest’s official pages show a sophisticated digital customer environment, which supports the interpretation that TM’s EastWest work sits inside a meaningful production environment. SU011, SU012, SU014
CU012 OpenAI’s partner page says Thinking Machines has served clients across financial services, retail, conglomerates, and civic organisations, reinforcing vertical breadth even though logos are sparse. SU003
CU013 Customer acquisition appears consultative and enterprise-led rather than transactional, as shown by the contact flow and services-led site structure. SU017, SU018, SU019
CU014 Partner channels probably matter for customer acquisition and expansion because both OpenAI and Temus place Thinking Machines inside broader enterprise transformation flows. SU003, SU004, SU018
CU015 The public case-study library is lighter than the raw client-count claims would suggest. SU003, SU009, SU016
CU016 That imbalance could mean a confidentiality-heavy enterprise base, but it could also mean public proof is underdeveloped relative to underlying customer activity. SU003, SU004, SU009, SU016
CU017 Thinking Machines’ public customer evidence supports strong buyer-quality signals, but not enough density to map logo-by-logo sector concentration. SU003, SU004, SU010, SU024
CU018 No public source in the fetched set discloses top-customer revenue concentration, top-10 customer share, or geography-by-revenue. SU003, SU004, SU020, SU021
CU019 No public NRR, GRR, churn, renewal-rate, or contract-length data were found for Thinking Machines. SU001, SU003, SU004, SU009
CU020 As a result, retention and durability can only be inferred indirectly through production references, follow-on customer behavior, and expanding partner confidence. SU004, SU011, SU012
CU021 Swarm’s evidence that 65 percent of Philippine enterprises remain in pilot mode suggests procurement friction and deployment bottlenecks remain material in the overall customer environment. SU023
CU022 Trade.gov’s emphasis on IT-BPM and enterprise-AI demand supports the view that TM’s customer base is built around sizable, workflow-heavy organizations. SU022
CU023 The GenAI materials describe banking operations use cases and branch-officer adoption, implying users inside customer accounts include frontline staff, branch personnel, and operational teams. SU001, SU019
CU024 The customer base likely values TM most where it solves core operating workflows rather than generic experimentation. SU001, SU011, SU012, SU019
CU025 UNICEF Venture Fund says TM’s geospatial analytics line won one of the company’s biggest contracts to date: satellite-imagery analysis for a major telecommunications company in Southeast Asia. SU024
CU026 That telecom proof expands customer diversity beyond banking, even though the customer name was not publicly disclosed. SU024, SU025
CU027 UNICEF and UNDP-linked materials also show public/civic-development demand for TM’s geospatial and climate-adjacent work. SU024, SU025
CU028 The stories directory says TM has developed dozens of custom GenAI apps, which supports account activity breadth even without named logos for each project. SU009
CU029 Public proof is stronger on depth of a few reference stories than on broad deployment transparency across the full client base. SU009, SU010, SU011, SU012
CU030 There is no clear public evidence of customer churn or failed production deployments in the fetched materials. SU009, SU016, SU021
CU031 Absence of visible churn evidence is not the same as proof of strong retention. SU019, SU021
CU032 Customer satisfaction evidence for Thinking Machines itself is limited to qualitative testimonials and partner/customer quotes rather than systematic scoring. SU001, SU003, SU011
CU033 The Temus transaction likely improves TM’s ability to cross-sell into Singapore and larger regional enterprise accounts. SU004, SU005, SU015
CU034 Expansion loops are most plausible where TM lands with data-platform or AI-adoption work first and later broadens into customer-facing or GenAI workflows. SU001, SU010, SU019
CU035 Because public customer evidence is sparse relative to total logos, top-customer concentration risk remains a material but unresolved diligence item. SU003, SU009, SU018
CU036 Overall, Thinking Machines appears to have a high-quality enterprise customer base with credible production proof, but not enough public retention or concentration data to underwrite durability tightly. SU003, SU004, SU011, SU012, SU024
CR001 Swarm identifies talent scarcity as a top barrier to enterprise AI scale in the Philippines, cited by 57 percent of respondents. SR002
CR002 Swarm also identifies security and privacy concerns as a top barrier, cited by 40 percent of respondents. SR002
CR003 Swarm reports that 65 percent of organizations remain at proof-of-concept stage, creating a real execution and conversion risk for AI vendors. SR002
CR004 Trade.gov says only 14.9 percent of Philippine firms use AI technologies, reinforcing how early and uneven the market still is. SR001
CR005 OECD’s NAISR profile lists limited local use cases, scarce compute and talent, insufficient data-strategy capacity, and legal uncertainty as adoption barriers. SR008
CR006 The NPC advisory makes it clear that the Data Privacy Act applies to AI development, training, testing, and deployment when personal data is involved. SR003
CR007 The NPC advisory requires transparency, lawful basis, minimization, governance, bias monitoring, and meaningful human intervention for AI systems. SR003
CR008 BSP Memorandum M-2026-031 is voluntary today, but it establishes supervisory expectations for BSFIs that could tighten over time. SR004, SR005
CR009 Gorriceta and eLegal both frame NAISR 2.0 as a supportive but evolving policy environment rather than a settled AI rulebook. SR006, SR007
CR010 Digital in Asia and Nexdigm both indicate that the Philippines remains earlier in infrastructure maturity than regional leaders and still faces power and scale constraints. SR009, SR010
CR011 Nexdigm explicitly cites high electricity cost and power reliability as constraints on AI infrastructure expansion in the Philippines. SR010
CR012 Thinking Machines’ public offer depends materially on OpenAI and public-cloud ecosystems for important parts of its GenAI and deployment stack. SR011, SR012
CR013 The Temus transaction creates upside in scale but also integration, governance, and channel-dependence risk. SR013, SR014
CR014 Customer concentration is unresolved because public sources do not disclose top-customer share, top-10 share, or contract tenure. SR012, SR014, SR029
CR015 The EastWest reference is high quality, but it also highlights how much of the public customer narrative leans on one clearly named account. SR011, SR029
CR016 Financial risk is amplified by opacity: public evidence does not disclose revenue, gross margin, burn, runway, or debt. SR014, SR016
CR017 Services-model competition creates a real margin-compression risk because global integrators, hyperscalers, and local peers can all contest implementation work. SR001, SR002, SR013
CR018 World Bank data says the Philippines remains one of the region’s faster-growing economies, but still faces global slowdown and trade-policy uncertainty. SR020
CR019 World Bank also emphasizes the country’s exposure to natural hazards and climate resilience needs, which can affect enterprise continuity and public-sector priorities. SR020
CR020 BSP maintains Philippine-peso per U.S. dollar exchange-rate monitoring, which matters because cloud/tool costs and regional contracts can create FX sensitivity for a multi-country AI services firm. SR021
CR021 The IMF country page reinforces that macro surveillance remains relevant for the Philippines, especially as external conditions shift. SR022
CR022 Thinking Machines' public geospatial repositories expose maintenance activity through open GitHub issue trackers rather than through any published enterprise support SLA. SR023, SR024, SR025
CR023 That makes the open-source tooling a positive technical-culture signal, but not a substitute for contracted delivery support when enterprise deployments run into production issues. SR023, SR024, SR025, SR028, SR031
CR024 UNICEF’s account of a major telecom contract and geospatial data work implies dependency on external data access and rights for some solution lines. SR030
CR025 The company’s public materials do not confirm formal security certifications or public reliability artifacts, leaving trust-control completeness unresolved. SR003, SR011, SR012
CR026 Blocked or limited access to some Philippine business press and official filing channels increases diligence friction and leaves blind spots in external verification. SR016, SR026, SR027
CR027 Stephanie Sy remains highly central to company narrative, founder identity, and external trust, which makes key-person concentration a real execution risk. SR018, SR019
CR028 HBSP and Ignition both suggest a founder-led culture that values learnability and storytelling, which is positive culturally but also reinforces leadership concentration. SR018, SR019
CR029 LeadIQ’s 51-200 employee signal conflicts with other public hints of larger operating footprint, reinforcing that organizational-scale visibility is imperfect. SR017
CR030 Headcount ambiguity matters because services businesses fail through under-hiring, utilization stress, or leadership-bench thinness more often than through hardware failure. SR002, SR017
CR031 Public-sector and development-oriented work can create procurement-timing and budget-cycle risk even when it improves strategic credibility. SR020, SR030
CR032 OpenAI partner status is a real go-to-market asset, but it can dilute as partner networks broaden or if model-vendor priorities shift. SR011, SR012
CR033 Temus lowers some go-to-market risk by adding scale, but it raises integration and strategic-control questions at the same time. SR013
CR034 There were no public litigation, enforcement, recall, or incident records found in the fetched set, which is reassuring but not definitive. SR003, SR004, SR026, SR027
CR035 Residual regulatory exposure remains meaningful because TM works in customer-data, document, and financial-services-adjacent workflows where AI rules are becoming more explicit. SR003, SR004, SR005, SR006
CR036 Residual operational exposure remains meaningful because adoption bottlenecks, infrastructure constraints, and repo-level best-effort maintenance signals can all slow delivery. SR002, SR010, SR023, SR024
CR037 Residual financial-model exposure remains high until revenue quality, utilization, margin, and concentration are disclosed. SR014, SR016
CR038 One thesis-break trigger would be loss or dilution of key partner status without offsetting proprietary customer pull. SR012, SR013
CR039 Another thesis-break trigger would be evidence that customer concentration is materially higher than expected or that marquee references fail to expand. SR014, SR015
CR040 Another thesis-break trigger would be regulatory action or customer resistance that blocks AI deployment in BFSI and customer-data workflows. SR003, SR004, SR005
CR041 Another thesis-break trigger would be inability to hire or retain enough senior technical talent to support regional delivery ambitions. SR001, SR002, SR017
CR042 Overall, the risk picture is manageable but real: the biggest unresolved risks are partner dependence, talent and delivery scale, customer concentration opacity, and regulatory tightening in data-rich verticals. SR002, SR003, SR013, SR014, SR017
CV001 Thinking Machines is evidenced publicly as a Philippines-rooted applied AI and data-science firm serving enterprise and development use cases, not as a frontier-model lab. SV001, SV002, SV009
CV002 Temus publicly announced a strategic investment in Thinking Machines on 30 July 2026, and independent coverage matched that description. SV004, SV005, SV006
CV003 Across the retrieved public record, Thinking Machines still does not disclose revenue, gross margin, retention, pricing, or cash metrics. SV001, SV002, SV003, SV004, SV012
CV004 No fetched official, partner, press, or filing-access source in this run substantiates a $10B+ valuation or a 2025 Series C led by a16z and Google for Thinking Machines Data Science, Inc. SV004, SV005, SV006, SV012
CV005 That means any aggressive private-market price today would be paying mostly for strategic optionality rather than disclosed operating fundamentals. SV004, SV012
CV006 The public record makes Thinking Machines look more like a services-heavy applied-AI company with platform upside than a pure software or frontier-model scarcity asset. SV002, SV007, SV008
CV007 OpenAI partner status improves third-party credibility, but it does not by itself prove monetization quality, customer depth, or moat durability. SV003, SV001
CV008 Named proof from EastWest and UNICEF reduces zero-traction risk by showing that Thinking Machines has shipped into recognizable organizations. SV007, SV008, SV009
CV009 The company can justify some premium to generic outsourcing because it appears to operate in governed enterprise AI and data-platform work rather than pure labor arbitrage. SV002, SV003, SV010
CV010 At the current evidence level, the most defensible recommendation is research-more rather than an affirmative buy call. SV004, SV012, SV011
CV011 Endava traded at about $0.15B market cap against about $0.98B TTM revenue in August 2026, implying a roughly 0.15x revenue multiple. SV021, SV022
CV012 Globant traded at about $1.60B market cap against about $2.45B TTM revenue in August 2026, implying roughly 0.65x revenue. SV023, SV024
CV013 EPAM traded at about $5.02B market cap against about $5.55B TTM revenue in August 2026, implying roughly 0.90x revenue. SV025, SV026
CV014 Genpact traded at about $5.59B market cap against about $5.16B TTM revenue in August 2026, implying roughly 1.08x revenue. SV027, SV028
CV015 Accenture traded at about $107.42B market cap against about $73.10B TTM revenue in August 2026, implying roughly 1.47x revenue. SV029, SV030
CV016 The five-comparable median public revenue multiple is about 0.9x, which is a restrained backdrop for pricing services-led enterprise-AI businesses. SV021, SV022, SV023, SV024, SV025, SV026, SV027, SV028, SV029, SV030
CV017 Thinking Machines could deserve a premium to those public comparables only if its AI specialization translates into better growth or more repeatable revenue than ordinary services firms. SV002, SV003, SV023, SV025
CV018 The Temus transaction anchors strategic relevance, but because public sources do not disclose round size or price mechanics clearly, it anchors credibility more than valuation. SV004, SV005, SV006, SV020
CV019 Public evidence still does not show liquidation preferences, share classes, dilution overhang, or other cap-table terms that could change investor returns materially. SV012, SV005
CV020 The cleanest comparable lens is therefore blended: high-skill digital-engineering and analytics companies with some AI premium, not frontier-model labs. SV002, SV021, SV023, SV025
CV021 Public comp dispersion shows how quickly market values compress when growth moderates and services mix dominates the story. SV021, SV023, SV025
CV022 The market-cap histories embedded in the fetched comp pages show that several comparables have de-rated sharply from 2024-2025 peaks into 2026. SV021, SV023, SV025, SV029
CV023 Any valuation case above low-single-digit revenue multiples would require proof of recurring software revenue, exceptional retention, or uniquely monetizable IP that public sources do not yet show. SV002, SV013, SV014, SV012
CV024 The bear case assumes Thinking Machines remains primarily a high-skill project and implementation business valued on conservative service multiples. SV002, SV007, SV021
CV025 The base case assumes the company proves repeatable enterprise AI delivery and earns a moderate strategic premium above generic consulting. SV003, SV004, SV023
CV026 The bull case assumes Temus, OpenAI, and visible customer proof compound into a more productized regional AI-platform narrative with stronger repeat revenue. SV003, SV004, SV008
CV027 Under conservative assumptions, a bear-case supportable value band is roughly $40m to $150m. SV021, SV022, SV012
CV028 A base-case supportable value band is roughly $180m to $500m if Thinking Machines proves better-than-consulting economics. SV023, SV025, SV004
CV029 A favorable strategic-premium case could support roughly $600m to $1.3b, but only if recurring economics and regional scale become demonstrably stronger. SV003, SV004, SV029
CV030 Even that favorable band remains far below the unsupported $10B-style narrative in the user prompt. SV004, SV005, SV012
CV031 Downside triggers include failure to show repeat revenue, concentration surprises, partner-halo decay, and services-style margin compression. SV012, SV007, SV004
CV032 Upside triggers include disclosed recurring revenue, repeatable product attach, broader customer proof, and clearer regional monetization. SV002, SV008, SV004
CV033 The recommendation could improve materially if management demonstrates software-like economics rather than one-off project economics. SV002, SV013, SV014
CV034 On public evidence alone, IPO readiness is low because financial, governance, and term-structure visibility are insufficient. SV012, SV001, SV019
CV035 A strategic sale or later-stage private financing is more plausible than a near-term IPO. SV004, SV005, SV019
CV036 Natural acquirer classes include regional IT-services consolidators, consulting/cloud partners, and data-platform buyers seeking Southeast Asian enterprise AI capacity. SV003, SV004, SV002
CV037 The five most important diligence asks are revenue quality, margin/utilization, retention/concentration, cap-table terms, and partner economics. SV012, SV004, SV003, SV008
CV038 One thesis-break trigger would be evidence that founder and a few marquee accounts carry more of the business than the organization can institutionalize. SV016, SV015, SV008
CV039 Another thesis-break trigger would be if the Temus strategic halo does not translate into measurable commercial expansion or delivery leverage. SV004, SV020, SV017
CV040 A third thesis-break trigger would be slowing regulated-industry AI adoption combined with rising talent and delivery costs. SV010, SV011, SV019
CV041 Overall risk rating is high because valuation support rests on sparse financial disclosure plus real execution and concentration uncertainty. SV012, SV007, SV015
CV042 Overall confidence is medium: the company-quality evidence is too strong for dismissal, but the valuation-input evidence is too incomplete for conviction pricing. SV004, SV003, SV012
来源
编号出版方标题引文
SO001 Thinking Machines Data Science Thinking Machines Data Science | From Data Foundations To Advanced (Gen)AI
SO002 Thinking Machines Data Science Thinking Machines Data Science | About
SO003 Thinking Machines Data Science AI, Data, & Cloud Computing Solutions | Thinking Machines Data Science
SO004 Thinking Machines Data Science Cloud Data & Analytics Platform | Thinking Machines Data Science
SO005 Thinking Machines Data Science Customer Intelligence Solution | Thinking Machines Data Science
SO006 Thinking Machines Data Science Document Artificial Intelligence Solution | Thinking Machines Data Science
SO007 Thinking Machines Data Science Location Intelligence Solution | Thinking Machines Data Science
SO008 Thinking Machines Data Science Thinking Machines Data Science | Generative AI
SO009 Thinking Machines Data Science Press Room | Thinking Machines Data Science
SO010 Thinking Machines Data Science Contact us | Thinking Machines Data Science
SO011 Temus Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SO012 OpenAI Thinking Machines Data Science
SO013 TNGlobal Singapore's Temus invests in Philippines' Thinking Machines, OpenAI's first APAC services partner
SO014 Context.ph Temasek-backed deal lifts Philippine AI startup ambitions
SO015 Media OutReach Newswire Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SO016 Business Daily Media Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SO017 The Sun Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SO018 Hustleshare Podcast Ep 355 - The Hustle Behind Thinking Machines
SO019 Forbes Stephanie Sy
SO020 Harvard Business Publishing This Data Scientist Left Silicon Valley to Start Her Own Company in the Philippines
SO021 Ignition Ignition Stories with Stephanie Sy of Thinking Machines
SO022 UNICEF Venture Fund Graduate: Thinking Machines
SO023 UNICEF Office of Innovation Thinking Machines Data Science
SO024 UNDP Digital X Digital X Solution: Thinking Machines Data Science
SO025 U.S. International Trade Administration Philippines Artificial Intelligence
SO026 TechEDT Thinking Machines partners with OpenAI to accelerate AI adoption in Asia Pacific
SO027 Makati Business Club GLOBAL IDEAS: UNLOCKING YOUR BUSINESS WITH AI
SO028 LeadIQ Thinking Machines Data Science Employee Directory, Headcount & Staff
SM001 U.S. International Trade Administration Philippines Artificial Intelligence
SM002 U.S. International Trade Administration Philippines - Strategic Technologies
SM003 OECD.AI National AI Strategy Roadmap 2.0 (NAISR 2.0)
SM004 National Privacy Commission Advisory No. 2024-04 Guidelines on Artificial Intelligence Systems Processing Personal Data
SM005 Baker McKenzie Philippines: BSP Releases AI Governance Framework
SM006 Swarm Philippine AI Report 2025: 92% Adoption, Yet 65% of Enterprises Are Still in Pilot Mode
SM007 eLegal Philippines DTI Unveils National AI Strategy Roadmap 2.0, Center for AI Research
SM008 Source of Asia Artificial Intelligence (AI) in Southeast Asia 2025-2026
SM009 Digital in Asia Who is Building AI Data Centres in Southeast Asia in 2026?
SM010 Nexdigm Philippines AI Infrastructure Industry Size, Industry Share, Compute and Data Center Expansion
SM011 Singapore Economic Development Board AI in Southeast Asia: An Era of Opportunity
SM012 Gorriceta Powered by AI – The Philippines’ National AI Strategy Roadmap 2.0
SM013 Bangko Sentral ng Pilipinas Governance Principles for Artificial Intelligence (AI) in Financial Services (M-2026-031)
SM014 Ajentik ASEAN AI Adoption: From AI-First to AI-Native in 2026
SM015 Bangko Sentral ng Pilipinas BSP issues AI governance principles for BSFIs
SM016 Thinking Machines Data Science Thinking Machines Data Science | From Data Foundations To Advanced (Gen)AI
SM017 Thinking Machines Data Science Cloud Data & Analytics Platform | Thinking Machines Data Science
SM018 Thinking Machines Data Science Thinking Machines Data Science | Generative AI
SM019 OpenAI Thinking Machines Data Science
SM020 Temus Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SM021 TNGlobal Singapore's Temus invests in Philippines' Thinking Machines, OpenAI's first APAC services partner
SM022 Context.ph Temasek-backed deal lifts Philippine AI startup ambitions
SM023 Business Daily Media Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SM024 Media OutReach Newswire Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SM025 Thinking Machines Data Science Thinking Machines Data Science | About
SP001 Thinking Machines Data Science Thinking Machines Data Science | From Data Foundations To Advanced (Gen)AI
SP002 Thinking Machines Data Science Cloud Data & Analytics Platform | Thinking Machines Data Science
SP003 OpenAI Thinking Machines Data Science
SP004 Temus Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SP005 TNGlobal Singapore's Temus invests in Philippines' Thinking Machines, OpenAI's first APAC services partner
SP006 Context.ph Temasek-backed deal lifts Philippine AI startup ambitions
SP007 Accenture Artificial Intelligence (AI) Services & Solutions
SP008 Accenture About Our Company | Accenture
SP009 IBM Data and AI Consulting Services | IBM
SP010 IBM About | IBM
SP011 AWS AWS Professional Services
SP012 Google Cloud Google Cloud Consulting
SP013 NCS NCS Singapore | Digital and Technology Services
SP014 NCS About Us | Consulting, AI, Digital, Cloud Services | NCS SG
SP015 Stratpoint Stratpoint Technologies | Accelerate your digital transformation
SP016 Stratpoint About Stratpoint | Leaders of accelerating digital transformation
SP017 Exist Software Labs Exist Software Labs Inc | Enterprise IT Solutions | Exist
SP018 Exist Software Labs Data & AI Solutions | Exist
SP019 Senti AI Artificial Intelligence Company - Senti AI
SP020 Senti AI Artificial Intelligence Solutions - Senti AI
SP021 Senti AI About the leading AI company in the Philippines - Senti AI
SP022 U.S. International Trade Administration Philippines Artificial Intelligence
SP023 Swarm Philippine AI Report 2025: 92% Adoption, Yet 65% of Enterprises Are Still in Pilot Mode
SP024 Singapore Economic Development Board AI in Southeast Asia: An Era of Opportunity
SP025 Thinking Machines Data Science Thinking Machines Data Science | About
SI001 Thinking Machines Data Science Thinking Machines Data Science | From Data Foundations To Advanced (Gen)AI
SI002 Thinking Machines Data Science Thinking Machines Data Science | About
SI003 Thinking Machines Data Science AI, Data, & Cloud Computing Solutions | Thinking Machines Data Science
SI004 Thinking Machines Data Science Contact us | Thinking Machines Data Science
SI005 Thinking Machines Data Science Cloud Data & Analytics Platform | Thinking Machines Data Science
SI006 Thinking Machines Data Science Customer Intelligence | Thinking Machines Data Science
SI007 Thinking Machines Data Science Document Intelligence | Thinking Machines Data Science
SI008 Thinking Machines Data Science Location Intelligence | Thinking Machines Data Science
SI009 OpenAI Thinking Machines Data Science
SI010 Temus Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SI011 Thinking Machines Stories Temus and Thinking Machines Strategic Partnership FAQs
SI012 TNGlobal Singapore's Temus invests in Philippines' Thinking Machines, OpenAI's first APAC services partner
SI013 Context.ph Temasek-backed deal lifts Philippine AI startup ambitions
SI014 Dealroom Temasek-backed Temus invests in Philippine AI firm Thinking Machines
SI015 DealStreetAsia Temasek-backed Temus invests in PH's Thinking Machines
SI016 Companies House PH THINKING MACHINES DATA SCIENCE INC.
SI017 Philippines SEC SEC Express System
SI018 Forbes Stephanie Sy
SI019 Hustleshare Stephanie Sy on building Thinking Machines
SI020 UNICEF Venture Fund Thinking Machines Data Science (graduate)
SI021 Swarm Philippine AI Report 2025: 92% Adoption, Yet 65% of Enterprises Are Still in Pilot Mode
SI022 U.S. International Trade Administration Philippines Artificial Intelligence
SI023 Media OutReach Newswire Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SI024 IBM Investor Relations Financial Reporting - Investor Relations | IBM
SI025 Accenture Investor Relations Annual Reports | Accenture
SE001 Thinking Machines Data Science Thinking Machines Data Science | From Data Foundations To Advanced (Gen)AI
SE002 Thinking Machines Data Science Cloud Data & Analytics Platform | Thinking Machines Data Science
SE003 Thinking Machines Data Science Customer Intelligence | Thinking Machines Data Science
SE004 Thinking Machines Data Science Document Intelligence | Thinking Machines Data Science
SE005 Thinking Machines Data Science Location Intelligence | Thinking Machines Data Science
SE006 Thinking Machines Data Science Generative AI for Work | Thinking Machines Data Science
SE007 Thinking Machines Data Science AI, Data, & Cloud Computing Solutions | Thinking Machines Data Science
SE008 OpenAI Thinking Machines Data Science
SE009 TechEDT Thinking Machines partners with OpenAI to accelerate AI adoption in Asia Pacific
SE010 Temus Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SE011 National Privacy Commission Advisory No. 2024-04 Guidelines on Artificial Intelligence Systems Processing Personal Data
SE012 UNICEF Venture Fund Thinking Machines Data Science (graduate)
SE013 UNICEF Innovation 30 Thinking Machines Data Science
SE014 UNDP DigitalX Thinking Machines Data Science
SE015 GitHub Thinking Machines Data Science organization
SE016 TNGlobal Singapore's Temus invests in Philippines' Thinking Machines, OpenAI's first APAC services partner
SE017 Context.ph Temasek-backed deal lifts Philippine AI startup ambitions
SE018 GitHub thinkingmachines/geomancer issues
SE019 Thinking Machines Stories Thinking Machines opens Bangkok office
SE020 GitHub thinkingmachines/tiffany releases
SE021 GitHub thinkingmachines/geomancer
SE022 GitHub thinkingmachines/geowrangler
SE023 GitHub thinkingmachines/tiffany
SE024 GitHub thinkingmachines/geowrangler releases
SE025 Thinking Machines Data Science Thinking Machines Data Science | About
SU001 Thinking Machines Data Science Thinking Machines Data Science | From Data Foundations To Advanced (Gen)AI
SU002 Thinking Machines Data Science Thinking Machines Data Science | About
SU003 OpenAI Thinking Machines Data Science
SU004 Temus Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SU005 TNGlobal Singapore's Temus invests in Philippines' Thinking Machines, OpenAI's first APAC services partner
SU006 Context.ph Temasek-backed deal lifts Philippine AI startup ambitions
SU007 Media OutReach Newswire Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SU008 Business Daily Media Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SU009 Thinking Machines Data Science Stories Thinking Machines Data Science | Data Stories and Case Studies
SU010 Thinking Machines Data Science Stories Transforming EastWest Bank’s Operations With a Smart Data Platform
SU011 EastWest Bank EastWest Builds Smarter, More Relevant Banking for Customers
SU012 EastWest Bank EastWest Bank Wins Three Digital CX Awards for ESTA
SU013 EastWest Bank News & Advisories | EastWest Bank
SU014 EastWest Bank Best Bank in the Philippines for Achieving Your Dreams
SU015 VIR Temus acquires Thinking Machines to scale enterprise AI
SU016 BusinessMirror Homegrown success: Thinking Machines’ decade of AI excellence attracts Singapore investment
SU017 Thinking Machines Data Science Contact us | Thinking Machines Data Science
SU018 Thinking Machines Data Science Customer Intelligence | Thinking Machines Data Science
SU019 Thinking Machines Data Science Generative AI for Work | Thinking Machines Data Science
SU020 Dealroom Temasek-backed Temus invests in Philippine AI firm Thinking Machines
SU021 DealStreetAsia Temasek-backed Temus invests in PH's Thinking Machines
SU022 U.S. International Trade Administration Philippines Artificial Intelligence
SU023 Swarm Philippine AI Report 2025: 92% Adoption, Yet 65% of Enterprises Are Still in Pilot Mode
SU024 UNICEF Venture Fund Thinking Machines Data Science (graduate)
SU025 UNDP DigitalX Thinking Machines Data Science
SR001 U.S. International Trade Administration Philippines Artificial Intelligence
SR002 Swarm Philippine AI Report 2025: 92% Adoption, Yet 65% of Enterprises Are Still in Pilot Mode
SR003 National Privacy Commission Advisory No. 2024-04 Guidelines on Artificial Intelligence Systems Processing Personal Data
SR004 Bangko Sentral ng Pilipinas Governance Principles for Artificial Intelligence (AI) in Financial Services (M-2026-031)
SR005 Baker McKenzie Philippines: BSP Releases AI Governance Framework
SR006 Gorriceta Powered by AI – The Philippines’ National AI Strategy Roadmap 2.0
SR007 eLegal Philippines DTI Unveils National AI Strategy Roadmap 2.0, Center for AI Research
SR008 OECD.AI National AI Strategy Roadmap 2.0 (NAISR 2.0)
SR009 Digital in Asia Who is Building AI Data Centres in Southeast Asia in 2026?
SR010 Nexdigm Philippines AI Infrastructure Industry Size, Industry Share, Compute and Data Center Expansion
SR011 Thinking Machines Data Science Thinking Machines Data Science | From Data Foundations To Advanced (Gen)AI
SR012 OpenAI Thinking Machines Data Science
SR013 Temus Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SR014 DealStreetAsia Temasek-backed Temus invests in PH's Thinking Machines
SR015 EastWest Bank News & Advisories | EastWest Bank
SR016 Philippines SEC SEC Express System
SR017 LeadIQ Thinking Machines Data Science employee directory
SR018 Harvard Business Publishing Stephanie Sy profile
SR019 Ignition Ignition Stories with Stephanie Sy of Thinking Machines
SR020 World Bank Group Philippines | World Bank Group
SR021 Bangko Sentral ng Pilipinas Statistics - Exchange Rate
SR022 IMF Philippines and the IMF
SR023 GitHub Issues · thinkingmachines/geomancer
SR024 GitHub Issues · thinkingmachines/geowrangler
SR025 GitHub Issues · thinkingmachines/tiffany
SR026 BusinessMirror Temus backs Filipina-founded Thinking Machines to scale enterprise AI across SEA
SR027 BusinessMirror Homegrown success: Thinking Machines’ decade of AI excellence attracts Singapore investment
SR028 GitHub Pull requests · thinkingmachines/geowrangler
SR029 EastWest Bank EastWest Builds Smarter, More Relevant Banking for Customers
SR030 UNICEF Venture Fund Thinking Machines Data Science (graduate)
SR031 GitHub Pull requests · thinkingmachines/geomancer
SV001 Thinking Machines Data Science Thinking Machines Data Science | About
SV002 Thinking Machines Data Science Cloud Data & Analytics Platform | Thinking Machines Data Science
SV003 OpenAI Thinking Machines Data Science
SV004 Temus Temus and Thinking Machines Data Science join forces to scale enterprise AI across Southeast Asia
SV005 DealStreetAsia Temasek-backed Temus invests in PH's Thinking Machines
SV006 Dealroom Temasek-backed Temus invests in Philippine AI firm Thinking Machines
SV007 EastWest Bank EastWest Builds Smarter, More Relevant Banking for Customers
SV008 Thinking Machines Data Science Stories Transforming EastWest Bank’s Operations With a Smart Data Platform
SV009 UNICEF Venture Fund Thinking Machines Data Science (graduate)
SV010 U.S. International Trade Administration Philippines Artificial Intelligence
SV011 Swarm Philippine AI Report 2025: 92% Adoption, Yet 65% of Enterprises Are Still in Pilot Mode
SV012 Philippines SEC SEC Express System
SV013 GitHub thinkingmachines/geowrangler
SV014 GitHub thinkingmachines/geowrangler releases
SV015 LeadIQ Thinking Machines Data Science employee directory
SV016 Ignition Ignition Stories with Stephanie Sy of Thinking Machines
SV017 BusinessMirror Temus backs Filipina-founded Thinking Machines to scale enterprise AI across SEA
SV018 BusinessMirror Homegrown success: Thinking Machines’ decade of AI excellence attracts Singapore investment
SV019 World Bank Group Philippines | World Bank Group
SV020 Thinking Machines Stories Temus and Thinking Machines Strategic Partnership FAQs
SV021 CompaniesMarketCap Endava (DAVA) - Market capitalization
SV022 CompaniesMarketCap Endava (DAVA) - Revenue
SV023 CompaniesMarketCap Globant (GLOB) - Market capitalization
SV024 CompaniesMarketCap Globant (GLOB) - Revenue
SV025 CompaniesMarketCap EPAM Systems (EPAM) - Market capitalization
SV026 CompaniesMarketCap EPAM Systems (EPAM) - Revenue
SV027 CompaniesMarketCap Genpact (G) - Market capitalization
SV028 CompaniesMarketCap Genpact (G) - Revenue
SV029 CompaniesMarketCap Accenture (ACN) - Market capitalization
SV030 CompaniesMarketCap Accenture (ACN) - Revenue