初创公司尽调
尽调报告 industrial AI / engineering simulation Series C 2026-06-22

PhysicsX

PhysicsX 尽调报告

PhysicsX 在 AI 原生工程上有可信的技术和商业势能,但公开披露中绝对收入、利润率、留存和客户集中度都太少,仅凭公开证据很难支持 $2.4B 的 Series C 定价。

封面要素

成立时间 01
2019 [CO007]
累计融资 02
467 USD M [CO026]
最新估值 03
2400 USD M [CO023]
员工数 04
300 + employees [CO030]
已确认收入增长 05
2x YoY [CO027]
已具名付费客户 06
3 [CU046]

公司概况

PhysicsX 是一家总部在伦敦的工业 AI 公司,正在为物理世界设计流程搭建 AI 原生工程平台。它的公开定位围绕 Large Physics Models、仿真数据编排和面向工程师的应用展开,帮助工业团队用更快的 AI 辅助分析,替代受求解器速度限制的迭代。到 2026 年中,公司已从 2019 年注册、2020 年发布,推进到由 Temasek 领投的 $300 million Series C;公开客户和合作伙伴证据显示,它已在半导体、数据中心电力系统、高阶汽车和高性能工程项目中取得牵引。核心尽调约束仍是披露质量:外部人士可以验证融资事件和战略相关性,但无法仅凭公开证据足够精确地验证底层软件经济性,从而完全支撑新的后期价格。

官网
www.physicsx.ai
成立时间
2019-08-01
创始人
Jacomo Corbo, Robin Tuluie
创立地点
London, UK
总部
London, UK
产品
AI 原生工程软件栈,覆盖仿真数据编排、Large Physics Model 开发,以及可接入 Ansys、CATIA、Siemens NX、OpenFOAM、STAR-CCM+ 等既有 CAE 工具的工程应用。
客户
航空航天与国防、半导体、工业机械、汽车、材料和能源领域的大型工业企业及工程团队。
商业模式
企业软件平台,面向关键任务工业项目销售,并配套高接触部署、模型开发和前置工程支持。
阶段
Series C
融资情况
2026-06-08 宣布超额认购的 $300 million Series C,估值约 $2.4 billion,使公开披露股权融资至少达到 $467 million。
[CO003, CO007, CO010, CO017, CO023, CO026, CO033, CU001]

执行摘要

主要优势

  • PhysicsX 用 Large Physics Models 和工作流软件压缩重仿真的工程周期,切中真实工业瓶颈。
  • 公司在当前阶段拿到罕见强的战略背书:Temasek、Siemens、NVIDIA、Applied Materials,以及其他工业或基础设施相关投资方和合作伙伴。
  • 来自 Siemens、Microsoft 和 GB1 的公开客户工作负载证据显示,平台用于关键工程问题,不只是沙盒试点。

主要风险

  • 关于绝对收入、毛利率、留存、服务占比和股权结构经济条款的公开披露仍太薄,无法有把握承销 $2.4B 估值。
  • 客户集中度可能不低;独立识别到的公开付费客户只有三家,积压订单和收入披露也很少。
  • 技术可复用性仍是现实风险,因为 PhysicsX 自家材料承认,部分 neural-operator 方法在激波、不规则几何、严格边界条件和其他硬物理工作负载上会吃力。

未决问题

  • 2025-2026 已确认收入、ARR、毛利率,以及按客户队列或垂直行业拆分的软件与服务占比。
  • 留存、扩张和集中度数据,包括最大客户收入占比、合同期限,以及部署能跨项目或站点扩张的证据。
  • 2026 Series C 的股权结构和优先权细节,包括清算优先权、是否有老股交易,以及把价格与可重复软件经济性对齐所需的证据。

目录

Chapter 01

01公司概览

1.1 身份、产品与布局

PhysicsX 将自己定位为一家英国创立的物理 AI 公司,为工业组织搭建 AI 原生工程平台。法律实体于 2019 年 8 月 1 日注册,最初名为 Motodynamics Ltd;后续公司材料称 PhysicsX 于 2020 年启动。到 2026 年 6 月,公开公司页面和注册记录显示总部在伦敦,注册地址为 Victoria House, 1 Leonard Circus,运营办公室位于伦敦和纽约。2026 年 6 月的 Series C 新闻稿还称,公司正在扩大 Bay Area 和 Singapore 布局;这一点重要,因为该轮融资明确绑定全球扩张,而非狭窄的英国足迹。 产品层面,当前官网和平台材料描述的是一个统一软件栈,覆盖产品生命周期中的仿真、物理 AI、数据和工程应用。披露的行业重点很广,但逻辑一致:航空航天与国防、半导体、材料、汽车和能源。因此,公司卖的是进入硬科技垂直行业的横向工程软件层,而不是单点设计工具。公开材料在工作流定位和集成广度上很强,但仍没有给出精确的 2026 年收入基数、ARR 或客户集中度画像。 [CO001, CO002, CO003, CO004, CO005, CO006]

Snapshot KPI 表
指标数值 / 状态日期置信度缺口 / 备注
法律注册1 Aug 20192019-08-01注册信息支持;公司营销更常引用 2020 年启动 / 创立叙事
总部London, United Kingdom2026-06-08英国注册处明确披露注册地址
办公布局London 和 New York;扩展 Bay Area 和 Singapore2026-06-08扩张足迹由公司披露,未独立细化
最新融资$300M Series C,估值约 $2.4B2026-06-08由 Temasek 领投
已披露股权融资公开轮次合计 $467M2026-06-08部分 2026 年报道将累计融资列为约 $500M
员工数官方 300+;2026 年 6 月独立报道约 3502026-06-08公开数字是方向性,而非精确薪资名册数据
确认收入增长同比翻倍2026-06-08绝对收入未披露
预订收入增长同比增长三倍2026-06-08积压转化和服务强度仍需尽调
客户增长同比增加一倍以上2026-06-08保留的官方来源未披露精确客户数
2026 年精确 ARR / 收入2026索取董事会材料或管理账,以锚定估值倍数

公开来源支持轮次规模和增长率,但精确 ARR / 收入、利润率和客户集中度仍未披露;null 表示未保留精确公开值。

[CO003, CO005, CO007, CO023, CO026, CO027]
FO002: PhysicsX 公司快照逻辑

PhysicsX 把资本、AI 基础设施、工业工作流和创始人主导的技术可信度,连成一条 GTM 闭环。

[CO001, CO032, CO033, CO034, CO036, CO040]

1.2 创始人、领导层与治理

公开的创始人叙事有两点清楚,一点模糊。独立报道、融资公告和 2023 年退出隐身阶段的材料一致将 Jacomo Corbo 和 Robin Tuluie 识别为创始组合。当前材料中,Corbo 是 CEO 兼联合创始人;Tuluie 被呈现为创始人或联合创始人兼董事长。两人都带来异常直接的 founder-market fit:Tuluie 曾在 Renault/Alpine、Mercedes Formula 1 和 Bentley 任职;Corbo 曾联合创立 QuantumBlack,并做过 Formula 1 比赛策略。这对一家向复杂工程组织销售仿真提速和优化收益的公司来说,相关性异常强。 2023 年以来,治理更专业化,但仍围绕创始人展开。Companies House 显示,Corbo 于 2023 年 1 月 1 日被任命为高级职员,Jim Baum 于 2023 年 10 月 20 日加入董事会,Laura Connell 在 Atomico 领投 Series B 后于 2025 年 6 月 20 日加入董事会。About 页面还列出 Nicolas Haag 为联合创始人兼仿真工程董事,这拓宽了技术创始班底,但也与多数融资报道使用的双创始人叙事之间形成轻微的创始人名单歧义。COO Alexander Dreismann 和北美负责人 Mark Huntington 显示出公开可见的运营厚度,但关键人依赖仍集中在 Corbo 和 Tuluie 身上,因为资本市场、技术愿景和外部合作伙伴信息主要由他们主导。 [CO010, CO011, CO012, CO013, CO014, CO015]

领导层与创始人表
人物职位背景创始人-市场匹配 / 职能覆盖关键人物依赖
Robin TuluieFounder / Chairman前 Renault(Alpine)和 Mercedes F1 工程负责人;后任 Bentley 车辆技术总监为工业工作流带来深厚仿真和顶级性能工程可信度高 - 技术愿景;资本叙事;伙伴信号
Jacomo CorboCEO & Co-Founder前 QuantumBlack 联合创始人 / 首席科学家;McKinsey 合伙人;Renault F1 比赛策略师连接前沿 AI 方法与企业工业部署高 - CEO;融资牵头;产品-市场叙事
Nicolas Haag联合创始人兼仿真工程负责人About 页面列为技术联合创始人,聚焦仿真工程拓宽技术创始团队,并把平台连接到领域工作流中 - 领域深度重要,但外部可见度较低
Alexander DreismannCOOAbout 页面和早期领导层材料列名的运营高管把运营和交付规模化推进到创始人带宽之外中 - 执行深度,而非市场身份
Jim BaumBoard Member / DirectorSeries A 阶段披露中加入董事会;此前在 Netezza、Endeca 和 PTC 有运营经历增加企业软件规模化和治理经验中 - 治理和规模化支持
Laura Connell董事 / Atomico 合伙人2025 年 6 月 20 日被任命为董事,当时 Atomico 领投 Series B来自成长阶段领投方的董事会级投资人监督低 - 治理影响力,不是日常运营者

覆盖不完整,因为公司仍是私营企业,公开创始人叙事在官网页面和融资材料之间并不完全一致。

[CO010, CO011, CO012, CO013, CO015, CO016]

1.3 融资、投资人和规模

PhysicsX 已完成三轮公开披露的股权融资,规模快速放大。它于 2023 年 11 月以 General Catalyst 领投的 $32 million Series A 退出隐身阶段,随后在 2025 年 6 月 22 日宣布由 Atomico 领投的 $135 million Series B,又在 2026 年 6 月 8 日宣布由 Temasek 领投、估值约 $2.4 billion 的 $300 million Series C。三轮披露融资额合计至少 $467 million;部分 2026 年新闻报道将总额四舍五入为约 $500 million。2026 年投资人名单在战略上值得注意,因为它把主权资本、工业巨头和 AI 基础设施敞口混在一起,包含 Siemens、NVIDIA、Applied Materials、NGP 等名字。 最新一轮还伴随强但仍不完整的规模披露。官方材料称,已确认收入同比翻倍,预订收入增长两倍,客户基数扩大一倍以上,员工数在 12 个月翻倍后超过 300 人。独立媒体将 2026 年 6 月的员工数估在接近 350,并把需求与半导体、数据中心电力和冷却、以及更广泛的工业基础设施联系起来。因此,增长叙事很有吸引力,但缺少绝对收入、ARR、毛利率或客户集中度披露,意味着投资人仍需要私下证据,才能只按软件式经济性来承销该估值。 [CO019, CO020, CO021, CO022, CO023, CO024]

利益相关方或投资人图谱
利益相关方角色控制权 / 经济重要性尽调问题
TemasekSeries B 投资人和 Series C 领投方最新 $300M 轮次的领投方;显示主权级信心和 Singapore 扩张支持确认持股比例、治理权,以及任何 pro rata 或否决条款
AtomicoSeries B 领投方催化 2025 年估值跃升轮次,并通过 Laura Connell 拥有董事会代表确认董事会权利、保留事项,以及对未来流动性选项的支持
General CatalystSeries A 领投方 / 持续投资人公开记录中最早的领投机构支持者;后续轮次继续支持确认当前稀释、持股,以及早期轮次是否存在结构化条款
Siemens战略投资人和工业伙伴投资人叠加 2026 年数据中心电力合作;可能在工业软件中带来分发可信度区分股权上行、商业依赖和排他性风险
NVIDIA战略投资人 / 生态伙伴投资敞口叠加围绕 physics AI 和 LPMs 的标准与基础设施协同厘清关系带来的是渠道杠杆,还是技术集中风险
Applied Materials战略投资人半导体相邻投资人支持「芯片制造工作流是核心需求驱动」这一论点评估是否能进入半导体 OEM 管线,而不只是被动资产负债表支持
NGPSeries A 起的投资人可见的重复支持者,也转发了公司融资材料厘清长期持有意愿,以及 GTM 中是否有能源转型相邻性
M&G Investments新 Series C 投资人最新轮次加入传统机构成长资本了解预期持有期和估值纪律
Intrepid Growth Partners新 Series C 投资人最新 cap table 中的新成长投资人确认持股、董事会权利,以及是否有集中的回报预期
CoreWeave基础设施伙伴保留来源中未确认股权关系,但对训练和部署私有 Large Physics Models 具有战略重要性判断算力关系是否有批量折扣、排他性或可替代性

覆盖不完整,因为未保留公开 cap table、二级出售披露或债务安排;表格聚焦最可见的财务和战略利益相关方。

[CO019, CO020, CO023, CO024, CO025, CO036]
FO003: PhysicsX 快照 KPI

公开 KPI 披露强在增长方向和融资,弱在绝对财务基数。

[CO023, CO026, CO027, CO029, CO031, CO040]

1.4 里程碑、合作伙伴和未解缺口

2026 年的里程碑模式显示,PhysicsX 想成为基础设施,而不只是一个有趣的工程应用供应商。2026 年 3 月的一周内,它宣布了 CoreWeave 云合作、GB1 America's Cup 部署、NVIDIA 标准倡议,以及围绕数据中心电力基础设施的 Siemens 协作。合起来看,这些里程碑意味着三条商业化路径并行:战略算力和分发、借顶级性能工程建立品牌、以及直接部署到 AI 供应链瓶颈中的工业场景。2026 年 6 月 Series C 则为下一阶段扩张提供资金,包括 Singapore 和持续美国扩张。 保留证据中最强的不利信号不是法律或监管,而是估值纪律。NewMarketPitch 认为,只有当 PhysicsX 把 backlog 转化为大得多的已实现收入,并证明可复用平台模式而非服务密集型工程实践时,$2.4 billion 价格才说得过去。这一批评与真实的公开数据缺口一致:预订收入增长快于已确认收入,Corbo 也称公司受供给侧约束。因此,进一步尽调应聚焦完整控制权图谱、2025/2026 年绝对财务数据、任何二级交易或债务,以及跨客户部署的可复用性。 [CO018, CO023, CO025, CO036, CO037, CO038]

里程碑表
日期事件类型金额 / 估值 / 状态参与方含义
2019-08-01注册为 Motodynamics Ltd创立成立为活跃英国私营公司创立法律实体锚定 2019 年法律起点,尽管营销后来强调 2020 年启动
2020-07-09从 Motodynamics Ltd 更名为 PhysicsX治理完成更名PHYSICSX LIMITED显示从预发布壳公司转向品牌化运营公司
2020公司材料称 PhysicsX 已启动创立公开起源故事Robin Tuluie / Jacomo Corbo / 早期团队解释为什么部分来源使用 2020 年创立叙事,而不是法律注册日期
2023-01-01Corbo 高管任命治理记录高管任命Jacomo Corbo在公开走出隐身前,正式化领导层演进
2023-10-20Jim Baum 被任命为董事治理记录董事会新增成员Jim Baum在更大规模融资前加入上市公司运营经验
2023-11-27PhysicsX 走出隐身并宣布 Series A融资$32M Series AGeneral Catalyst、Standard Investments、NGP、Radius、Henry Kravis 等投资方公司公开亮相,并为早期商业规模化提供资金
2025-06-22宣布 Series B融资$135M;累计融资近 $170MAtomico、Temasek、Siemens、Applied Materials、July Fund、General Catalyst、NGP、Radius、Standard、Allen & Co 等投资方把 PhysicsX 推入后期 deeptech 区间,并扩大了战略投资人阵容
2026-03-11宣布 CoreWeave 合作合作战略基础设施合作CoreWeave将 PhysicsX LPM 训练和部署连接到聚焦 AI 的云基础设施
2026-03-13宣布 GB1 合作合作官方 AI Engineering Platform PartnerGB1 / 英国 America’s Cup 赛队展示顶级性能工程部署和品牌光环
2026-03-16宣布 NVIDIA 标准合作产品开放标准 / Opora / LPM 叙事NVIDIA将 PhysicsX 定位为 physics-AI 架构惯例的塑造者
2026-03-17宣布 Siemens Smart Infrastructure 合作合作数据中心电力优化项目Siemens Smart Infrastructure瞄准 AI 工厂电力基础设施这一快速增长的工业瓶颈
2026-06-08宣布 Series C融资$300M,估值约 $2.4BTemasek、M&G、Intrepid、Applied Materials、Atomico、General Catalyst、July Fund、NGP、NVIDIA、Radius、Siemens 等投资方资助美国和 Singapore 扩张,并巩固 PhysicsX 作为旗舰工业 AI 融资案例的地位

这是公开保留里程碑的记录年表;若无法获得具体日期,表格保留公开叙事中的期间级表述。

[CO007, CO008, CO009, CO014, CO015, CO018]
FO001: PhysicsX 里程碑时间线

融资和 2026 年 3 月的合作公告,标出 PhysicsX 公开规模化叙事中的关键拐点。

[CO018, CO023, CO025, CO036, CO037, CO038]

1.5 图表

Chapter 02

02市场分析

2.1 市场边界、纳入支出和现状替代方案

PhysicsX 更适合被界定为工程仿真之上的 AI 原生层,而不是泛工业 AI 公司。它自己的产品描述聚焦于把仿真、物理 AI、数据和工程应用统一到产品生命周期中;合作伙伴和面向客户的材料,则把它放在航空航天、半导体、汽车、材料和能源项目里,物理预测在这些项目中塑造设计决策。这意味着相关市场包括 CAE、由求解器支撑的工程分析、高价值数字孪生验证,以及压缩这些工作流的 AI surrogate 工具。它不包括大多数泛企业 AI、工厂自动化硬件和非工程 SaaS。现状不是空白地带:Ansys、Siemens/Altair、Synopsys、Cadence 等既有厂商已经横跨流体、结构、电子、多物理场和数字孪生工作流,因此 PhysicsX 必须作为叠加层或工作流升级胜出,不能假设买家从零开始。[CM001, CM003, CM004, CM005, CM006, CM012]

市场定义表
细分 / 类别包含支出排除支出买方 / 付款方与 PhysicsX 的相关性
核心工程仿真 / CAEFEA、CFD、结构、热、电磁和系统级仿真软件通用办公 AI 和非工程 SaaS工程领导层、仿真团队、产品开发PhysicsX 必须在工作流层面附着或替代的核心存量支出。
Physics-AI / 替代层加速 solver 设置、筛选、优化和设计空间探索的 AI 模型没有基于物理工作流集成的纯聊天机器人CTO、工程 VP、仿真负责人、高级 R&D最接近 PhysicsX 价值主张的表达。
EDA 相邻的物理系统分析芯片-封装-系统热分析、先进封装、3D-IC 和电子可靠性分析纯逻辑综合和非物理数字设计任务半导体和电子项目负责人重要,因为 PhysicsX 瞄准半导体,而从 silicon 到 systems 的边界正在变模糊。
数字孪生 / 验证层运营孪生、虚拟调试,以及与工程资产绑定的基于模型验证与工程模型脱节的通用 BI 仪表盘运营工程、产品工程、平台负责人在有支持的地方,把 PhysicsX 从设计延伸到运营。
云 / HPC 支撑的仿真交付弹性算力、安全浏览器部署、工程 solver 工作流自动化与工程工作流无关的商品化云支出工程 IT、平台工程、仿真 COE往往是让 AI 辅助仿真可规模化部署的预算路径。
排除的通用工业 AI独立 copilot、通用企业分析和自动化硬件CIO 或自动化预算这些相邻,但不是 PhysicsX 需求的干净代理。

边界逻辑把以 solver 为中心的工程支出,与更宽泛的工业 AI 叙事区分开。PhysicsX 主要参与物理驱动工程决策和验证已经重要的场景。

[CM001, CM003, CM004, CM005, CM006, CM012]
FM001: 市场规模观察框架

机会从全部仿真软件收窄到一组更小、可由 AI 增强的工程工作流,再进一步收窄到更小的初始商业切片。

[CM003, CM006, CM044, CM046, CM051, CM052]

2.2 TAM/SAM/SOM 必须用多重视角划边界,不能只用一个泛 TAM

公开市场规模只能支撑一个区间,而不是单一权威数字。较窄的 CAE 估算在 2026 年约 USD 7.64 billion 至 USD 11.53 billion 之间聚集;更宽的仿真软件估算在 2026 年达到 USD 15.46 billion,若纳入相邻软件类别,总量会明显更大。差异主要来自范围,而非增长判断:有些发布方只统计核心 CAE 工具,另一些则纳入更广的仿真软件、服务、云交付和数字孪生工作流。对 PhysicsX 来说,宽口径 TAM 因此是仿真软件市场;但更可辩护的 SAM,是一组更小的工程项目:虚拟验证已具战略意义,且买家能够支撑数据和算力要求。谨慎的 SOM 还要更窄:这些项目中愿意在既有软件栈上增加 AI 层的那一部分。证据不支持精确的 PhysicsX 专属 TAM 点估计,这种不确定性应被保留。[CM019, CM020, CM021, CM022, CM023, CM024]

TAM / SAM / SOM 或规模测算视角表
发布方年份地域数值 / 指标CAGR方法置信度局限
Mordor Intelligence2026全球2026 年 USD 15.46B;2031 年 USD 28.59B13.08%宽口径仿真软件市场模型比核心 CAE 更宽;包含相邻仿真类别。
Grand View Research2024 / 2030全球2024 年 USD 23.56B;2030 年 USD 51.11B14.0% (2025-2030)宽口径仿真软件市场模型通过 archive 获取;若使用 2026 年点位,需要插值。
Business Research Insights2026全球2026 年 USD 11.53B;2035 年 USD 23.81B8.39%CAE 仿真软件市场模型方法透明度有限。
Global Growth Insights 市场报告2026全球2026 年 USD 7.64B;2035 年 USD 15.13B7.9%窄口径 CAE 市场模型口径可能比更宽的仿真软件报告更窄。
SEMI2026全球半导体2026 年晶圆厂设备支出 USD 130B 至 USD 133B2026 年同比增长 18%半导体资本开支可作为重物理设计与验证需求的代理指标这是资本开支代理指标,不是仿真软件收入。
IEA2025 / 2026全球数据中心2025 年大型科技公司资本开支超过 USD 400B,2026 年预计增长 75%;2025 年数据中心用电量增长 17%n/a能源与基础设施需求代理指标这是能源和资本开支代理指标,不是软件 TAM。
本报告(约束性框架)2026仅限与 PhysicsX 相关的细分市场没有可支撑的单点 TAM;SAM 是高价值、仿真驱动工程工作流的一个子集;SOM 还要更窄n/a边界逻辑锚定在行业和工作流证据上可用于尽调,不适合精确估值计算。

本表有意混合直接市场规模估算和相邻需求代理指标,因为没有公开出版方把工业工程中的物理 AI 作为独立市场来清晰测算。

[CM019, CM020, CM021, CM022, CM023, CM024]
FM002: 市场估算区间

已审阅发布方给出的 2026 年市场区间很宽,取决于它们指的是狭义 CAE,还是更广义的仿真软件。

Grand View 的 2026 年点位,是根据其发布的 2024 年规模、2030 年规模和注明的 2025-2030 CAGR 插值得出;图表旨在展示口径驱动的分散度,而不是共识点估计。

[CM019, CM020, CM021, CM022, CM023, CM024]

2.3 买方、用户、付款方和行业采纳路径

最可信的买方图谱,穿过那些物理性能、可靠性和上市时间重要到足以支撑重仿真支出的行业。多份市场研究中,汽车仍是单一最大建模细分;航空航天与国防、半导体和先进电子则对多物理场和系统级验证有特别强的需求。PhysicsX 自己的材料和第三方故事显示,用例覆盖半导体设备开发、热设计和航空航天几何探索。实际日常用户是仿真专家、CAE 工程师、设计工程师,以及越来越多使用 copilots 或 surrogate 工具的更广泛产品团队。经济买方通常是资深工程负责人、CTO 办公室、产品开发预算负责人或垂直项目负责人,而不是泛 AI 预算。采纳路径也因细分而异:航空航天要求可追溯到求解器的验证;半导体买家关注封装和热复杂性;汽车买家强调虚拟验证和周期压缩;数据中心或电子项目则聚焦热、电力和基础设施约束。[CM002, CM010, CM011, CM014, CM018, CM025]

细分市场 / 买方图谱
细分市场买方用户付款方工作流预算负责人采用触发因素
航空航天与国防OEM 或主承包商工程组织空气动力、结构和系统工程师项目或先进工程预算概念探索、面向认证的仿真、数字任务工程工程副总裁 / 项目总工程师需要在不牺牲可追溯性和安全裕度的前提下压缩迭代。
半导体设备与电子设备制造商、芯片或封装团队、电子 OEM热、封装、PCB 和可靠性工程师研发或平台工程预算芯片-封装-系统热设计、先进封装、电子可靠性CTO、产品开发或封装负责人AI 芯片复杂度和热密度抬高了慢迭代的成本。
汽车与出行整车 OEM、一级供应商、自动驾驶项目CAE、电池、热、碰撞和控制工程师整车平台或数字工程预算虚拟验证、轻量化、电池安全、自动驾驶测试总工程师 / 车型线高管需要更快通过型式认证,降低原型成本,并跑更多 EV / 自动驾驶迭代。
数据中心与基础设施硬件服务器、冷却、电力或基础设施设计团队热、电力和系统工程师基础设施或产品工程预算冷却、气流和功率密度优化产品副总裁 / 基础设施工程AI 负载快速增长,功率包络也很紧。
工业机械 / 材料 / 能源工业 OEM 或流程运营商流程、热、机械和控制工程师卓越运营或资本开支预算设备优化、流程建模和数字孪生验证工厂技术或工程高管能耗、良率或停机成本足够高,足以支撑基于模型的优化。

公开证据同时显示既有仿真支出和在现有工作流之上加一层 AI 的可信需求时,本买方图谱才把这些位置纳入。

[CM002, CM010, CM011, CM018, CM025, CM026]
FM003: 买方 / 细分市场图

PhysicsX 卖给仿真强度本来就高的行业,但不同垂直领域里的用户、出资方和验证负担并不一样。

[CM010, CM018, CM025, CM028, CM029, CM030]

2.4 增长驱动、采纳约束和结构性矛盾

增长逻辑真实,但有条件。PhysicsX 受益于清晰的需求侧推动:surrogate 模型和 AI 增强工作流可以大幅压缩迭代时间;与 AI 芯片绑定的半导体资本开支仍高;数据中心建设迫使企业做出更复杂的热和电力决策;航空航天项目复杂度上升,但专家劳动力不足。云原生交付也降低了部分买家的门槛。但约束同样重要。成熟工程组织大多仍卡在试点阶段,调研证据显示,数据准备度、治理和互操作性比热情更能阻碍采纳。信任也没有解决:多数组织只在明确监督下允许 AI 驱动的通过 / 失败决策,风险规避行业仍会把 AI 输出与可信求解器和最终验证对照。算力和基础设施成本仍有意义,互操作性缺口保留了沉没成本锁定,出口管制叠加 AI 基础设施瓶颈让部署更复杂。Synopsys/Ansys、Siemens/Altair 和 Cadence 相邻资产的整合进一步抬高了创业公司成为 system-of-record 软件的门槛,即便这也为 AI 原生叠加层创造空间。[CM007, CM017, CM028, CM029, CM030, CM031]

增长驱动因素与约束表
驱动因素 / 约束方向时间含义尽调问题
替代模型与工作流 AI正向当前可把迭代周期从数小时或数天压到接近数秒,并扩大设计空间探索。哪些客户工作流已经用 AI 做筛选,而不是最终签核?
云原生与浏览器交付正向当前至中期买方不必都自建固定集群,也能更广泛地使用高性能仿真。PhysicsX 的采用在多大程度上取决于客户云政策和数据驻留要求?
半导体资本开支浪潮正向当前AI 芯片和封装复杂度强化了对热、可靠性和多物理场工程的需求。哪些半导体子工作流会最先转化为有预算的软件需求?
数据中心电力与散热瓶颈正向当前至中期功率密度和冷却权衡更复杂,抬高了更快设计迭代的价值。PhysicsX 卖进的是设施级设计、部件设计,还是运营优化?
航空航天复杂度与劳动力稀缺正向当前工程团队需要工具,让更少专家在可追溯前提下评估更多变体。哪些证据能证明 AI 输出已被受监管审查关口接受?
切换成本与沉没许可证负向当前现有平台让既有工作流保持粘性,也迫使初创公司走叠加集成路线。哪些集成能让 PhysicsX 成为增量能力,而不是要求客户推倒重来?
验证、信任与监督负担负向当前风险厌恶行业仍需要与可信求解器做基准对比,并控制部署范围。哪些独立证据能显示接近求解器的精度已在生产环境成立?
互操作性、治理与数据准备度负向当前工程数据杂乱、工具链不兼容,会拖慢从试点到生产的扩展。推广过程中,客户要承担哪些数据和模型治理负担?
计算、出口管制与基础设施瓶颈负向当前至中期AI 基础设施仍受芯片管制、电力、电网和硬件可得性约束。PhysicsX 对先进计算限制或 GPU 稀缺有多大暴露?
现有厂商整合混合当前至中期大平台拿到了分发和信任,但整合也给专用 AI 叠加层留下空白地带。PhysicsX 能否比现有厂商复制功能更快地切入大平台技术栈合作?

本表把每个驱动因素或约束都连到时间和尽调含义,而不是假设 AI 采用加速就会自动转化为即时软件支出。

[CM007, CM017, CM028, CM029, CM030, CM031]
FM004: 采用漏斗或价值链图

最合理的购买路径,是用 AI 加速前段流程,同时在后段工作流里保留高保真验证。

[CM007, CM017, CM036, CM038, CM047, CM048]

2.5 规模测算矛盾和应保留的尽调缺口

市场章节应明确保留证据尚无法解决的问题。第一,没有针对「工业工程物理 AI」且方法可审计的独立公开市场估算,因此任何精确的 PhysicsX TAM 都是伪精确。第二,即便半导体 capex 或数据中心扩张等相邻信号很强,如果不知道有多少份额流向仿真工作流,而不是硬件、服务或内部工程劳动力,也无法干净转化为软件 SAM。第三,当前买方证据仍指向叠加层采纳,而非替代求解器,意味着 SOM 很大程度取决于集成、验证和变革管理,而这些大多是私下信息。最后,已发布估算冲突足够大,投资人应把它们视为边界标尺,而不是共识。这些矛盾有用:它们揭示了物理 AI 一旦获得信任,公司可能超预期的地方,也揭示了保守工作流和既有生态继续主导时,采纳可能保持狭窄的地方。[CM023, CM024, CM044, CM045, CM046, CM047]

Chapter 03

03竞争对手

3.1 格局和解决方案类别

PhysicsX 竞争的不是一两家 AI 原生创业公司。买家至少可以通过六条路径解决同一任务:Synopsys/Ansys 和 Siemens 的既有 CAE 套件;Cadence 等相邻仿真软件栈;Rescale 等工作流控制平面平台;BeyondMath 等 AI 原生同业;Akselos、Monolith、nTop 等垂直专家;以及基于 OpenFOAM、SU2、NVIDIA PhysicsNeMo 的内部自建路径。这一点重要,因为采购很少从白纸开始。大型工程团队已经拥有求解器席位、脚本、验证流程和 CAD/PLM 连接,所以真实替代集合比创业公司同业名单更宽。 PhysicsX 自己的证据也说明,竞争框架必须包括替代方案。平台接入既有工具,而不是要求 rip-and-replace;案例研究使用 OpenFOAM 生成的求解器数据;GTM 倚重嵌入式工程师和合作伙伴渠道。换句话说,买家常常是在决定是否把 AI 原生加速层加到已安装的仿真软件栈上,而不是决定是否抛弃仿真这一类别。这个框架让开源和内部自建的重要性远高于典型 SaaS 竞争页面所暗示的水平。[CP001, CP002, CP004, CP010, CP011, CP016]

竞争对手画像表
竞争对手 / 路径类别规模或牵引信号目标买方差异化局限
PhysicsXAI 原生物理平台2026 年 Series C 轮估值约 $2.4B;收入和客户数同比增长超过一倍航空航天、汽车、半导体、能源、材料领域的工业工程团队面向客户定制的大型物理模型,加上嵌入式部署公开定价不透明,公开认证细节也比现有厂商更薄
Synopsys / Ansys现有套件厂商2025 年后形成从硅到系统的合并技术栈,并有明确 AI 路线图大型企业仿真和半导体客户装机基础、广工作流覆盖、捆绑能力AI 姿态仍叠在经典套件之上,集成还在推进
Siemens Simcenter + Altair现有套件厂商多领域产品组合,包含 Simcenter X 和 Altair units 模型NX / Teamcenter 客户、工业集团、受监管工程团队深度数字主线锁定、tokens / units 灵活性、宽仿真传承在共同客户中也可能成为 PhysicsX 的守门人或替代者
Cadence Fidelity相邻现有厂商从电子 / EDA 基础延伸出的强 CFD 与 GPU 计算叙事航空航天、汽车、涡轮机械、电子热 / EM 买方GPU 加速和软硬件集成PhysicsX 式客户定制大型物理模型的证据较少
Rescale工作流控制平面同类数百家企业客户,平台年化 HPC 支出超过 $1B管理大型仿真资产的研发负责人HPC 编排、AI 代理、广合规姿态不掌握最深的专有求解器或物理模型护城河
Monolith AI相邻专业厂商测试数据 AI,具备汽车背景和 NAFEMS 能见度验证、测试和产品性能团队基于真实测试数据的自学习模型和主动测试规划当前更贴近测试 / 验证,而不是直接面向多物理场仿真
nTop上游工作流专业厂商成熟的计算设计平台,已用于先进工程设计和几何自动化团队位于仿真上游的设计空间探索和几何自动化不是直接求解器,也不能替代替代物理模型
Akselos垂直专业厂商实时结构性能管理,且有量化能源成效关键基础设施、油气、LNG、海上风电运营商领域专用结构 ROI 和运营集成威胁最强的地方是狭窄垂直领域,而不是广义工业 CAE
BeyondMathAI 原生直接同类新融资,加上 Formula 1 和航空航天验证点需要快速设计探索的高性能工程团队基础物理定位,速度主张很强商业成熟度和定价在公开层面仍不够清晰
开源 + 内部自建现状 / 替代方案OpenFOAM、SU2 和 PhysicsNeMo 免费且持续维护成熟 OEM 和内部 HPC / ML 团队最大控制权、无软件加价、复用现有求解器数据需要稀缺人才、MLOps 和部署加固,很多团队不具备

各行归纳了严肃 PhysicsX 买方在 2026 年最可能纳入决策的替代项,包括替代方案和内部自建。

[CP001, CP005, CP010, CP011, CP014, CP015]
FP001: 竞争定位图

这是一张序位图,用来比较最重要替代路径里的分发杠杆和 AI-native 物理优势。

坐标轴是基于产品范围、渠道触达和公开部署姿态推导的证据支撑序位分数,不是已发布的第三方排名。

[CP009, CP010, CP011, CP016, CP022, CP023]

3.2 能力、信任和分发对比

核心比较不只是「谁有 AI」。既有厂商仍在广度、嵌入式工作流和采购肌肉上占优。Synopsys/Ansys 和 Siemens 可以把 AI 功能打包进更大的仿真、CAD、PLM 和 EDA 资产;Cadence 则从 GPU 加速的经典仿真切入。Rescale 从另一个角度竞争:它把仿真执行、AI agents 和算力经济性包成一个控制平面,可以架在多种求解器选择之上。PhysicsX 的反向定位,是在困难工业物理问题上做更深的 AI 原生加速,并配套客户特定微调和更紧密的工程工作流。 信任和监管姿态同样不对称。Rescale 公开展示 FedRAMP、SOC 2、ISO 27001 和 ITAR 信号;Siemens 和 Synopsys 继承了几十年的企业与认证可信度;PhysicsX 的公开信号在主权和工业合作伙伴上更强,在外部可读的认证细节上更弱。这不意味着 PhysicsX 在信任上薄弱,但确实意味着,在国防、航空航天、半导体或关键基础设施采购中,买家往往能比面对 AI 原生新进入者时更快通过既有厂商和基础设施供应商的尽调。[CP003, CP006, CP007, CP008, CP009, CP011]

功能 / 能力矩阵
购买标准PhysicsX现有套件RescaleBeyondMath开源自建垂直专业厂商
客户定制微调闭环部分部分不一不一
广泛求解器与工作流资产低-中
数量级提升的替代模型推理不一
嵌入式工程交付
公开合规可见度
买方灵活性 / 自建选项

评分是基于已审阅语料的有证据序位判断;未知或证据较弱的领域有意不夸大。

[CP001, CP003, CP009, CP011, CP013, CP017]
信任 / 切换成本比较
维度PhysicsXSynopsys / AnsysSiemens SimcenterRescale含义
工作流嵌入与现有工具集成,并嵌入工程师深厚的传统装机基础和 EDA / CAE 脚本深度 CAD / PLM / CAE 数字主线控制平面可位于多种工具之上PhysicsX 可以在不推倒重来的情况下落地,但现有厂商仍掌握周边工作流
公开合规可见度主权云和合作伙伴信号公开长期企业级 / 合规声誉长期企业级 / 合规声誉FedRAMP / SOC 2 / ISO 27001 / ITAR 明确披露受监管买家用既有厂商和 Rescale 做尽调,流程能更快通过
许可证池锁定未公开暴露多产品打包带来议价杠杆token 加 Siemens 更广资产平台粘性绑定作业历史和治理PhysicsX 必须靠更快产出取胜,不能只靠许可证灵活性
合作伙伴依赖对 Siemens、NVIDIA、Deutsche Telekom 依赖高较低;它们掌握更多技术栈低;它们掌握更多技术栈中等,主要在超大云厂商和软件生态合作伙伴渠道扩大触达,但也可能压住战略自由度
内部自建替代中等风险,因为平台层可以被部分复制较低,因为覆盖面极广较低,因为覆盖面极广中等,因为超大云厂商可能绕开中间层相比最宽的套件,PhysicsX 和 Rescale 面临更可信的自建压力
采购熟悉度新兴供应商极高极高HPC 重度账户中较高熟悉的采购肌肉仍是既有厂商的真实优势

最高切换成本来自工作流所有权和采购熟悉度,而不只是某一个求解器模型。

[CP002, CP006, CP007, CP010, CP011, CP018]
FP002: 功能广度 / 能力图

从能力视角看,买方评估仿真加速时,PhysicsX、套件和替代方案的差异主要落在哪里。

这些取值来自对已审阅来源集的分析判断,并刻意把工作流广度与 AI-native 模型强度拆开。

[CP003, CP013, CP017, CP018, CP023, CP027]

3.3 定价、切换成本和内部自建替代方案

在这个格局中,公开定价披露是例外,不是常态。Siemens 和 Altair 解释 token 和 unit 机制,但不公布费率。PhysicsX、Rescale、Monolith 和 BeyondMath 在公开材料中都像销售主导模式。因此,最有用的定价信号不是标价,而是打包结构和买方经济性。PhysicsX 本质上要求买家为速度、迭代能力和工作流压缩付费;既有厂商要求买家留在更广的许可证池内;开源路径则要求买家吸收人才和集成成本,而不是软件加价。 因此,内部自建路径是真实存在的,尤其适合已经拥有求解器经验和 GPU 预算的成熟工程组织。OpenFOAM 和 SU2 仍然免费、成熟且技术可信;PhysicsNeMo 则在 PhysicsX 瞄准的同一批广泛物理族上开源。PhysicsX 自己的案例研究强化而非削弱了这一结论,因为它们显示产品训练于精英内部团队也可以生成的求解器输出。公司最好的防御不是内部自建不可能,而是对多数买家来说,搭建带有专有数据管线、不确定性管理、部署加固和持续模型运营的工业级软件栈,仍然又贵又慢。[CP004, CP012, CP014, CP019, CP024, CP025]

定价 / 包装比较
路径价格 / 合同模型公开可见内容主要未知对买方的含义
PhysicsX定制企业合同无公开费率卡;叙事是咨询式平台 + 部署实际 ACV、使用量指标和续约经济性必须直接尽调定价权,不能从网站倒推
Synopsys / Ansys企业套件订阅AI 产品公开;定价仍由销售主导AI 模块如何相对传统捆绑包定价捆绑能力可能比求解器标价更重要
Siemens Simcenter命名用户 + 浮动 tokenstoken 机制公开,但 token 费率不公开每个工作流或每个计算小时的有效成本灵活许可降低了既有 Siemens 客户的摩擦
Altair Units可在 180+ 产品间共享的 units 池units 模型公开;定价仍需联系销售真实 unit 经济性和跨产品组合折扣广泛 unit 池可把多种工具压进一份合同,从而提高切换成本
Rescale企业平台 + 使用量经济性定价仍不透明;财务控制和版本公开转嫁利润率与软件加价之间的差异控制平面价值与计算治理混在一起,因此很难按席位与求解器厂商逐项比较
Monolith AI演示驱动的企业销售无公开定价部署规模和支持成本最适合视为卖进特定验证痛点的项目或平台销售
BeyondMath早期企业 / 项目访问融资和案例研究主张公开;定价不公开产品是否已有标准化包装商业成熟度仍是尽调问题的一部分
开源自建无软件许可证,但有人才和基础设施成本OpenFOAM 和 SU2 免费;PhysicsNeMo 开源真实内部工程和 MLOps 负担软件便宜不等于总成本低;缺少专业人才时仍会很贵

本表强调包装机制,因为整个竞争集的公开标价都很少。

[CP012, CP014, CP019, CP024, CP025, CP026]

3.4 护城河耐久性和不利信号

PhysicsX 的护城河真实存在,但有条件。最强论点是,它把专有工业数据、客户特定微调、深度集成和前置交付整合进一个产品化工作流。对通用既有厂商或内部团队来说,在定制化、高价值多物理场工作流上快速复制这一点并不容易。较弱版本的护城河叙事是「只有 PhysicsX 能构建这些模型」,因为 PhysicsX 自己的合作伙伴策略和 NVIDIA 的开放工具,都会让这一说法随时间变得不那么站得住。 因此,不利情形可信,应在尽调中保持突出。Siemens 可以从渠道变成替代者。Synopsys/Ansys 可以把 AI 打包进更大的装机基础。Rescale 可以赢下编排层。OpenFOAM、SU2 和 PhysicsNeMo 保持 DIY 路径可行。垂直或上游专家也可以在不匹配完整平台广度的情况下,分走工作流切片。PhysicsX 仍可能跑赢这个领域,但其耐久性取决于同时守住数据优势、工作流速度和部署信任,而不是假设其余软件栈原地不动。[CP008, CP018, CP020, CP022, CP029, CP032]

护城河耐久度 / 竞争风险登记表
护城河主张威胁严重性为什么可信尽调或缓释
专有工业数据和微调开源求解器 + PhysicsNeMo 内部自建模型层越来越开放,买家本来就拥有求解器数据验证参考客户留下来,是因为数据飞轮和部署负担,而不是因为模型无法复刻
合作伙伴驱动分销Siemens 从渠道变成替代者Simcenter 集成可能同时加速采用和未来替换在战略合作中厘清排他性、数据边界和账户控制规则
嵌入式交付质量高接触 GTM 带来的可扩展性压力中高前置部署工程师很强,但吃人力衡量每名部署工程师收入和参考客户扩张效率
AI 原生速度优势既有厂商把“够用”的 AI 加进更大的套件Synopsys/Ansys 和 Siemens 已经在营销 AI 辅助工作流验证 PhysicsX 是否仍能交付数量级提升,足以支撑买家引入第二家供应商
工作流宽度对比局部工具预算分散到 Akselos、Monolith、nTop 等专业厂商专业厂商能赢下窄但高价值的切片按工作流切片梳理丢单,而不只按头部具名竞争对手归因
信任姿态公开认证细节仍比 Rescale 或大型套件更薄采购可能偏好公开合规界面更明确的供应商尽调早期索取安全资料包和受监管客户证明

严重性衡量未来 12-24 个月 PhysicsX 被替代或利润率承压的风险,而不是长期生存概率。

[CP008, CP018, CP027, CP029, CP031, CP034]
FP003: 护城河 / 就绪度 KPI

这张紧凑记分卡概括 PhysicsX 当前相对套件、替代方案和合作伙伴驱动替代风险的耐久性。

这些取值概括了从已审阅语料得出的竞争耐久性判断;它们不是外部发布的分数。

[CP029, CP032, CP034, CP038, CP039]

3.5 图表

Chapter 04

04财务

4.1 收入来源、定价模型和 GTM 动作

PhysicsX 的公开材料指向协商式企业模式,而不是透明 SaaS 标价业务。公司销售一个 AI 原生工程平台,覆盖仿真管理、模型开发和部署到工程工作流;但它也称前置工程师会直接嵌入实时客户项目,并按客户场景定制模型架构、数据管线和优化循环。这个组合强烈暗示,收入来自平台订阅或许可证、实施和模型开发工作,以及部署后向更广工作流所有权扩张的混合体。唯一公开的产品主导定价线索是 Ai.rplane,这是 LGM-Aero 工具的免费基础版本,看起来更像漏斗顶端获客界面,而非有意义的收入线。由于没有官方价目表、最低承诺或合同期限细节公开,定价纪律和收入确认时点仍是尽调事项,而不是可承销事实。[CI001, CI003, CI007, CI008, CI010, CI011]

收入流表
收入流机制单位 / 合同基础当前公开状态收入质量判断尽调要求
企业平台部署横跨生命周期的核心 AI 原生工程平台协商式企业合同官方平台页面明确活跃如果复用占主导,可能具备扩展性索取样本 MSA 和收入确认政策
前置部署工程嵌入客户项目交付和工作流定制可能是限定范围服务或打包实施官方披露为嵌入式动作提高赢单率,但可能稀释软件利润率索取交付附加率和实施利润率
定制模型开发 / 微调定制模型架构、仿真管线、优化循环项目或扩张 SOW官方披露活动,定价未公开ACV 更高,但可重复性不清楚索取 SOW 示例和续约路径
仿真和数据编排Simulation Workbench、数据血缘、模型训练环境平台席位、用量或项目费未披露产品模块公开,变现方式未公开可能形成粘性的记录系统经济性索取 SKU 图谱,以及模块是否单独销售
主权 / 合作伙伴云部署合作伙伴计算基础设施上的 PhysicsX 应用层可能是平台费加托管部署经济性Deutsche Telekom 和 NVIDIA 已有部署证据战略分销,但云经济性未知索取合作伙伴计费结构和利润分成
产品驱动的漏斗入口Ai.rplane 基础公开工具免费访问发现的唯一公开免费产品线索获取,不是实质收入索取免费工具到付费账户的转化漏斗

各行区分明确公开的信息和推断信息。PhysicsX 未披露收入流级别的收入结构,因此收入质量结论仍是方向性判断。

[CI001, CI007, CI011, CI013, CI025]
定价 / 变现表
产品 / 层公开价格观察到的合同单位实际已知信息关键未知项来源或尽调路径
核心企业平台未披露可能是年度或多年期企业合同官方界面未找到公开价目表最低承诺、期限和扩张触发条件审查近期 MSA / 订单表
前置部署交付未披露可能是限定范围实施或打包交付关于页面确认现场项目中有嵌入式工程师是单独计费,还是在平台 ACV 内补贴审查发票上的服务行项目
定制模型开发未披露可能按项目或里程碑计费平台 FAQ 确认可做客户特定定制复用经济性对比一次性经济性审查 SOW 和复用政策
合作伙伴 / 主权云部署未披露平台费与基础设施转嫁的组合未知Deutsche Telekom 和 NVIDIA 已有部署证据利润分成、市场费、预留容量义务审查合作伙伴协议经济条款
Ai.rplane / LGM-Aero 入口免费基础版面向公开用户免费访问The Next Web 报道 Ai.rplane 免费转化为付费企业部署的路径审查漏斗指标和后续 ACV
模块级打包未披露模块是否单独销售未知平台页面列出不同工作台和服务SKU 粒度和附加率索取产品 / 价格目录

这张表刻意呈现定价不透明度,因为公开记录只透露打包线索,没有透露实际定价或折扣。Null 已替换为明确的“未披露”措辞,避免尽调要求含糊。

[CI010, CI011, CI047]
FI001: 收入模式桥

公开证据显示,PhysicsX 会把试点需求转化为平台加交付的混合项目,再扩展到更广的工作流所有权,而不是卖一个简单的自助式 SKU。

[CI001, CI007, CI008, CI013, CI025]

4.2 收入质量、需求信号和公开牵引

公开牵引强于多数私营工业 AI 创业公司,但仍不完整。官方 Series C 披露称,已确认收入同比翻倍,预订收入增长两倍,客户数增加一倍以上,员工数超过 300。外部报道补充称,2026 年收入应接近 $50 million,目标是在 2027 年翻一倍以上,且公司拥有约六个月的客户需求 backlog。具名客户包括 Applied Materials、Siemens 和 Stellantis;与 Deutsche Telekom 和 NVIDIA 的主权云部署证据,支撑了 PhysicsX 正在落地真实工业工作负载,而不是只跑试点项目。收入质量的核心细微点在于,预订收入增长快于已确认收入。如果实施变得更可复用,这可能是利好;但也可能意味着交付能力和客户特定工作仍是瓶颈。[CI013, CI014, CI015, CI016, CI017, CI018]

FI003: 财务估算区间

最强的公开数字锚点是管理层的收入目标和累计资本基础,而不是利润率或 runway。

公开来源只给出单一披露金额或下限时,精确数字以点状区间展示。

[CI019, CI020, CI035, CI036]

4.3 成本结构和毛利率驱动因素

从成本结构看,PhysicsX 不像一家纯软件公司。官方产品和合作伙伴材料显示,其软件栈依赖 GPU 加速模型训练和推理、大量仿真数据生成、客户特定工作流集成,以及在客户项目内部工作的交付团队。Deutsche Telekom 与 NVIDIA 的主权云合作显示了公司想接入的基础设施规模:底层云包含 1,000 多套 DGX B200 系统和最多 10,000 个 Blackwell GPUs。汽车案例研究进一步凸显了部署前可能的成本负担:用 250 多个基线设计生成训练数据时,使用了 20,000 多次 CFD 仿真。同时,PhysicsX 的平台 FAQ 强调不确定性量化、主动学习和客户特定数据隔离,这些都意味着部署后还需要持续验证和维护。已审阅的公开来源没有披露毛利率,也没有拆分可复用软件毛利与前置交付劳动力。[CI024, CI025, CI026, CI027, CI028, CI029]

单位经济性表
指标公开数值 / 状态置信度为什么重要尽调要求
已确认收入增长同比翻倍最好的官方收入质量信号绑定实际美元基数和收入确认政策
预订收入增长同比增至三倍显示需求形成早于收入确认按季度把预订收入桥接到已确认收入
客户数增长增长超过一倍支撑客户基础拓宽,而不只是单一客户增长索取新增客户名单和头部账户集中度
积压订单约 6 个月需求积压暗示产能约束,以及可能提前拉动营运资本索取已预订积压的账龄和预期转化日期
员工规模官方 300+;第三方报道约 350人力强度是毛利率的主要决定因素索取 R&D、交付、S&M、G&A 的职能拆分
毛利率未披露判断估值是否匹配软件经济性的关键按收入流和客户类型索取毛利率
CAC / 回本期未披露判断企业销售效率所必需按细分市场索取销售漏斗、赢单率和回本期
NRR / 流失未披露决定扩张是否持久,还是项目制驱动索取客户队列留存和续约瀑布表
服务与软件组合未披露影响利润率和估值质量的最重要驱动因素索取把平台和服务拆开的收入桥

只有前五行有公开证据支撑。其余都是明确的私有指标缺口,并直接影响承销判断。

[CI015, CI016, CI021, CI023, CI031, CI032]
FI002: 单位经济桥

公开可见的主要成本桥,从仿真数据生成和 GPU 基础设施一路延伸到交付人力,最后落到未知的毛利。

这座桥是定性的,因为 PhysicsX 没有披露毛利率、托管成本或 COGS 中交付人力的占比。

[CI025, CI026, CI027, CI028, CI029, CI030]

4.4 资本充足性和融资依赖

PhysicsX 资金强于多数同业,但资本需求也大于简单企业软件故事所暗示的水平。公开来源显示,公司在 2025 年 6 月完成 $135 million 初始 Series B,2025 年 11 月的延期融资把 Series B 总额推至超过 $155 million,估值接近 $1 billion;2026 年 6 月又以约 $2.4 billion 估值完成 $300 million Series C。二手来源将总融资额放在约 $487 million 至 $500 million。Companies House 记录还显示,2025 年和 2026 年有多次股权表更新,包括截至 2024 年的集团账目,以及 2026 年 2 月和 4 月新的 SH01 配股活动。Series C 资金用途很明确:全球扩张、加深美国布局、开设 Singapore、建设平台能力,并继续对更大物理模型做前沿研究。缺失的是实际现金余额、烧钱速度、runway 或任何类似债务的算力承诺。因此,融资依赖是运营性而非生存性:如果 backlog、招聘和算力强度继续比可复用软件经济性上升更快,即便需求强劲,下一轮也可能被提前拉来。[CI021, CI033, CI034, CI035, CI036, CI037]

资本充足性表
项目数值 / 状态截至为什么重要来源 / 尽调备注
Series B 初始轮$135M2025-06在规模化年份之前确立增长融资官方 Series B 新闻稿
Series B 延展后总额估值接近 $1B,融资 >$155M2025-11显示 Series C 前的估值上台阶,并加入 NVIDIA 风投支持官方延展公告加 MarketScreener
Series C 轮估值约 $2.4B,融资 $300M2026-06最大现金注入,也是当前估值锚官方 Series C 公告和外部报道
公开披露累计融资~$487M-$500M2026-06勾勒达到当前规模所需资本量CB Insights 和 tech.eu
资金用途美国扩张、新加坡办公室、平台能力扩张、更大的物理模型2026-06解释新资本预计会消耗在哪里官方 Series C 和 tech.eu
英国申报证据2024 年集团账目,以及 2026 年 2 月 / 4 月新的 SH01 配股2026-06 审查日期为持续股权结构表活动提供独立证据Companies House 申报历史
手头现金未披露Series C 后跑道分析的关键缺失输入索取最新资产负债表
烧钱 / 跑道未披露2026无法判断需求是自我供血,还是依赖融资轮索取月度烧钱额和预算
债务 / 计算承诺未公开披露2026隐性的预留容量承诺可能像债务一样运作审查云 / GPU 预留协议

已融资金额清楚;资本充足性并不清楚。缺失项是交割后现金余额、烧钱曲线,以及任何类似债务的计算承诺。

[CI033, CI034, CI035, CI036, CI038, CI039]
FI004: 资本强度 / 现金流图

公开证据指向一家需求很强、但人力和算力强度也真实存在的公司,因此 Series C 后的现金规划仍然重要。

[CI021, CI026, CI027, CI040, CI041, CI043]

4.5 财务结论和尽调阻塞项

财务图景足以支持严肃尽调,但还不足以仅凭公开证据承销利润率或资本效率。增长、客户质量和融资可信度都是真实的:具名工业客户存在,主权云合作伙伴公开,管理层披露了私营公司少见的牵引指标。弱点在于,几乎所有能区分可扩展软件平台与技术服务组织的承销变量仍是私下信息。已审阅的公开来源没有披露 ARR、经常性收入与服务收入占比、毛利率、CAC、回本期、NRR、现金或 runway。这也是为什么不利情形重要。New Market Pitch 认为,最新估值隐含约 48x 当年收入,核心未解问题是 PhysicsX 正在成为可复用软件层,还是停留为服务密集型工程公司。因此,正确的尽调路径很直接:在接受溢价估值的表面数字前,验证实际定价、收入组合、毛利率、cohort 留存和 Series C 后现金规划。[CI031, CI032, CI041, CI042, CI043, CI044]

公开财务缺口表
缺失的私有指标当前公开状态对承销的影响精确尽调路径
ARR 和经常性收入组合未披露无法判断多少收入是真正的软件经常性收入按产品和客户索取月度经常性收入桥
毛利率和 COGS 拆分未披露无法判断 PhysicsX 是像软件还是像服务一样扩张索取管理层 P&L,包含计算、人力和服务 COGS
实际定价和折扣未披露公开打包线索看不出 ACV 纪律或回本期审查覆盖试点、扩张和战略账户的三份近期合同
现金、烧钱和跑道未披露Series C 后资本充足性仍未证明审查最新资产负债表、13 周现金流和 FY2026 预算
NRR、GRR、流失和集中度未披露增长质量无法与项目赢单拆开索取客户队列分析和头部客户清单
类债务计算或预留容量义务公开未披露潜在表外承诺可能压缩灵活性审查云预留、GPU 和主权云容量协议

仅靠开源证据,无法承销收入质量、利润率路径和融资依赖;这些就是明确堵点。

[CI031, CI032, CI041, CI045, CI046]
Chapter 05

05产品与技术

5.1 产品定义和工作流界面

PhysicsX 并不把自己呈现为单一 surrogate 模型或一次性仿真器。公开产品故事是一套分层工程软件栈:先做仿真和数据编排,再加入物理 AI 模型开发,最后落到面向工程师、适配既有产品开发工作流的应用。按客户语言,公司卖的是更快的概念探索、由求解器信息支撑的优化,以及跨设计、制造和运营的 AI 辅助运营决策支持,而不只是一个模型 API。今天公开可见的最具体模块包括 Simulation Workbench、AI Workbench、Engineering Applications、平台服务,以及 LGM-Aero 和 Ai.rplane 这一公开展示组合。工作流很明确:摄取 CAD、仿真和运营数据;构建可追溯数据集;训练或微调物理模型;通过应用、APIs 或边缘场景部署;并让人在验证和决策中保留参与。这个框架重要,因为它让 PhysicsX 更接近工程操作系统,而不是独立 foundation-model 实验室;但它也意味着产品价值取决于集成纪律和交付执行,不只取决于模型质量。[CE001, CE002, CE003, CE004, CE005, CE009]

产品模块 / 资产矩阵
模块 / 资产主要用户交付内容状态 / 成熟度差异化尽调缺口
Simulation Workbench仿真、数据和平台团队可追踪的仿真数据骨干、自动化、编排、血缘在线核心平台模块把孤立的求解器输出变成可复用、可用于 ML 的资产没有公开吞吐量、正常运行时间或模式文档
AI WorkbenchML 工程师、领域专家、模型所有者开发、微调、部署 DPMs/LPMs 和第三方模型在线核心平台模块支持 low-code 和编程式工作流,也支持私有基础模型没有公开基准测试看板或模型目录细节
工程应用工程师、技术人员、制造和运营用户用于推理、优化和决策支持的 Web、API 与边缘应用已上线的核心平台模块把 AI 封装进贴合工作流的界面,而不是只提供模型访问未公开定价、SLA 或应用目录
平台服务 / 部署底座IT、安全和平台买家多云、本地、隔离网络和主权部署模式,并集成 CAE已上线,但依赖合作伙伴买家采用 AI 时可保留既有 CAE/HPC 技术栈安全架构细节大多未公开
LGM-Aero + Ai.rplane 展示应用航空航天用户、潜在客户和内部 GTM面向几何生成、气动预测和优化的公开展示公开参考应用 / 展示样板具体展示 Large Geometry Model 工作流参考应用范围窄于完整工业平台

各行综合公开平台、技术和合作伙伴材料;模块边界来自公司描述,而不是正式公开的 SKU 表。

[CE002, CE007, CE009, CE010, CE019, CE024]
工作流 / 使用场景表
用户任务当前工作流PhysicsX 方案可衡量收益局限
物理产品概念探索按顺序跑 CAD 到求解器循环,等待数小时或数天出结果用平台应用和模型先筛选设计,再重跑求解器公司和合作伙伴称,数小时 / 数天可压缩到数秒收益仍偏方向性;公开 ROI 指标稀少
创建可用于 ML 的工程数据集手工跨工具收集日志、网格和 KPI 导出Simulation Workbench 自动处理血缘、归一化和结构化输出生成可搜索、可复用的训练语料未公开 schema 或连接器文档
构建特定领域的物理 AI从分散的仿真文件训练定制代理模型AI Workbench 用客户私有数据微调预训练 DPM/LPM 资产相比从零训练模型,数据负担更低私有模型治理细节未公开
把推理部署进真实工程运营依赖孤立的求解器团队或离线分析师Engineering Applications 在客户流程中提供 Web、API 或边缘工作流更快做 what-if 分析,并可能优化运营公开披露的生产案例仍有选择性
生产动作前验证 AI信任单一模型输出,或慢速重跑求解器使用不确定性量化、主动学习、参考仿真和工程师复核提高置信度,并有针对性地生成数据验证负担仍因领域而异,公开基准也不完整

收益来自公开声明或合作伙伴轶事,不是覆盖装机基数的标准化 ROI 研究。

[CE005, CE011, CE014, CE018, CE038, CE048]
FE001: 产品架构图

PhysicsX 公开架构从数据编排向上分层,延伸到模型、应用和部署服务。

[CE002, CE003, CE009, CE010, CE011]
FE002: 客户工作流 / 运行流

产品流程从工程数据采集走到模型部署,同时保留求解器支撑的验证和人工评审。

[CE014, CE015, CE017, CE038, CE047]

5.2 架构、运营模型和集成栈

公开架构指向一个相当有主张的运营模型。Simulation Workbench 是数据和编排层:PhysicsX 将其描述为把仿真、实验和运营数据规范化、打标签、版本化,并在几何、网格、配置和结果之间保持可追溯的地方。AI Workbench 则架在该数据骨干之上,作为 Deep Physics Models、私有 foundation models 和第三方模型家族的训练与部署环境。Engineering Applications 是消费层,通过网页界面、APIs、专用边缘或制造部署暴露结果。产品设计成接入标准工程工具链,而不是直接替代它们;公开引用包括 ANSYS、CATIA、Siemens NX、OpenFOAM、STAR-CCM+,以及客户云和 air-gapped 部署模式。PhysicsX 自己的技术写作异常明确地说,这套软件栈不只是模型外面的 UI 包装:它依赖编排、数据血缘、主动学习和反复的求解器支撑验证。这在技术上可信,也适合企业,但也揭示了实施负担:PhysicsX 必须在异构求解器生态、变化的云目标和定制客户环境之间,让抽象层保持稳定。[CE006, CE007, CE008, CE011, CE012, CE013]

技术 / 运营架构表
层 / 组件角色关键依赖公开证据质量主要风险
Simulation Workbench自动化仿真流水线,并保存可追溯数据CAD/CAE 工具、网格流水线、HPC、元数据 schema官方技术说明支撑强工具异质性越高,集成工作量越大
AI Workbench开发、微调并部署 DPM/LPM私有数据、预训练模型、第三方模型族平台和架构文章支撑中高基准可比性大多仍未公开
Engineering Applications把模型输出转成面向工程师的决策Web 应用、API、边缘目标、特定工作流 UX平台和 Microsoft 材料支撑中等运营可靠性和应用级控制未公开
数据血缘与编排底座连接几何、设置、求解、结果和再训练工作流引擎、版本控制、存储、转换服务仿真自动化文章支撑强客户定制系统可能拖慢实施
部署底座运行托管、客户云、本地、隔离网络和主权版本AWS、Azure、CoreWeave、客户 HPC、Deutsche Telekom/T-Systems平台和合作伙伴来源支撑强对合作伙伴的依赖会影响路线图和经济性
验证与交付闭环让模型与工程师、参考仿真和主动学习配套交付工程师、客户 SME、验证数据官方和 CDFAM 材料支撑中高人力密集的验证会限制产品化扩展

证据质量反映公开材料对该层描述的具体程度;“高”仍不等于经过审计或独立基准验证。

[CE006, CE007, CE011, CE012, CE013, CE015]
FE003: 关键依赖图

PhysicsX 的产品既依赖自己的模型,也同样依赖外部 CAE 套件、云平台和联合交付伙伴。

[CE006, CE007, CE015, CE026, CE028, CE029]

5.3 Large Physics Models、部署模式和差异化

PhysicsX 的公开差异化论点,建立在把 foundation-style 模型与工业部署基础设施耦合起来。LGM-Aero 和 Ai.rplane 是最清晰的公开成果:公司和独立报道描述了一个 100M 参数几何模型,训练于数千万个形状以及大量 CFD 和 FEA 数据,能够在几秒内生成并评分飞机概念,而不必等待反复数值求解。展示之外,PhysicsX 主张同一产品模式可推广到以专有数据训练的私有客户 foundation models,由平台处理编排、带不确定性感知的再训练和工作流集成。合作伙伴让这种运营模型更容易部署:Siemens 把 PhysicsX 延伸进企业 CFD 和 CAE 工作流,Microsoft 带来 Discovery 和 Azure 分发,Deutsche Telekom 与 T-Systems 带来欧洲主权工业部署,CoreWeave 带来大规模 GPU 训练。这是强 go-to-market 架构,因为它降低买方摩擦,并把 PhysicsX 锚定在既有企业工具链中。代价是,部分护城河来自生态位置,而非纯技术排他性。扩大触达的同一张合作伙伴网络,也让 PhysicsX 暴露在平台依赖,以及既有厂商在自家套件内产品化相邻 AI 仿真能力的风险下。[CE015, CE019, CE020, CE021, CE022, CE023]

路线图 / 发布 / 开发阶段表
日期 / 阶段功能或里程碑状态含义来源
2024-12 发布LGM-Aero 和 Ai.rplane 亮相公开参考应用已上线为几何和物理基础模型工作流提供可触摸的展示样板PhysicsX、Siemens、Engineering.com 等来源
2024-12 合作里程碑Siemens 深度物理仿真合作公开合作进行中强化数据生成和 CAE 集成叙事Siemens + PhysicsX 发布
2025-05 合作里程碑宣布 Microsoft Discovery 集成和 Azure Marketplace 私有发布合作进行中把平台延伸进智能体和以 Azure 为中心的企业工作流PhysicsX + Microsoft 专题
2025-07 工作流扩展宣布 Simcenter X 合作工作流扩展进行中增加基于 SaaS 的 CFD/HPC 路径,用于 AI 辅助工程PhysicsX 发布
2026-02 部署里程碑平台在 Deutsche Telekom Industrial AI Cloud 上线当前主权部署增加欧洲主权计算运营模式和联合交付模型PhysicsX + Technology Magazine
2026-03 基础设施里程碑宣布 CoreWeave 合作当前训练 / 部署合作为前沿和私有 LPM 增加大规模 GPU 底座PhysicsX + Trending Topics
基准披露2024 年文章承诺发布完整 LGM-Aero 技术论文在已审阅语料中仍未看到造成公开基准细节上的尽调缺口PhysicsX 技术深度解析

路线图项目混合了发布、部署里程碑和仍未披露的缺口,因为 PhysicsX 的公开叙事由发布驱动,而不是按传统产品路线图发布。

[CE019, CE022, CE026, CE028, CE029, CE030]
FE004: 产品成熟度 / 能力图

PhysicsX 的公开产品在编排和部署灵活性上看起来最强,但基准透明度和边缘案例泛化仍没那么成熟。

[CE023, CE033, CE039, CE040, CE041, CE042]

5.4 信任、安全、验证负担和未解技术缺口

与许多 deeptech 创业公司相比,PhysicsX 有更多公开信任材料,但对一家瞄准关键任务工程工作流的公司来说,这套材料仍不完整。正面信号包括 ISO 27001 认证、引用英国数据保护法和 GDPR 的已发布隐私通知、明确的客户数据隔离主张、公开支持托管、客户云和 air-gapped 部署,以及反复强调不确定性量化、主动学习、求解器支撑验证和人类判断。这些都是工业 AI 平台方向上不错的控制措施。未解决风险在于,公开证据仍比营销雄心薄。PhysicsX 自己的技术写作点明了负担为何真实:普通 FNO-style operators 在冲击、严格边界条件、不规则几何、稀缺数据和更高维问题上表现吃力;其 LGM 材料提到有损压缩、latent-space 假设,以及针对小众几何微调的需求。公司还称完整 LGM-Aero 基准论文仍待发布;已审阅公开语料没有说明 uptime 指标、外部评估审计、客户工程数据的企业留存政策,或国防相邻工作负载的出口管制处理。对买家或投资人而言,这意味着平台看起来技术认真,但生产级承销仍取决于围绕基准、治理和特定领域验证的私下尽调。[CE016, CE017, CE034, CE035, CE036, CE037]

信任 / 质量 / 合规表
控制 / 信号状态范围作用缺口
ISO 27001已公开宣布组织层面的信息安全管理展示正式安全管理基线未公开映射到具体产品模块的控制
隐私声明 / GDPR 引用公开 PDF,最后更新于 2022 年 10 月网站和服务中的个人数据处理为买家提供法律 / 隐私起点这不是现代企业 AI 治理白皮书
客户专属模型训练已公开声称跨客户训练隔离降低跨客户模型泄露顾虑未公开保留或删除工作流细节
托管 / 客户云 / 隔离网络部署已公开声称企业部署拓扑选项支持高密级和主权使用场景未公开事故或正常运行时间历史
不确定性量化和主动学习技术材料已有公开描述模型置信度和数据获取闭环改善信任校准,并有针对性地再训练未公开校准指标或验收阈值
人在回路验证反复强调工程师复核加求解器支撑验证限制对纯模型输出的过度信任扩展和采购会比营销暗示的更慢
出口管制 / 国防处理计划未公开描述航空航天与国防、主权云场景对敏感工作负载可能重要公开语料未显示明确出口管制框架

本表区分公开信号是否存在,以及是否足以支撑企业尽调;几项控制方向积极,但对采购门槛高的环境仍不完整。

[CE008, CE014, CE034, CE035, CE036, CE037]

5.5 图表

Chapter 06

06客户

6.1 客户细分很宽,但各垂直领域的证据质量差异很大

PhysicsX 2026 年 6 月融资新闻稿称,平台已部署在航空航天与国防、半导体、工业机械、汽车、能源和材料领域;同一新闻稿称,过去一年客户数增加一倍以上,已确认收入翻倍,预订收入增长两倍。这证明了广泛行业触达和不小的商业动能,但本身并不能说明哪些细分是可复用软件部署,哪些是定制工程项目。最干净的细分证据来自具体工作流:Siemens 的数据中心电力基础设施、GB1 的顶级性能海事工程、Microsoft Surface 的设备热设计,以及一个未具名的采矿与金属项目用于铜提取。半导体也有强证据,因为 Bloomberg 将 Applied Materials 列为客户,Microsoft 描述了半导体原型工作,但公开记录仍缺少 Applied Materials 工作负载的客户侧描述。汽车证据混合:Stellantis 被具名,技术笔记显示 PhysicsX 在思考真实客户汽车设计,Microsoft 和 CoreWeave 都把公司放入汽车工作流,但具名 OEM 用例仍薄。航空航天与国防被反复列出,但公开证据只有 GB1 和一个未具名航空航天客户;已审阅来源没有点名国防客户或合同载体。[CU001, CU002, CU003, CU004, CU005, CU024]

客户分层表
细分市场买家 / 用户 / 付款方公开证明新鲜度缺口
半导体设备买家:设备研发负责人;用户:仿真和产品工程师;付款方:中央工程 / 制造预算Bloomberg 点名 Applied Materials;Microsoft 称 PhysicsX 缩短半导体制造中的原型开发时间高 — 2026 年 6 月和 2026 年 2 月Applied Materials 工作负载、规模和生产状态仍未披露
汽车 OEM 工程买家:车辆开发和气动负责人;用户:CFD、系统和测试工程师;付款方:工程项目Bloomberg 点名 Stellantis;PhysicsX 技术说明称,真实客户设计不同于基准数据集高 — 2026 年 6 月和 2026 年 3 月未公开 Stellantis 案例研究、支出或结果指标
航空航天 / 高性能工程买家:项目工程负责人;用户:气动、结构和仿真团队;付款方:项目预算GB1 嵌入式部署,以及 Observer 报道的未具名航空航天成果高 — 2026 年 3 月和 2025 年 6 月未点名国防买家,也未公开航空航天合同条款
数据中心基础设施买家:基础设施运营商和电力系统所有者;用户:电气和热工程师;付款方:capex / 基础设施预算Siemens Smart Infrastructure 具体配电工作流高 — 2026 年 3 月未披露合同经济性和重复部署广度
材料 / 采矿和金属买家:运营或工艺工程负责人;用户:工艺工程师;付款方:工厂优化预算Microsoft 称 PhysicsX 与一家全球领先企业合作,提高铜提取效率高 — 2026 年 2 月客户身份和实测提升未公开
欧洲工业云渠道买家:T-Systems 和客户 IT/工程负责人;用户:工业 AI 和工程团队;付款方:云 + 转型预算Deutsche Telekom / T-Systems 入驻,加上客户现场团队高 — 2025 年 11 月至 2026 年 2 月渠道收入分成和客户转化未披露
计算 / 平台赋能买家:企业 AI 和基础设施负责人;用户:模型训练和部署团队;付款方:计算和平台预算CoreWeave 和 Microsoft 提供安全部署路径和企业级基础设施高 — 2026 年 2 月至 3 月这些是赋能关系,本身不是直接终端客户支出证明

各行混合了直接客户证明、渠道证明和平台赋能,因为 PhysicsX 面向复杂工业工作流销售,而不是简单的自助 SaaS 模式。公开买家 / 付款方描述根据各来源中的工作流推断。

[CU001, CU005, CU011, CU014, CU019, CU021]
FU003: 客户证明矩阵

不只比较部署具体度,也比较每条公开证明是否有独立确认、留存可见度和参考独立性风险。

定性排序只反映公开证据。高表示已有具体工作负载,且至少有一个交易对手或双来源佐证;留存可见性为「无」表示未找到公开续约或队列数据。

[CU038, CU039, CU042, CU043, CU044, CU048]

6.2 具名客户证据中,Siemens、GB1 和 Microsoft 最强;Applied Materials 和 Stellantis 描述仍薄

最高质量的具名证据是 Siemens。PhysicsX 和 Siemens 共同描述了一个 2026 年 3 月面向下一代 AI 数据中心的实时工作流;Siemens 独立表示,PhysicsX 帮助工程师实时预测 busway 热行为,把过去需要数天的分析缩短到亚秒级迭代。这读起来像活跃的生产级工程部署,而不是投机性试点。GB1 是次强证据:PhysicsX 和 GB1 都表示,公司被部署为英国 America's Cup 挑战的官方 AI Engineering Platform 合作伙伴,在 Portsmouth 有嵌入式工程师,GB1 Head of Design 的交易对手引用也提到更高保真模型和更快迭代。Microsoft 是有意义但更窄的证据点,因为 Microsoft 公开称 PhysicsX 改善了 Surface 设备热行为,并缩短了半导体原型开发时间,意味着内部参考使用,而不只是合作伙伴身份。相比之下,Applied Materials 和 Stellantis 只由 Bloomberg syndication 点名;两者都没有公开案例研究,工作负载没有披露,公开记录也没有说明任一账户处于试点、生产还是更广企业推广。Deutsche Telekom、T-Systems、CoreWeave、NVIDIA 和 Agentic Launchpad program 强化了分发和基础设施可信度,但更适合被归为使能或渠道关系,而非终端客户收入的独立证据。[CU005, CU010, CU011, CU012, CU013, CU021]

具名客户证明表
交易对方垂直公开工作负载状态结果 / 交易对方证明证据新鲜度关键局限
Siemens Smart Infrastructure数据中心基础设施面向下一代 AI 数据中心的配电系统设计和运营生产部署 / 活跃工程工作流Siemens 称 PhysicsX 可实时预测热行为,并把多日分析压缩到不到一秒的迭代高 — 2026 年 3 月未披露合同价值,也未披露在 Siemens 客户群中的覆盖广度
GB1高性能海事 / 类航空航天工程工程平台嵌入第 38 届 America’s Cup 赛队活跃嵌入式部署GB1 设计负责人提到更高保真模型、更低数据成本和更快设计迭代高 — 2026 年 3 月运动工程参考可信,但本身不能代表经常性工业收入
Microsoft Surface电子 / 设备Surface 设备冷却风扇设计的热行为优化已命名的内部用例Microsoft 称,工程师可以测试更多设计变体高 — Feb 2026未披露支出、规模或续约细节
Applied Materials半导体未披露仅为已命名客户Bloomberg 联合稿将 Applied Materials 列为客户高 — Jun 2026无公开案例研究;试点还是生产部署不明
Stellantis汽车未披露仅为已命名客户Bloomberg 联合稿将 Stellantis 列为客户高 — Jun 2026未公开工作负载、结果或部署深度
Deutsche Telekom / T-Systems欧洲工业渠道云上导入和客户现场部署支持合作伙伴 / 渠道,不是终端客户证明DT 和 T-Systems 描述了为工业客户导入云服务,并在客户现场调动小队高 — Nov 2025 to Feb 2026证据支持的是分销和赋能,而非终端客户直接支出
CoreWeave计算基础设施面向私有大型物理模型的安全企业部署路径基础设施伙伴,不是终端客户证明CoreWeave 称,PhysicsX 已在其云上交付生产级物理 AI高 — Mar 2026证明指向基础设施和部署能力,不指向直接应用买方需求
全球矿业 / 金属龙头(未具名)材料提升铜提取效率活跃但未具名的用例Microsoft 称 PhysicsX 正与一家全球龙头合作推进该工作流高 — Feb 2026客户身份和量化提升未披露
未具名航空航天客户航空航天喷气发动机涡轮叶片质保工作流有结果证明,但客户未具名Observer 报道废品率下降 70%中 — Jun 2025客户未具名,限制了参考质量和可重复性分析

状态标签区分活跃部署、仅命名客户和赋能型合作伙伴关系。只有标识、没有工作负载细节的案例,不视为生产证明。

[CU005, CU011, CU012, CU013, CU015, CU019]

6.3 采纳动能仍在当前周期,合作伙伴驱动的扩张路径清晰可见

尽管 PhysicsX 不披露精确账户数和 ARR,但其采纳轨迹已足够公开,能显示真实动能。公司在 2026 年 6 月称,客户数增加一倍以上,已确认收入翻倍,预订收入增长两倍,团队在过去一年增至 300 多人。Bloomberg syndication 补充了两个重要细节:2026 年收入应接近 $50 million,半导体预计将在第二季度末成为最大细分。这种增长不只是抽象 pipeline。Deutsche Telekom 称 PhysicsX 是其 Industrial AI Cloud 的启动合作伙伴;PhysicsX 后来称平台已在那里上线;双方都描述了由 T-Systems 牵头的入驻,以及客户现场的前置部署小队。CoreWeave 另称 PhysicsX 已经在其云上交付生产级 physical AI。Microsoft Launchpad 入选增加的是 go-to-market 层,而不是直接支出,但重要之处在于,它把分发拓宽到企业 Azure 渠道。因此,扩张模式看起来是旗舰账户中的直接嵌入式工程项目,与通过主权云、GPU 基础设施和平台合作伙伴实现的渠道辅助扩张相混合。[CU002, CU003, CU004, CU006, CU008, CU014]

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
客户数量增长较上一年增长超过一倍2026-06PhysicsX Series C;Digital Engineering 24/7 报道显示当前商业动能已不止少数几个 logo绝对客户数量未披露
已确认收入增长同比翻倍2026-06PhysicsX Series C;Digital Engineering 24/7 报道表明客户基数并非纯试点阶段起始收入基数未披露
已签约收入增长同比增至三倍2026-06PhysicsX Series C;Digital Engineering 24/7 报道暗示管线和新增签约强劲新 logo 与扩张的混合比例未披露
2026 年收入展望接近 $50M2026-06Bloomberg 辛迪加稿即使具名客户不多,规模也足以让集中度变得重要ARR 与项目收入拆分未披露
2027 年增长目标较 2026 年收入翻倍以上2026-06Bloomberg 辛迪加稿管理层预期继续快速扩张目标可能依赖招聘和交付能力
积压订单约六个月客户需求2026-06Bloomberg 辛迪加稿需求看起来领先于交付能力未披露按客户、垂直和交易阶段拆分的积压订单
细分市场结构拐点半导体预计成为最大细分市场2026-06Bloomberg 辛迪加稿支持半导体主导扩张的论点未披露各细分市场收入占比
Industrial AI Cloud 上线平台已上线,并采用客户现场入驻模型2026-02PhysicsX / Deutsche Telekom显示渠道辅助部署路径已经运转入驻客户数量未披露
招聘版图交付、产品和研究方向共有 34 个开放职位2026-06PhysicsX 招聘页面暗示部署能力仍在扩建开放职位不能显示当前可计费利用率或填补周期

采用指标结合了公司自报披露和 Bloomberg 独立辛迪加稿。它们显示动能,但没有把经常性软件收入与项目或服务占比高的工程工作拆开。

[CU002, CU003, CU004, CU006, CU007, CU008]
FU001: 客户旅程图

PhysicsX 公开的客户推进路径,看起来是从高风险工程痛点发现开始,在现有工具内做技术验证,再进入嵌入式交付、安全部署,最后借渠道扩张。

[CU014, CU018, CU019, CU027, CU032, CU045]
FU002: 采用 / 部署漏斗

公开记录支持广泛认知和许多合作伙伴触点,但只有更窄的一组关系有具体工作负载细节,更小的子集看起来明显达到生产级。

取值是从公开证据密度推导的相对指数点,不来自 PhysicsX 的转化数据。该漏斗意在展示,证明质量如何从广泛相关性收窄到耐久收入证据。

[CU001, CU015, CU021, CU023, CU039, CU043]

6.4 工作流嵌入让耐久性看似合理,但公开留存证据几乎缺失

PhysicsX 的公开证据在采纳上远强于耐久性。已审阅来源没有披露 NRR、GRR、流失、合同期限、续约节奏或 cohort 表现。因此,本章节可以说公司有需求,也有一些看起来粘性的集成,但无法像软件尽调流程通常要求的那样,证明重复收入质量。最好的正面耐久性信号是结构性的。Siemens 与 Teamcenter 和 Simcenter 的集成贴近既有工程工作流。CoreWeave 和 Microsoft 都强调安全企业部署和关键任务环境,这通常与更高切换成本和更长实施周期相关。GB1 也更像嵌入式项目,而非一次性 logo。不过,结构性粘性不等于已观察到的留存。六个月 backlog 和供给侧限制可能意味着需求极好,但也造成盲点:如果既有账户扩张受到节流,公开增长指标就无法干净区分 land-and-expand 和延迟交付。实际结论是,耐久性方向上为正,但尚未在公开数据中得到证明。[CU009, CU018, CU019, CU023, CU027, CU038]

留存 / 重复使用 / 满意度表
指标数值 / null分群置信度尽调问题
净收入留存(NRR)未披露全部客户要求按头部客户和垂直行业拆分队列 NRR
总收入留存(GRR)未披露全部客户要求按队列年份提供客户标识总留存和总收入留存
合同期限未披露全部客户获取平均初始期限、续约机制和通知期
续约 / 流失事件未发现公开流失或续约数据全部客户询问续约率、流失账户,以及扩张与收缩队列
公开客户满意度数据集已审阅来源中未发现 G2 / Gartner 式评论语料全部客户寻求客户背调电话或内部 NPS 数据
工作流黏性代理指标正面但间接Siemens、GB1、Microsoft、渠道主导账户验证集成深度是否与多年期合同和重复预订相关
现有客户扩张能力受供给侧限制和节制式推出约束现有账户要求按客户列出实施积压,并说明延迟推出是否影响续约

公开耐久性数据稀缺,所以本表有意保留大量 null。章节可以从嵌入式工作流和安全环境推断黏性,但没有私有收入队列证据,无法证明留存质量。

[CU009, CU018, CU023, CU027, CU038, CU043]

6.5 集中度风险和企业采购摩擦是客户侧主要尽调问题

客户章节的主要红旗是集中度不透明和企业实施摩擦,而不是缺少参考 logo。公开层面,只有 Applied Materials、Siemens 和 Stellantis 是独立具名客户;但 Bloomberg syndication 指向约 $50 million 的 2026 年收入。这并不证明存在危险集中度,但足以让头部账户敞口成为重要尽调问题。投资人重叠会放大集中度风险:Siemens 和 Applied Materials 既是财务支持者,也是客户参考;NVIDIA 既是投资人,也是生态锚点。因此,当交易对手直接发声时,参考质量会提高,Siemens、GB1、Deutsche Telekom、T-Systems、Microsoft 和 CoreWeave 都是如此;但如果关系只存在于融资报道中,参考质量就会变弱。采购摩擦也相当大。PhysicsX 必须进入关键任务工程工作流,证明安全性和准确性,接入 Teamcenter、Simcenter 或私有云环境,并为前置部署项目配备人员。Sifted 关于 rollout 能力的引用显示,需求跑在交付能力前面,这好过需求疲软,但仍会限制扩张和续约。广泛的航空航天与国防主张也值得谨慎,因为没有找到公开国防客户、合同或合规路径。[CU007, CU009, CU017, CU018, CU027, CU031]

扩张与集中度风险表
驱动因素 / 风险类别影响尽调路径
到 2026 年 Q2,半导体成为最大分群扩张驱动因素高 — 支撑公司深入一个难以被替代的工业预算池按半导体账户,以及晶圆厂设备客户与芯片设计客户,拆分预订和续约
Deutsche Telekom / T-Systems 云渠道扩张驱动因素高 — 带来欧洲分销和客户现场实施杠杆衡量渠道来源商机管线、转化率和收入占比
CoreWeave 和 Azure 企业部署路径扩张驱动因素中高 — 降低客户安全推出的技术门槛测试云端赋能是否真正缩短销售或部署周期
公开具名客户只有三家集中度风险高 — 即使客户总数更多,具名客户集合很小也可能意味着实质收入集中要求提供 top-1 / top-3 / top-10 收入集中度和 HHI
投资者与客户参考重叠参考质量风险中 — 战略协同有帮助,但会削弱证言独立性背调时区分投资者客户和非投资者客户
六个月积压和节制式推出执行 / 集中度风险中高 — 需求强,但实施延迟可能拖慢现有账户扩张要求按账户列出积压账龄、实施人员配置和收入转化时间
无公开留存指标耐久性风险中 — 削弱对早期试点转化为持久经常性收入的信心在 NDA 下审阅续约队列和扩张预订

由于没有公开客户收入拆分,影响级别属于定性判断。最强的公开风险信号不是需求疲软,而是集中度不透明叠加实施产能受限。

[CU006, CU007, CU008, CU014, CU019, CU036]
采购与部署摩擦表
摩擦点公开证据受影响分群当前缓释措施剩余缺口
任务关键型准确性和安全性证明Startup Fortune 称,安全、成本和性能一旦上桌,工业客户不会为含糊的 AI 承诺买单所有严肃工业账户PhysicsX 发布技术说明,并依赖深度嵌入工作流未公开把模型准确性与客户验收标准绑定的验证框架
与既有工具链的工作流集成PhysicsX 称其可接入 Teamcenter 和 Simcenter,且不打断现有工作流工业工程团队原生集成和仿真在环数据生成未公开每个账户的达产时间或集成周期数据
安全企业部署和受监管环境CoreWeave 和 Microsoft 强调安全企业环境与受监管云运营大型企业、主权或受监管客户CoreWeave 和 Azure 上的私有 LPM 部署路径未公开合规工件,也没有具名国防 / 政府生产客户
数据中心工作负载的基础设施和电网约束Siemens 称,电网容量和并网时间表限制数据中心增长数据中心和高能耗客户PhysicsX + Siemens + Fluence + Emerald AI 生态公开证明最强的是一个 Siemens 工作流,而不是广泛的多站点推出
实施产能Sifted 报道称,PhysicsX 受供给侧限制,因此节制推出;招聘页面显示仍在持续招聘现有和新增企业账户在交付和安全岗位招聘,并借助 T-Systems / 客户现场小队未公开交付团队利用率、部署时间或流失率指标
国防 / 出口管制不透明PhysicsX 多次列出航空航天和国防,但已审阅来源均未点名国防客户或公开合同工具国防和高度受监管买方行业营销加嵌入式工程模型需要具名国防证明,否则应将国防视为未验证

摩擦项来自公开证据,而不是假设性的企业软件清单。公司已有若干缓释措施,但几乎都需要私有尽调证明其能扩展到旗舰账户之外。

[CU014, CU017, CU018, CU019, CU027, CU032]
Chapter 07

07风险

7.1 按严重程度排序的风险概览

PhysicsX 的风险栈,最前面不是已知法律爆雷,而是验证与商业化。公司明确卖向航空航天与国防、半导体、汽车、能源、材料和数据中心基础设施,并把平台定位成关键任务、关乎安全的工程软件。与此同时,它自己的技术材料也说得很清楚:Fourier Neural Operators 等核心替代模型方法,在边界条件、几何、冲击和校准上都有边界;平台 FAQ 也承认,即便有不确定性量化,输出仍是近似值。因此最高严重度风险很直接:一旦模型在安全、质量或正常运行时间关键的工作流中,被信任到超出已验证运行包线,错误会比普通企业软件 bug 更快传导为客户责任、采用放缓或合同流失。 下一层风险来自依赖与规模。PhysicsX 当前增长计划依赖大型云和算力伙伴,包括 Azure、AWS、CoreWeave、Deutsche Telekom 搭载 NVIDIA 的 Industrial AI Cloud,以及与 Siemens 相关的工具链和应用。这些合作能缩短采购、拿到 GPU 基础设施,是优势;但也把议价权和执行风险集中到 PhysicsX 控制之外。融资上,公司从 2025 年 6 月 $135 million Series B,到 2025 年 11 月扩展后超过 $155 million,再到 2026 年 6 月约 $2.4 billion 估值、超额认购的 $300 million Series C。资本底座降低了近期破产风险,却显著抬高了市场对规模化、试点转化、交付吞吐和最终软件经济性的期待。 法律和监管暴露真实存在,但更多是公开资料不足,而非已经出现负面事件。PhysicsX 有公开隐私声明和 ISO 27001 认证,但隐私声明日期停在 2022 年 10 月,仍指向旧 Shoreditch 地址;Companies House 则显示后续办公地址已变更。公开材料没有找到产品级 DPA、SLA、信任门户、事故记录、出口许可统计,也没有保留的诉讼或执法案件。正确解读不是风险不存在,而是尽调负担仍然很重:相对于 PhysicsX 瞄准行业的雄心,公开缓释证据明显偏薄。[CR001, CR003, CR004, CR005, CR006, CR009]

FR001: 风险热力图

PhysicsX 主要风险簇的发生可能性、影响、缓释成熟度和剩余敞口。

单元格是基于公开技术、法律、合作伙伴和融资证据综合得出的定性判断,并非来自公司披露的风险评分。

[CR003, CR004, CR005, CR006, CR031, CR041]

7.2 法律、监管与责任风险

PhysicsX 的法律与监管画像,不是由某一起已知执法事件定义,而是由雄心很大的行业覆盖、陈旧的公开法律披露和产品责任不确定性共同塑造。公司公开称业务覆盖航空航天与国防、半导体、能源等先进工业场景;在这些场景里,出口管制、制裁、隐私和合同责任分配都可能成为门槛。BIS 指引明确要求出口商判断哪些事项受 EAR 管辖、哪些需要许可证或例外;OFAC 则在与战略产业供应链直接相关的多个司法辖区维持制裁项目。PhysicsX 还在欧洲推广主权云部署,并支持高密级或混合 / 本地环境,因此出口筛查和辖区合规不是理论负担,而是公司若干最有吸引力终端市场的 go-to-market 组成部分。 隐私和企业合同风险,比公开诉讼风险更看得见。PhysicsX 网站公开的隐私声明称其遵守 UK Data Protection Act 2018 和 GDPR,并引导用户向 ICO 投诉。这是有用的基线,但不等于完整公开的企业保证包。同一份文件最后更新于 2022 年 10 月,仍列出公司旧 Shoreditch 地址;Companies House 则显示,2024 和 2025 年地址更新后,当前注册地址已在 Victoria House。对成熟买家来说,这个错位虽小但真实:公开法律卫生看起来没有明显跟上公司的规模、融资和受监管行业野心。 最难的法律问题,是安全关键工程中的模型责任分配。PhysicsX 称其平台面向关键任务工程挑战,但自家技术文章也说明,标准 neural-operator 方法不保证严格满足边界条件,也不适合冲击、不连续、非规则几何和部分高维问题。因此,保证条款、验证协议、人工签核和责任限制,都是核心尽调项。2026 年本章采用的检索路径没有发现保留的公开诉讼或执法行动,但这只是证据缺口:登记页面和官网不能替代外部律师、案卷检索和客户合同抽样。[CR005, CR006, CR007, CR008, CR010, CR011]

监管 / 法律风险台账
风险司法辖区 / 来源当前公开信号可能性严重性缓释措施剩余暴露尽调路径
安全关键工作流中的模型责任客户合同;工程 QA;任务关键型用例任务关键型定位,加上关于模型局限的明确技术提示UQ、主动学习、客户专属模型、人工复核在抽样合同和验证协议之前,剩余暴露为高要求按用例提供责任上限、验证 SOP、保修排除和签字归属。
出口管制和制裁筛查BIS EAR 基础规则;OFAC 制裁制度;航空航天 / 半导体行业组合战略行业定位和主权云商业化路径让跨境筛查变得重要混合 / 本地部署和客户专属环境可以把工作流本地化中到高,因为国家收入组合、被拒方控制和许可历史未公开审阅国家收入组合、受限方筛查工作流,以及任何出口许可意见或拒绝记录。
隐私和数据保护合规PhysicsX 隐私通知;ICO AI 指引公开隐私通知存在,但最后更新于 October 2022,与当前申报相比显得陈旧ISO 27001、已声明的 GDPR / Data Protection Act 依据,以及客户专属训练隔离中,因为公开材料未披露 DPA、留存、子处理方和事件条款要求提供当前 DPA、子处理方、留存时间表、模型训练限制和隐私治理负责人。
诉讼 / 执法可见性缺口Companies House;官网;2026 年网页发现路径未发现留存公开案件,但公开来源很薄,不能替代案卷审查低到中公司状态为 active,留存材料中无明显公开执法事件中,因为没有公开命中并不证明不存在请律师针对头部客户和行业开展英国 / 美国案卷检查、产品责任审查和索赔历史核查。
IP 和合同防御性公开产品主张;合作伙伴集成;法律包强产品主张和合作伙伴主张已公开,但企业责任条款和专利姿态未公开嵌入客户数据 / 工作流和 ISO 流程纪律,可能提高切换成本中,因为公开证据不足以判断专利、赔偿和源代码 / 托管条款审阅专利 / 申请清单、开源使用、客户赔偿,以及入站 / 出站 IP 转让条款。

各行按剩余严重性排序。未发现留存的公开诉讼或执法案件;该行记录的是证据缺口,不是不存在的证明。

[CR005, CR006, CR007, CR008, CR010, CR011]

7.3 运营、模型可靠性与安全风险

从运营看,PhysicsX 的核心暴露在于,它把加速推理卖进了错误答案代价很高的工作流。平台承诺把仿真周期从数小时或数天压缩到数秒,并支持关键任务工程问题。这正是价值主张,但公司自己的材料反复加了限定:输出是近似值,模型准确度取决于用例和训练数据,不确定性需要校准;当几何、不连续、高频细节或严格边界条件占主导时,部分 neural-operator 家族会失效。这不是完全回避公司的理由,而是把验证纪律列为第一运营控制的理由。 PhysicsX 确实有运营缓释。平台 FAQ 称模型按客户定制,不跨客户联合训练;出现高不确定性区域时支持主动学习;高密级场景可用混合或本地部署。公司还宣布取得 ISO 27001 认证,并称覆盖财务数据、知识产权、员工信息和第三方数据。对企业采用来说,这些控制有意义,尤其是在敏感工业环境里。但公开证据仍留下明显空洞:没有公开状态页、没有可见事故档案、没有公开产品 SLA,也没有公开的数据泄露或正常运行时间指标,投资人无法判断韧性是结构性强,还是只是未披露。 商业上,PhysicsX 的运营模型仍然服务密集。公司称前置部署工程师会直接嵌入客户项目,试点由双方共同界定、持续一到三个月,客户最终需要逐步成为独立用户。这带来典型 deeptech 软件风险:公司能在高接触部署中证明价值,却仍可能难以把价值转化为可重复、可扩展的软件使用。如果试点仍高度依赖稀缺内部人才,经营杠杆和毛利率扩张会落后于融资与估值叙事。[CR003, CR004, CR013, CR016, CR018, CR019]

运营 / 质量 / 安全风险台账
失效模式可能性严重性缓释成熟度剩余暴露未解决缺口
替代模型在已验证运行边界之外仍被信任中等:UQ、主动学习和客户专属模型已公开任务关键型工程工作流中的一次错误预测,可能带来直接责任并损害采用未公开验证协议、误差预算或用例验收标准。
FNO 风格架构在冲击、不规则几何或严格边界条件下表现不佳中等:PhysicsX 公开记录了这些方法的短板技术边缘案例可能迫使客户昂贵地退回经典仿真,或限制产品范围需要客户层面证据,说明回退率以及几何感知替代方案已部署在何处。
不确定性校准失准,制造虚假信心低到中:UQ 已公开,但校准证据未公开置信区间若不能贴合现实,可能比可见的不确定性更糟需要按工作流提供校准指标、部署后监控和升级阈值。
尽管有 ISO 27001,仍发生安全或隐私事件低到中中等:ISO 27001 和客户专属模型训练可见战略行业中的一次泄露就会伤害信任和扩张未发现公开事件档案、状态页或产品 SLA 历史。
高接触试点无法扩展成可重复产品使用中到高中等:试点纪律和前置交付已被明确提出交付重的采用模式会压住毛利率,并拖慢可重复性需要试点到生产转化、部署周期和客户自助指标。

运营风险主要来自验证和规模化纪律,而非已知公开宕机或召回历史。

[CR003, CR004, CR013, CR016, CR018, CR019]
FR002: 风险传导图

模型验证和保障缺口如何传导到采用、续约和估值。

边表示的因果链,来自 PhysicsX 技术披露和试点驱动部署模式最直接指向的影响。

[CR004, CR020, CR021, CR022, CR023, CR042]

7.4 伙伴、云与商业依赖风险

PhysicsX 的合作网络很亮眼,但也是最清晰的非技术剩余风险来源。平台称自己云无关,可在 AWS、Azure、混合或本地基础设施上运行;但公开增长叙事反复依赖具名算力和渠道伙伴。AWS 将 PhysicsX 选入 Generative AI Accelerator。Microsoft 让 PhysicsX 成为 Microsoft Discovery 的发布伙伴,并推出私有 Azure Marketplace 版本,同时另行称 Azure 是关键任务工程工作负载的基础。Deutsche Telekom 和 T-Systems 给 PhysicsX 带来欧洲主权分发和搭载 NVIDIA 的 Industrial AI Cloud。CoreWeave 为前沿模型训练提供专用 GPU 云。Siemens 把公司延伸进工业工作流和数据中心电力应用。每个伙伴都扩大触达;每个伙伴也都成为依赖节点。 实际问题在于,算力、采购和客户语境的控制权集中。CoreWeave 和 Deutsche Telekom 重要,是因为 PhysicsX 明确表示,更大的 Large Physics Models 需要在海量工作负载上持续获得高吞吐 GPU 算力。Microsoft、AWS 和 Siemens 重要,是因为它们能加速企业分发,也会调解客户已经购买、保护和运营技术软件的环境。这意味着,即便产品需求健康,伙伴经济性、路线图变化、云可用性或战略一致性改变,也可能拖慢 PhysicsX。欧洲主权云就是好例子:它强化 go-to-market,但也把部分投资逻辑绑到特定基础设施推出和伙伴主导的 onboarding 动作上。 还有一层更隐蔽的依赖:既有工程工具链。PhysicsX 推广 ANSYS、CATIA、Siemens NX、OpenFOAM 和 Siemens Star CCM+ 连接器,这很务实,因为工程师不想拆掉现有系统。但这也意味着,集成复杂度、客户 IT 约束或第三方工具经济性变化,都可能拖慢采用。PhysicsX 越是靠嵌入其他生态成功,投资人越需要监控:公司掌握的是战略层,还是只是改良了别人的技术栈。[CR025, CR026, CR027, CR028, CR029, CR030]

合作伙伴 / 依赖风险台账
依赖交易对方角色集中度失效场景严重性缓释措施剩余暴露
Azure / MicrosoftMicrosoft基础设施、市场和发布合作伙伴渠道合作伙伴优先级变化,或 Azure 经济性压缩 PhysicsX 在大客户中的杠杆云无关姿态和混合部署主张高,因为 Microsoft 既是赋能方,也是强大的相邻平台所有者
AWSAmazon Web Services训练支持、生态接入和加速器网络AWS 仍有助于早期接入,但没有变成持久差异化渠道多云支持和替代计算伙伴中,因为 AWS 依赖真实存在但并非唯一
主权工业 AI 云Deutsche Telekom / T-Systems / NVIDIA面向主权部署的欧洲计算、导入和商业化路径推出延迟或战略优先级重排拖慢模型训练和欧洲企业采用三年合作伙伴关系和一站式部署动作高,因为主权云定位现在已是欧洲论点的一部分
前沿训练计算CoreWeave面向大模型训练和部署的高吞吐 GPU 云计算稀缺、定价或可用性约束前沿模型进展云多样性,以及部分工作负载的本地 / 混合选项高,因为前沿 LPM 训练明确需要持续 GPU 吞吐
工业工作流和工具链接入Siemens 和既有 CAE 工具仿真数据、工作流集成和工业分销语境中到高连接器或路线图摩擦限制在客户环境中的采用中到高接入多个 CAE 工具的连接器和客户专属部署中,因为 PhysicsX 仍依赖既有生态,在工程师工作的地方触达他们

依赖风险集中在计算接入、采购渠道和既有工作流所有权,而不是单一经销商关系。

[CR025, CR026, CR027, CR028, CR029, CR030]
FR003: 依赖关系图

塑造 PhysicsX 部署、训练和客户导入的关键外部伙伴与基础设施节点。

这张图强调基础设施和渠道依赖,而不是收入占比;实践中,多个合作伙伴角色会重叠。

[CR025, CR026, CR027, CR028, CR029, CR030]

7.5 财务、人才与投资逻辑击穿触发器

PhysicsX 的财务故事,需求最容易相信,软件经济性最难承销。公开材料支持强劲的收入动能:公司称确认收入同比翻倍、预订收入同比增长三倍、客户数增加一倍以上,2026 年 6 月员工数超过 300 人;tech.eu 报道约 350 名员工、累计融资约 $500 million。这些信号很强,但没有回答 $2.4 billion 估值下最关键的问题:毛利率、烧钱速度、 runway、客户集中度、续约质量,以及收入中有多少来自高接触交付而非可扩展产品使用,在本次审阅的公开材料中都未披露。 人才和执行风险,就在财务不透明之下。PhysicsX 依赖异常稀缺的跨学科人才:仿真工程师、数据科学家、机器学习工程师、产品建设者,以及能嵌入硬核工业项目的前置客户团队。管理层厚度有所增强,Chris Wigley 加入担任 COO,Alexander Dreismann 转向战略;但这项变化本身也说明,公司在全球商业规模化上仍需证明很多。招聘页面还强调英国和美国搬迁与签证支持,再次提醒投资人:招聘是国际化且竞争激烈的,不是常规动作。 缓释因素是自洽的:ISO 27001、客户专属模型、不确定性量化、混合部署、主权云伙伴和更深的运营班底,都能降低执行风险。但只要出现三类情况之一,投资逻辑就应被击穿。第一,生产工程工作流中出现验证或安全失败,会直接攻击信任。第二,伙伴或算力中断若实质性拖慢训练、部署或 onboarding,就会暴露公司外部杠杆有多大。第三,如果公开或私下尽调显示,试点无法转化为可重复的软件经济性,当前估值就会显得远远跑在商业模式前面。投资人应把这些监控项视为不可谈判。[CR035, CR036, CR037, CR038, CR039, CR040]

人才 / 执行风险登记表
角色 / 职能依赖或缺口可能性严重性缓释措施尽调路径
仿真 + ML 人才横跨物理、软件和交付的复合型招聘池稀缺品牌、使命和资本有助于招聘按核心技术岗位审查流失率、招聘周期和薪酬压力。
前置部署工程团队客户成功靠嵌入式操盘手,不只是软件席位中到高试点纪律和模板可提升可复制性要求提供试点人员配比、服务收入结构和部署利用率。
领导层扩容快速全球扩张要求新增 COO 能力,并持续调整角色定义中到高Chris Wigley 入职及更广泛的骨干补强审查组织设计、销售领导覆盖,以及产品、交付和 GTM 之间的决策权。
跨境招聘 / 调动英国-美国布局和签证支持,使移民执行成为规模化的一部分伦敦和纽约枢纽加上调动支持检查优先团队的招聘结构、签证依赖和招聘提前期。
商业化纪律需要把高难度试点转成可复制的软件使用和续约价值客户赋能和一体化平台路线图检查试点转化、扩张和客户自助采用指标。

人才风险格外关键,因为公司同时要招深度科学人才、把软件产品化,并完成前置部署式工业交付。

[CR022, CR023, CR042, CR043, CR044, CR045]
缓释与否决标准表
风险可监测触发项阈值 / 事件行动含义
模型验证 / 责任生产用例中的验证误差纪律安全、质量或在线时间关键的部署出现超出批准边界的重大偏差暂停承销,直到验证治理、责任分配和后备仿真控制得到核实。
云 / GPU 依赖合作伙伴可用性和接入节奏CoreWeave、Azure 或 Deutsche Telekom 延误,实质拖慢训练、部署或客户接入下调护城河评分,因为算力杠杆被证明来自外部,而非自有能力。
隐私 / 安全 / 信任公开与客户保证界面发生安全事件,或尽调发现 DPA、分包处理方或事件治理薄弱假设采购变慢,向战略或受监管行业扩张的空间降低。
商业化质量试点转生产和续约行为试点仍高度定制,转化滞后,或软件使用没有超出嵌入式项目将业务标记为服务偏重,并质疑估值倍数。
融资 / 估值纪律资本效率和经济性透明度燃烧率、毛利率、现金跑道或集中度数据相对 $2.4B 价格令人失望在提高确信度前,把工作从增长承销转向经济性证明尽调。
骨干规模 / 领导层执行留任和组织杠杆全球扩张期间关键技术或运营负责人离职,或招聘提前期明显拉长将执行风险视为会与商业化风险叠加,而不是独立的人才问题。

本表聚焦尽调期间和下一次更新可监测、足以打破论点的条件,而非泛泛的观察项。

[CR041, CR047, CR048, CR050, CR051, CR052]

7.6 附录

Chapter 08

08估值

8.1 融资背景与证明缺口

PhysicsX 显然已经跨过一个重要融资里程碑:公司自己 2026 年 6 月 8 日公告和 Pulse 2.0 都报道,其完成超额认购的 $300 million Series C,估值约 $2.4 billion,由 Temasek 领投,M&G Investments 和 Intrepid Growth Partners 加入,既有投资人包括 Applied Materials、Atomico、NVIDIA、Siemens 等。此前 2025 年 Series B 扩展把估值推向 $1 billion,因此头部估值约一年内提升了约 2.4x。问题不在于融资是否发生——它已经发生。问题在于,外部投资人能否独立支撑这个新价格。公开来源提供了相对增长表述、300 多名员工和 2026 年新股配发证据,但没有披露绝对收入基数、软件与服务 mix、毛利率、留存或优先权结构;这些才决定该价格是高溢价但公平,还是只是太早。按今天估值,证明负担已经从愿景叙事转向硬经济证据。[CV001, CV002, CV004, CV005, CV007, CV008]

可比估值表
可比公司背景收入锚点估值 / 倍数相关性局限
PhysicsX2026 年私有 Series C未公开披露约 $2.4B 投后估值当前直接入场点。无公开收入、利润率、留存或优先权结构披露。
Cadence公开 EDA / 工程软件TTM 收入约 $5.213B(2025 年 9 月)2026 年 6 月市值约 $106.84B;约 20.5x 市值 / 收入与工程工作流高度相关的高端软件类比。规模远大于 PhysicsX,且为公开、成熟度更高的公司。
Ansys公开仿真软件TTM 收入约 $2.468B(2024 年 9 月)2025 年 8 月市值约 $32.90B;约 13.3x 市值 / 收入最接近、以仿真为先的工业软件锚点。收购前 / 历史快照,且仍是成熟公开资产。
Autodesk公开设计与制造软件TTM 收入约 $6.888B(2025 年 10 月)2026 年 6 月市值约 $40.92B;约 5.9x 市值 / 收入展示没有前沿 AI 稀缺性时,广义工业设计软件的交易水平。品类结构更广,买方群体不同。
PTC公开工业产品生命周期软件TTM 收入约 $2.739B(2025 年 9 月)2026 年 6 月市值约 $13.25B;约 4.8x 市值 / 收入有用的低端工业软件锚点。不是纯物理 AI 公司,交易中计入的增长预期更低。

本表用粗略市值 / TTM 收入作为公开速记,而非精确 EV/NTM 倍数;面对经济性未披露的私有公司,它是纪律工具。

[CV004, CV027, CV028, CV029, CV030, CV031]
FV002: 估值敏感性

在不同经常性收入假设下,当前 $2.4B 标记对应的示意性估值倍数。

条形图用估值简单除以假设经常性收入;这是粗略的市值式速记,不是精确 EV/ARR。

[CV032, CV033]

8.2 为什么故事有吸引力,也为什么仍然脆弱

看多理由不是想象出来的。PhysicsX 围绕工业工程中的真实痛点建设:昂贵仿真瓶颈、缓慢设计循环,以及从受 solver 约束的工作流转向 AI 辅助迭代的需求。平台描述自洽,有具名模块、云和本地部署选项、明确的 CAE 集成,以及能嵌入客户项目的交付模型。行业足迹也有吸引力,覆盖航空航天、半导体、汽车、材料和能源;Siemens 与 NVIDIA 合作表明,成熟生态玩家看到了这套 stack 的战略价值。但反向论点同样重要。PhysicsX 自己的技术材料说明,部分 neural-operator 方法恰好会在投资人担心的环境里失效:冲击、非规则几何、严格边界条件和更高维工作负载。这不会否定业务,但意味着可重复性必须逐个工作负载证明。如果部署仍然狭窄、重专家或服务主导,平台叙事就不应享有类似公开市场软件公司的溢价。[CV013, CV014, CV015, CV016, CV017, CV018]

正方论点 / 反方论点表
立场论点哪些证据会改变判断
正方论点PhysicsX 瞄准真实的工业瓶颈,把仿真密集型工程流程压缩成更快的 AI 辅助闭环。证明这一流程优势能在多个工厂、项目或产品族中转化为可复制的经常性收入。
正方论点Siemens、NVIDIA、Temasek 和主要工业投资人的合作信号,说明其战略相关性不止于实验室演示。把合作伙伴叙事转成具名生产部署,并拿出扩张证据和利润率质量。
反方论点公司披露增长百分比,但不披露给后期轮定价所需的收入基数、毛利率、续约画像或服务结构。按垂直行业提供队列收入、留存、毛利率,以及软件与服务收入拆分。
反方论点PhysicsX 自己的技术文章指出,部分神经算子方法在冲击、严格边界条件、不规则几何和高维问题上会失效。证明商业上重要的工作负载落在平台足够耐用、可扩展的问题空间内。

反方论点刻意以证据为基础:来自已披露的信息缺口和 PhysicsX 自己表述的技术边界,而不是泛泛怀疑。

[CV013, CV017, CV018, CV020, CV022, CV034]
FV001: 建议逻辑

建议取决于:可信产品和伙伴信号,是否足以抵消当前价格下缺失的经济性证明。

[CV017, CV018, CV020, CV022, CV031, CV034]

8.3 可比背景、情景区间与价格纪律

合理的公开市场视角,是粗略市值 / 收入,而不是因为它完美,而是因为当企业价值、利润率和净现金无法在同一处稳定披露时,这是最好的公开速记。使用 CompaniesMarketCap 的 2026 年市值和 Macrotrends 的过去十二个月收入,Cadence 约 20.5x,Ansys 约 13.3x,Autodesk 约 5.9x,PTC 约 4.8x,四家公司区间约 4.8x 至 20.5x,中位数接近 9.6x。对照这一区间,PhysicsX 的 $2.4 billion 估值显得昂贵,除非其隐藏收入基数已经相当可观。如果年化经常性收入只有 $50 million,本轮隐含约 48x。即便达到 $100 million,仍约 24x,高于样本公开公司区间。只有当经常性收入大致超过 $120 million,价格才开始接近公开可比区间上沿;要接近中位数,则大约需要 $250 million。因此,下方宽情景区间应被视为说明性的纪律工具,而不是点估计。[CV027, CV028, CV029, CV030, CV031, CV032]

牛市 / 基准 / 熊市场景表
场景核心假设示例估值逻辑概率信号关键风险
牛市已签约增长转化为产品化经常性收入,合作伙伴渠道加深,PhysicsX 证明可在多个工业项目中复制。如果经常性收入达到约 $200M-$250M,且市场继续为品类龙头支付高端工业软件倍数,估值为 $3.5B-$5.0B。需要私下数据证明软件式利润率、续约和多站点扩张。如果增长仍偏服务,高端倍数会消失。
基准PhysicsX 具备战略相关性且仍在快速增长,但外部人只能看到经济性和可复制性的部分证明。如果经常性收入已高于约 $120M,但低于公开可比公司中位数区间隐含的规模,估值为 $2.0B-$3.0B。当前公开证据最支持这一判断:公司有吸引力,价格证明不足。即便平台不错,不透明的单位经济性也会限制上行。
熊市部署范围仍窄,技术边界比预期更重要,或公开工业软件倍数进一步压缩。如果市场在完整收入证明出现前,把业务重估到成熟工程软件区间,估值为 $1.2B-$1.8B。弱扩张、服务成分重,或可复制性证明延迟,都会触发该场景。后期轮资本结构可能放大新钱下行。

这些是示例估值区间,不是目标;它们锚定粗略的公开市值 / 收入区间,因为更好的私下输入尚未公开。

[CV031, CV032, CV033, CV035, CV036, CV037]
FV003: 估值 / 回报区间

熊市、基准和牛市情景下的示意性宽估值区间。

区间有意设得很宽,因为公开证据不足以对私营公司退出时点、稀释或企业价值给出虚假的精确度。

[CV049, CV050, CV051]

8.4 建议、尽调问题与投资逻辑击穿触发器

因此,基于公开证据的答案不是「买入」,而是「继续研究」。PhysicsX 未来可能配得上溢价:融资财团很强,终端市场很大,产品故事瞄准实体经济创新中的真实瓶颈。但在收入质量、客户集中度、毛利率、续约和 cap table 经济性没有公开披露的情况下,当前轮次应被视为一个需要尽调的价格,而不是可以背书的价格。信心为中等,因为公司有可信技术和合作信号;但投资判断仍取决于市场看不见的私有信息。风险为高,因为执行不仅要证明需求,还要证明可重复性和软件经济性。我的价格纪律很简单:只有在管理层能证明经常性软件收入已经大致超过 $100 million,且留存和利润率质量强,或入场价格充分重置、缺失数据不再承担那么多回报压力时,才重新审视。在那之前,工作是尽调,不是确信。[CV034, CV036, CV039, CV040, CV041, CV042]

建议摘要表
维度当前判断决策含义
建议继续研究不要只凭公开证据承销 Series C 价格。
确信度技术和合作伙伴质量真实存在,但核心经济性仍未公开。
风险评级执行和估值都依赖未公开披露的指标。
估值立场偏紧该轮在外部人能验证之前,就把高端软件经济性计入价格。
价格纪律只有拿到私下证明或更低入场价格才重新接触需要 >$100M 经常性软件收入,并具备强留存 / 利润率,或入场价格明显更便宜。
目标回报 / 持有逻辑公开数据无法支撑没有股权结构和收入质量数据,持有期 IRR 建模只是伪精确。

本表刻意给出判断而非数字,因为公开证据不足以支撑精确回报建模。

[CV039, CV040, CV041, CV042, CV043]
最终尽调问题表
主题缺失证据重要性负责人 / 尽调路径
收入质量2025 和 2026 年实际经常性收入、签约到确认收入的桥接,以及按垂直行业拆分的软件与服务结构。这是判断 $2.4B 究竟是高端但公平、还是单纯过早的核心缺失变量。CFO / 数据室请求;将管理层 KPI 材料与审计账目对齐。
留存和扩张毛留存、净留存、按客户队列的扩张,以及从部署到扩张的时间线。说明客户增长是在像软件一样复利,还是像咨询项目一样重置。CRO 访谈,加上来自财务系统的队列表。
毛利率和交付强度按产品和交付模式拆分的毛利率,以及前置部署工程人员负载。服务偏重的运营模式不应拿到高端软件倍数。CFO 和交付领导层尽调。
股权结构和优先权完整股份类别结构、清算优先权、反稀释保护和期权池计算。没有这些,新投资人的回报结果无法可信。外部律师参与的法律尽调和股权结构导出。
客户集中度头部客户收入占比、续约时间,以及部署能扩展到少数旗舰项目之外的证明。集中度过窄会让当前估值变脆。财务 + 销售运营导出,然后做客户推荐访谈。
退出准备度董事会层面对融资跑道、投行接触、二级流动性,以及现实 IPO 或战略退出时间的看法。后期入场价格下,持有期和下行保护更重要。董事会和投资人权利尽调。

这些问题按其对建议、确信度和价格纪律的直接影响排序。

[CV034, CV036, CV043, CV044, CV045]
FV004: 投资 KPI

IC 式评分偏好市场相关性和产品野心,但会折减估值支撑和证据质量。

[CV017, CV018, CV022, CV034, CV039, CV041]

8.5 什么会击穿投资逻辑

一章好的后期估值分析需要明确淘汰标准,因为这里的错误,是把强故事误当成强入场点。第一个击穿点,是有证据显示部署仍主要项目制,无法跨站点、工厂或产品线转化为可重复、可扩张的软件收入。第二个,是技术证据显示 PhysicsX 的最佳结果仅限于平滑、高度结构化的问题,而重要客户工作负载仍需要专家仿真团队和缓慢人工介入。第三个,是工程软件和 EDA 公开市场重置,而 PhysicsX 没有对冲性的披露,因为后期私有价格很少能永远脱离公开可比现实。第四个,是 cap table 或优先权结构让新钱几乎没有上行空间。这些不是理论担忧;它们是当前结论的实际理由:把公司留在尽调名单里,但把价格拴得很紧。[CV037, CV041, CV042, CV045, CV046, CV047]

论点击穿与否决触发项表
触发项阈值 / 事件对论点的传导行动含义
可复制性失败灯塔项目未能在 12-18 个月内转化为多站点或多项目扩张。削弱 PhysicsX 正在成为可复制软件层、而非专家服务的主张。从继续研究转为在当前价格下回避。
技术范围收窄商业胜利只集中在平滑、求解器友好的工作负载,而复杂工业场景仍需人工处理。说明核心 TAM 可能比平台叙事暗示的更窄。下调估值区间,并要求按垂直行业给出更严格证明。
公开可比重置工业 / EDA 软件倍数显著压缩,而 PhysicsX 没有披露收入来抵消。后期私有估值往往会重新锚定公开可比公司。在价格重置或经济性披露前,不再追加资本。
股权结构悬顶优先权结构、反稀释或期权池扩张实质削弱普通股上行。即便运营执行良好,也未必转化为新钱可接受的回报。进一步接触前重建回报模型。
证明缺口持续Series C 完成后,管理层仍无法披露队列收入、毛利率和留存。业务成熟却仍不透明时,确信度应下降。维持继续研究建议,或退出。

每个触发项都可监测,并与估值传导绑定,而不是泛泛的公司质量担忧。

[CV041, CV042, CV045, CV046, CV047, CV048]

8.6 附录

免责声明

本报告是基于公开证据的尽调快照,不构成投资建议。重要的财务、法律、技术和合同事实仍未公开;作出任何投资决定前,应直接向管理层和原始文件核验。

证据索引

结论
编号陈述可信度来源
CO001 PhysicsX describes itself as a physics-AI or AI-native engineering company focused on industrial hardware development. SO001, SO003, SO025
CO002 PhysicsX says its mission is to empower industrial organizations and accelerate hardware innovation through AI-native engineering workflows. SO001, SO003
CO003 PhysicsX is headquartered in London, United Kingdom. SO003, SO006
CO004 The registered office for PHYSICSX LIMITED is Victoria House, 1 Leonard Circus, London, EC2A 4DQ. SO006, SO002
CO005 PhysicsX publicly says it operates offices in London and New York. SO003, SO002
CO006 The June 2026 Series C announcement says PhysicsX is expanding its presence in the Bay Area and Singapore. SO003
CO007 PHYSICSX LIMITED was incorporated on 1 August 2019. SO006
CO008 The company operated as Motodynamics Ltd until its legal name changed to PHYSICSX LIMITED on 9 July 2020. SO006
CO009 A 2023 company leadership article says PhysicsX launched in 2020. SO005
CO010 Most financing materials and independent coverage identify Jacomo Corbo and Robin Tuluie as the founding pair behind PhysicsX. SO008, SO014, SO018, SO019, SO022
CO011 The current about page also lists Nicolas Haag as a co-founder and director of simulation engineering. SO004
CO012 Jacomo Corbo is presented as CEO and co-founder in 2026 materials. SO004, SO007, SO003
CO013 Robin Tuluie is presented as founder or co-founder plus chairman in 2026 materials. SO003, SO007, SO004
CO014 Companies House records show Jacomo Corbo was appointed as an officer on 1 January 2023. SO007
CO015 Jim Baum is a board member or director at PhysicsX and was appointed on 20 October 2023. SO004, SO007, SO008
CO016 Laura Connell became a director on 20 June 2025, adding board-level representation for Atomico after it led the Series B. SO007, SO009
CO017 Founder-market fit is unusually strong because Tuluie comes from elite Formula 1 engineering roles and Corbo comes from Formula 1 plus QuantumBlack and McKinsey industrial AI. SO009, SO014, SO005
CO018 PhysicsX publicly emerged from stealth in November 2023. SO008, SO014
CO019 The November 2023 Series A raised $32 million and was led by General Catalyst, with Standard Investments, NGP, Radius Capital, and Henry Kravis also participating. SO008, SO014, SO016
CO020 The June 2025 Series B raised $135 million and was led by Atomico, with Temasek, Siemens, Applied Materials, July Fund, General Catalyst, NGP, Radius, Standard Investments, and Allen & Co participating. SO009
CO021 Series B materials said total funding had reached nearly $170 million and headcount had grown past 150 by mid-2025. SO009
CO022 Series B materials said PhysicsX had more than quadrupled revenue over the previous two years. SO009
CO023 PhysicsX announced a $300 million Series C at an approximately $2.4 billion valuation on 8 June 2026. SO003, SO017, SO018, SO019, SO021, SO022
CO024 Temasek led the Series C, with M&G Investments and Intrepid Growth Partners joining as new investors and Applied Materials, Atomico, General Catalyst, July Fund, NGP, NVIDIA, Radius, and Siemens continuing. SO003, SO017, SO018, SO021
CO025 Temasek first invested in PhysicsX in 2025 before leading the 2026 Series C. SO003, SO017, SO021
CO026 Summing the publicly disclosed Series A, B, and C round sizes yields at least $467 million of equity funding, while some 2026 outlets round the total to roughly $500 million. SO008, SO009, SO003, SO018, SO019
CO027 Official 2026 materials say recognized revenue doubled year over year. SO003, SO017, SO021
CO028 Official 2026 materials say booked revenue tripled year over year. SO003, SO017, SO021
CO029 Official 2026 materials say customer count more than doubled over the prior year. SO003, SO017, SO021
CO030 Official 2026 materials say the team had grown to more than 300 people and doubled in size over the last twelve months. SO003, SO017, SO021
CO031 Independent June 2026 coverage places PhysicsX headcount at roughly 350 people. SO018, SO019, SO022
CO032 PhysicsX says its platform unifies simulation, physics AI, data, and engineering applications across the full product lifecycle. SO001, SO025
CO033 PhysicsX says it integrates with CAE tools including ANSYS, CATIA, Siemens NX, OpenFOAM, and Star CCM+. SO025
CO034 PhysicsX explicitly targets aerospace and defense, semiconductors, materials, automotive, and energy and renewables. SO001, SO026
CO035 Microsoft says PhysicsX has used its platform to improve thermal behavior in Surface devices and accelerate semiconductor equipment prototyping. SO015
CO036 The March 2026 CoreWeave partnership positioned PhysicsX to train and deploy private Large Physics Models on AI-focused GPU cloud infrastructure. SO013
CO037 The March 2026 NVIDIA collaboration focused on open standards, Opora, and the Large Physics Models narrative. SO011
CO038 The March 2026 Siemens collaboration targeted power-distribution design for next-generation AI data centers. SO010
CO039 The March 2026 GB1 partnership embedded PhysicsX in Britain’s America’s Cup engineering campaign. SO012
CO040 Public 2026 coverage increasingly ties PhysicsX demand to semiconductors, AI data-center infrastructure, and other industrial hardware bottlenecks. SO015, SO018, SO019, SO021
CO041 Key-person dependence remains material because public fundraising, technical vision, and partnership messaging are still centered on Corbo and Tuluie. SO003, SO009, SO019
CO042 Governance has broadened since 2023, but public disclosure on the full board, cap-table control, and founder roster remains incomplete. SO007, SO008, SO009, SO004
CO043 Execution risk remains visible because public sources pair tripled bookings with only doubled recognized revenue and report that demand is supply-side constrained. SO003, SO019, SO021, SO023
CO044 New Market Pitch argues the $2.4 billion valuation is only defendable if PhysicsX quickly converts demand into much larger realized revenue and proves software-style economics. SO023
CO045 New Market Pitch also argues investors still need proof that PhysicsX is more platform than services-heavy engineering shop. SO023
CO046 Sifted and The Next Web report that PhysicsX plans more US expansion and a Singapore office while remaining headquartered in London. SO019, SO022
CO047 Sifted says PhysicsX ranked second in its inaugural AI 100 of European AI startups. SO022
CO048 Public sources show an operating bench beyond the founders, including COO Alexander Dreismann and North America leader Mark Huntington. SO004, SO015
CM001 PhysicsX describes its product as an AI-native engineering software stack that unifies simulation, physics AI, data, and engineering applications across the product lifecycle. SM001, SM002
CM002 PhysicsX says its platform is deployed across aerospace and defense, semiconductors, industrial machinery, automotive, energy, materials, and other hardware-intensive sectors. SM002, SM004, SM005
CM003 The market relevant to PhysicsX is narrower than generic industrial AI because the product is sold into engineering workflows where physics prediction, validation, and optimization influence product decisions. SM001, SM003, SM025
CM004 Included spend for the PhysicsX opportunity includes simulation and CAE software, cloud and HPC-backed engineering runs, digital-twin validation workflows, and AI-surrogate layers embedded into those workflows. SM001, SM013, SM025
CM005 Excluded spend includes generic enterprise AI, factory automation hardware, and software not tied to physical-system design, validation, or operation. SM001, SM013
CM006 Status-quo substitutes are incumbent CAE and EDA-linked engineering stacks from Ansys, Siemens/Altair, Synopsys, and Cadence, plus internal solver and HPC workflows. SM008, SM009, SM010, SM012, SM027
CM007 Siemens says successful AI deployment into engineering depends on high-quality synthetic data, robust CAE-AI integration, and trust in the underlying technologies. SM003
CM008 PhysicsX and Siemens say LGM-Aero was trained on more than 25 million geometries and tens of thousands of CFD and FEA simulations. SM003, SM006
CM009 PhysicsX says its models predict physical behavior in seconds rather than hours or days and let teams evaluate orders of magnitude more design variants. SM002, SM004
CM010 Microsoft says PhysicsX is used to shorten semiconductor equipment prototype development and to improve thermal behavior in Microsoft Surface devices. SM004
CM011 AWS says PhysicsX addresses engineering challenges across automotive, aerospace and defense, materials, semiconductors, and energy. SM005
CM012 Synopsys describes itself as delivering engineering solutions from silicon to systems, signaling convergence between EDA and system-level engineering software. SM008
CM013 Cadence says its Intelligent System Design strategy expands beyond traditional chip design into full electromechanical systems. SM009
CM014 Cadence says its system-innovation pillar applies multiphysics analysis to printed circuit boards, advanced packaging, and 3D-ICs. SM009
CM015 Altair describes itself as a provider of simulation, HPC, data analytics, and AI software. SM010
CM016 Altair says demand for its software is expanding beyond simulation engineering specialists into additional verticals. SM010
CM017 Siemens and Altair say HyperWorks 2026 unifies AI, HPC, and multiphysics and can deploy physics-based AI models with results up to 1,000x faster than traditional solver simulations. SM012
CM018 JetZero says lower-HPC-demand aerodynamic tools are critical to gaining useful insights early enough for aerospace development schedules. SM012
CM019 Mordor Intelligence sizes the simulation software market at USD 15.46 billion in 2026 and USD 28.59 billion in 2031, a 13.08% CAGR. SM013
CM020 Grand View Research sizes the broader simulation software market at USD 23.56 billion in 2024 and USD 51.11 billion by 2030, a 14.0% CAGR from 2025 to 2030. SM014
CM021 Business Research Insights values the CAE simulation software market at USD 11.53 billion in 2026. SM015
CM022 Global Growth Insights values the CAE market at USD 7.64 billion in 2026. SM016
CM023 Public market estimates span at least roughly USD 7.6 billion to USD 15.5 billion for CAE or simulation in 2026, and still higher when broader simulation categories are included. SM013, SM015, SM016
CM024 The spread between published estimates is mostly methodological because some publishers count narrow CAE software while others include broader simulation, services, digital twin, or adjacent workflows. SM013, SM015, SM016, SM017
CM025 Mordor says automotive represented 28.32% of 2025 simulation software revenue and product design and engineering represented 41.35% of spend. SM013
CM026 Grand View says engineering, research, modeling, and simulated testing is the largest application segment, and automotive is the largest end-use segment. SM014
CM027 Grand View says U.S. growth sectors for simulation software include aerospace and defense, automotive, pharmaceuticals, and semiconductors. SM014
CM028 SEMI says worldwide 300mm fab equipment spending is expected to reach USD 133 billion in 2026 and USD 151 billion in 2027, driven by AI chip demand. SM024
CM029 SEMI says global fab equipment spending should reach USD 110 billion in 2025 and then rise 18% to USD 130 billion in 2026 due to HPC, memory, data-center, and edge-AI demand. SM023
CM030 The IEA says capex by five large technology companies exceeded USD 400 billion in 2025 and is set to rise a further 75% in 2026, while data-center electricity demand grew 17% in 2025. SM022
CM031 The IEA says electricity use from AI-focused data centers is set to triple by 2030 and that chips, turbines, transformers, planning systems, and grid connections are current bottlenecks. SM022
CM032 BCG says aerospace and defense organizations face rising demand, shrinking critical expertise, aging assets, and increasing technical complexity. SM020
CM033 BCG says regulatory constraints, fragmented data and tools, and difficulty scaling beyond pilots make AI adoption uniquely hard in aerospace and defense. SM020
CM034 SimScale’s 2026 survey says only 9% of organizations report a mature scaled engineering-AI program, while 80% are still in pilots or experimentation. SM018
CM035 SimScale says 74% of organizations cite data preparation and availability as the main barrier to scale, followed by governance and compliance at 48% and software interoperability at 42%. SM018
CM036 SimScale says 87% of organizations allow AI-driven pass/fail decisions only under defined oversight frameworks. SM018
CM037 SimScale says 99% of engineering leaders expect tangible ROI within 12 months and 90% report some use of agentic AI copilots or autonomous agents in workflows. SM018
CM038 TGM says surrogate models can accelerate simulation runtimes by 100x to 1000x but require clearly defined trust boundaries. SM025
CM039 TGM says PINNs remain mostly in research and pilot phases because training stability, scalability, and realistic three-dimensional deployment remain challenging. SM025
CM040 ENGtechnica says AI adoption in simulation is cautious, benchmarked against trusted solvers, and likely to follow a staged support-before-replacement path in risk-averse industries. SM026
CM041 Mordor says high HPC costs, interoperability gaps, IP security concerns, and talent scarcity restrain simulation adoption, and it cites H100 prices above USD 30,000 as one example of compute pressure. SM013
CM042 The 2025 Federal Register AI diffusion rule tightened controls on advanced computing ICs and AI model weights while creating new exceptions for approved data-center ecosystems. SM021
CM043 Cambashi says 2025–2026 market structure is being reshaped by major M&A and by blurring lines among CAE, industrial AI, EDA, and MBSE. SM017
CM044 The most defensible SAM for PhysicsX is not all simulation software but the upper end of simulation-led engineering workflows in sectors where physical validation is expensive and data plus compute are available. SM002, SM013, SM020, SM024
CM045 A precise PhysicsX-specific TAM point estimate is not supportable from public evidence, so the chapter should preserve a range and boundary logic instead of a single number. SM013, SM014, SM015, SM016, SM017
CM046 A cautious SOM framing is a narrow slice of aerospace, semiconductor, automotive, data-center, and industrial-machinery programs willing to buy AI layers on top of incumbent CAE rather than rip and replace entire engineering stacks. SM003, SM012, SM018, SM026
CM047 Switching costs are reinforced by sunk licenses, fragmented standards, and enterprises’ resistance to ecosystem lock-in. SM013, SM018
CM048 Conservative engineering workflows still require trusted high-fidelity solver benchmarking and final validation even when AI is used for screening, setup, or optimization. SM012, SM025, SM026
CM049 PhysicsX’s likely buyer motion starts with engineering leaders and simulation heads, but budget sponsorship can come from CTO, product, R&D, semiconductor-equipment, vehicle-platform, or data-center infrastructure programs. SM004, SM005, SM018
CM050 Incumbent consolidation increases the validation and distribution advantages of large platforms while also creating openings for AI-native overlays that integrate with those platforms rather than replace them. SM017, SM009, SM012
CM051 Siemens reports Digital Industries as a reportable segment inside a large automation and digitalization technology group, underscoring the scale of incumbent industrial-software competition around PhysicsX. SM011, SM012
CM052 Ansys markets simulation as a way to reduce physical testing and spans fluids, electronics, structures, chips and 3D-ICs, data centers, autonomy, and digital twins, illustrating how broad the incumbent problem surface already is. SM027, SM028
CP001 PhysicsX packages its offer as three layers—Simulation Workbench, AI Workbench, and Engineering Applications—rather than as a single point product. SP002, SP003
CP002 PhysicsX already integrates with incumbent tools including ANSYS, Siemens NX, OpenFOAM, and Siemens Star-CCM+, which lowers adoption friction at accounts standardized on legacy CAE stacks. SP002
CP003 PhysicsX frames private customer-specific models, built-in uncertainty quantification, and active learning as part of its switching-cost story because customer data and retraining loops compound over time. SP002, SP003
CP004 PhysicsX’s own brake-cooling case study uses OpenFOAM-generated CFD data and transfer learning, showing that the company layers AI on top of existing solvers rather than replacing the solver stack outright. SP004
CP005 PhysicsX’s public GTM evidence points to an embedded and forward-deployed sales motion rather than self-serve software, including its stated use of forward-deployed engineers and its embedded work with GB1. SP001, SP005
CP006 PhysicsX uses sovereign-compute and partner-led distribution as trust levers in Europe through the Deutsche Telekom Industrial AI Cloud. SP006
CP007 PhysicsX’s Simcenter X collaboration shows the company can ride incumbent channels today even while depending on a future competitor for distribution. SP007
CP008 PhysicsX’s NVIDIA standards collaboration expands ecosystem reach but also signals a willingness to open-source some architecture patterns through Opora-related tooling. SP008
CP009 Ansys now markets GeomAI, SimAI Pro, and SimAI Premium under its 2026 R1 release, showing that a major incumbent is explicitly pushing AI-first simulation workflows rather than only classical solvers. SP009, SP010
CP010 Synopsys completed its acquisition of Ansys in July 2025 and now pitches a silicon-to-systems engineering stack, which materially increases the distribution and bundling power facing AI-native startups. SP010, SP011
CP011 Siemens Simcenter remains a broad incumbent across mechanical, fluid, thermal, and systems simulation, with Xcelerator anchoring lock-in through the wider digital thread. SP012
CP012 Simcenter X Advanced packages multiple simulation domains under one named-user model plus floating tokens, which is a more flexible usage model than traditional point licenses even though rates stay private. SP013, SP014
CP013 Siemens’s own customer evidence shows AI can compress the equivalent of 2,000 turbine-design runs into minutes after training, but Siemens still describes neural operators and foundation models as next-stage goals. SP015
CP014 Altair’s absorption into Siemens and retention of Altair Units give Siemens more HPC, AI, and data-science coverage while making the incumbent bundle harder to dislodge. SP016, SP017
CP015 Cadence’s Fidelity CFD platform claims more than 10x simulation-process acceleration and AI-driven optimization, but it still attacks from an enhanced classical-CFD starting point rather than a Large Physics Model posture. SP018
CP016 Rescale is the broadest workflow-control-plane rival in the set, with hundreds of enterprise customers and more than $1 billion of annual HPC infrastructure spend transacting through the platform. SP019
CP017 Rescale’s 2026 agentic digital engineering launch makes it a direct workflow competitor to PhysicsX because it automates validation, troubleshooting, reporting, and hardware selection around simulation jobs. SP020
CP018 Rescale’s public FedRAMP, SOC 2, ISO 27001, and ITAR posture is more legible than most AI-native challengers and strengthens its position in regulated or defense-oriented procurements. SP021
CP019 Monolith is primarily an AI-for-test-data and validation platform today, so it competes most directly for engineering-analytics budgets rather than for PhysicsX’s core solver-surrogate slot. SP022
CP020 Monolith’s NAFEMS webinar on PINNs suggests an upmarket move toward simulation-adjacent workflows, which narrows the conceptual gap with CAE-native AI vendors over time. SP023
CP021 nTop is best understood as an upstream computational-design and geometry-automation platform that competes for earlier design workflow budget rather than for PhysicsX’s physics-inference slot. SP024
CP022 Akselos is a credible competitive threat in critical-infrastructure and energy accounts because it sells real-time structural performance management with quantified CAPEX and asset-life outcomes. SP025
CP023 BeyondMath is close enough to count as a direct AI-native peer because it markets a foundational physics model, cites 1000x performance discovery in Formula 1, and is using fresh capital to scale commercial deployment. SP026, SP027, SP033
CP024 BeyondMath still appears earlier in GTM maturity than PhysicsX because public evidence points to project-led commercialization and investor-backed scaling rather than a clearly exposed enterprise packaging model. SP026, SP033
CP025 OpenFOAM remains a powerful free substitute and frames proprietary CFD as roughly $50,000 per user per year, making cost avoidance a real alternative to buying another commercial layer. SP028
CP026 SU2 explicitly positions itself as free multiphysics software built to avoid proprietary or prohibitively expensive tools, reinforcing the credibility of internal-build paths for aerodynamic and optimization teams. SP029, SP030
CP027 NVIDIA PhysicsNeMo is openly available for CFD, structural mechanics, and electromagnetics model building, so the model-construction layer is increasingly commoditized for teams with talent and GPUs. SP031, SP032
CP028 The strongest internal-build substitute is a stack of existing solver data plus PhysicsNeMo plus GPU cloud, because PhysicsX itself trains on solver outputs and the open tools cover the same physics families. SP004, SP028, SP030, SP031
CP029 PhysicsX’s moat appears to rest more on proprietary industrial datasets, customer-specific fine-tuning, deployment hardening, and embedded delivery than on exclusive access to model architectures. SP001, SP003, SP006, SP008
CP030 Switching costs are highest against Siemens and Synopsys accounts because those incumbents control broader CAD, PLM, EDA, and licensing workflows rather than just individual solvers. SP011, SP012, SP014, SP016
CP031 Public pricing signals remain weak across the field: Siemens and Altair explain tokens and units but not rates, while PhysicsX, Rescale, Monolith, and BeyondMath all present as sales-led or opaque. SP013, SP014, SP017, SP019, SP022, SP026
CP032 PhysicsX is most likely to win where customers need orders-of-magnitude iteration speed on bespoke multiphysics problems and are willing to accept a high-touch deployment model. SP001, SP003, SP004, SP005
CP033 PhysicsX is most vulnerable where buyers already own broad incumbent suites or maintain strong internal ML and HPC teams, because those buyers can multi-home or build around existing solver data. SP002, SP012, SP020, SP028, SP031
CP034 Siemens is simultaneously a route to market and a displacement risk because it integrates PhysicsX today while expanding its own AI-enabled Simcenter roadmap. SP007, SP012, SP015, SP016
CP035 Adjacent entrants can attack from workflow-control-plane or hardware-software integration angles without matching PhysicsX’s model layer, as shown by Rescale and Cadence. SP018, SP020
CP036 Vertical specialists such as Akselos, Monolith, and nTop can siphon budget from PhysicsX even without matching breadth because each solves a high-value slice of the engineering workflow. SP022, SP024, SP025
CP037 Likely-entrant pressure extends beyond the named peer set because NVIDIA, Synopsys/Ansys, and Siemens all now have capital, distribution, or open tooling that can be pushed further into physics AI. SP008, SP010, SP016, SP031, SP032
CP038 Commoditization risk is material because open standards and open-source physics AI can narrow differentiation toward data, services, and deployment rather than model architecture alone. SP008, SP031, SP032
CP039 Public evidence still does not establish realized pricing, ACV, or formal sector-specific certification for PhysicsX itself at the same level of detail that Rescale and the incumbents expose. SP001, SP006, SP021
CP040 The CAE landscape is consolidating around larger suites at the same time that AI-native challengers proliferate, which raises the odds that buyers will compare PhysicsX against bundles, substitutes, and partial tools rather than against one clean peer set. SP010, SP016, SP020, SP023, SP027
CI001 PhysicsX sells an AI-native engineering platform that spans the full product lifecycle from design and simulation through manufacturing and operations. SI001, SI003
CI002 PhysicsX says its platform combines fast AI-driven physics inference with numerical simulation to cut simulation runtimes from hours or days to seconds. SI003, SI014
CI003 Public official materials describe Simulation Workbench, AI Workbench, Engineering Applications, a model catalog, data unification, and platform services as core platform components. SI003
CI004 PhysicsX says the platform supports AWS and Azure natively and can also be deployed in hybrid or on-prem environments. SI003
CI005 PhysicsX says the platform integrates with ANSYS, CATIA, Siemens NX, OpenFOAM, and Siemens Star CCM+ to fit existing engineering workflows. SI003
CI006 PhysicsX states that each customer’s data is used only for that customer’s models and is not used to train models for other clients. SI003
CI007 PhysicsX says it does not deploy software and step back; instead, forward-deployed engineers embed directly into customer programs that are already live. SI002
CI008 PhysicsX says its delivery team works with customers to tailor model architectures, simulation pipelines, and optimization loops to each customer context. SI003
CI009 Official and independent sources place PhysicsX across aerospace, defense, automotive, semiconductors, materials, energy, and industrial machinery use cases. SI001, SI006, SI016
CI010 No public list pricing or self-serve enterprise rate card was found on the official homepage, platform page, or newsroom materials reviewed for this chapter. SI001, SI003, SI005
CI011 The only public product-led pricing signal found is Ai.rplane, which The Next Web described as a free barebones version of PhysicsX’s LGM-Aero tool. SI027
CI012 Careers and Greenhouse materials confirm a UK headquarters with London and New York offices plus UK and US relocation support, consistent with a field-heavy enterprise buildout. SI004, SI020
CI013 PhysicsX’s Series B press release said the company’s technology was already embedded in the workflows of sophisticated engineering and manufacturing organizations. SI007
CI014 The Business Times and GuruFocus reported that Applied Materials, Siemens, and Stellantis are current PhysicsX customers. SI012, SI032
CI015 PhysicsX’s Series C press release said recognized revenue doubled year over year. SI006, SI015
CI016 PhysicsX’s Series C press release said booked revenue tripled over the prior year. SI006, SI016
CI017 PhysicsX’s Series C press release said customer count more than doubled over the prior year. SI006, SI015
CI018 PhysicsX’s Series C press release said the team had grown to more than 300 people and doubled in size in the previous twelve months. SI006, SI016
CI019 The Business Times and Yahoo Finance reported that PhysicsX expected revenue to be close to $50 million in 2026. SI012, SI028
CI020 The Business Times and Yahoo Finance reported that management aims to more than double revenue in 2027. SI012, SI028
CI021 The Business Times, Yahoo Finance, and GuruFocus reported that PhysicsX had roughly a six-month backlog of customer demand and needed more cash partly to expand staff. SI012, SI028, SI032
CI022 The Business Times and Yahoo Finance reported that semiconductors are expected to become PhysicsX’s largest segment by the end of Q2 2026. SI012, SI028
CI023 Tech.eu and The Next Web placed PhysicsX headcount around 350 by June 2026, above the company’s official “300+” phrasing. SI013, SI014
CI024 The Next Web argued that AI data-center infrastructure demand is a growth engine for PhysicsX because chip, cooling, and power systems all require heavy engineering simulation. SI014
CI025 Public evidence supports a hybrid commercial model in which enterprise software revenue is paired with material service-delivery labor because PhysicsX embeds engineers and customizes workflows. SI002, SI003, SI007
CI026 PhysicsX’s Deutsche Telekom press release says the platform runs natively on NVIDIA accelerated computing and CUDA-X and integrates directly with NVIDIA PhysicsNeMo. SI009, SI019
CI027 Deutsche Telekom said its Industrial AI Cloud, where PhysicsX is a software-layer partner, is being built with more than 1,000 DGX B200 systems and up to 10,000 Blackwell GPUs. SI018
CI028 PhysicsX’s automotive aerodynamics case study said its Data Factory generated more than 20,000 CFD simulations from more than 250 baseline vehicle designs. SI010, SI016
CI029 PhysicsX’s platform FAQ emphasizes uncertainty quantification and active learning, implying ongoing validation and retraining work rather than one-off model delivery. SI003
CI030 PhysicsX’s neural-operator research blog explicitly discusses where a Fourier Neural Operator falls short, showing the company still has to invest in core research to improve model performance. SI011
CI031 No reviewed public source disclosed gross margin or the split between software gross profit and service delivery cost. SI001, SI003, SI006, SI012, SI013
CI032 No reviewed public source disclosed CAC, payback, NRR, or churn. SI001, SI003, SI006, SI012, SI013
CI033 PhysicsX announced a $300 million Series C at an approximately $2.4 billion valuation led by Temasek. SI006, SI012, SI013
CI034 PhysicsX announced a $135 million initial Series B financing led by Atomico in June 2025. SI007, SI024
CI035 PhysicsX announced in November 2025 that its Series B total exceeded $155 million and valued the company at nearly $1 billion. SI008, SI024
CI036 Public secondary sources place PhysicsX’s total capital raised by mid-2026 at roughly $487 million to $500 million. SI013, SI031
CI037 Companies House search identifies PhysicsX Limited as company number 12134466, incorporated on 1 August 2019, at Victoria House, 1 Leonard Circus, London. SI022
CI038 Companies House filing history shows PhysicsX filed group accounts made up to 31 December 2024 on 29 October 2025. SI023
CI039 Companies House filing history shows SH01 statements of capital or share allotment activity on 5 February 2026 and 20 April 2026. SI023
CI040 Official and independent coverage say Series C proceeds will fund global growth, US expansion, a Singapore office, platform capability expansion, and frontier research on larger physics models. SI006, SI013
CI041 No reviewed public source disclosed cash on hand, burn rate, runway, debt, or project-finance obligations. SI006, SI012, SI013, SI023
CI042 New Market Pitch argued that a $2.4 billion valuation against roughly $50 million of 2026 revenue implies an approximately 48x current-year revenue multiple. SI029
CI043 New Market Pitch argued that the gap between tripled bookings and merely doubled recognized revenue could mean demand is outrunning delivery and that economics depend on deployments becoming repeatable software rather than bespoke engineering. SI029
CI044 New Market Pitch said the key unresolved question is whether PhysicsX is a scalable software platform or a services-heavy engineering shop. SI029
CI045 The public record supports a mixed financial verdict: growth and customer demand are credible, but margin quality and capital efficiency cannot be underwritten without private data on revenue mix, gross margin, and delivery intensity. SI006, SI012, SI029, SI031
CI046 No reviewed public source disclosed ARR or the split between recurring platform revenue and services revenue. SI006, SI012, SI013
CI047 Series A and Series B coverage both said new capital would be used to expand customer delivery, platform engineering or global expansion, and fundamental research. SI025, SI007
CE001 PhysicsX publicly positions its product as an AI-native engineering software stack spanning design, manufacturing, and operations. SE001, SE002, SE005
CE002 The public platform stack is organized around Simulation Workbench, AI Workbench, Engineering Applications, Model Catalog, Data Unification, and Platform Services. SE002
CE003 PhysicsX says its platform unifies simulation, physics AI, data, and engineering applications instead of treating AI as an add-on to legacy tools. SE001, SE002, SE005
CE004 PhysicsX says the platform is purpose-built for the full product lifecycle from concept and design through manufacturing and operations. SE001, SE002
CE005 The product promise is to combine fast AI-driven physics inference with numerical simulation so engineers can iterate faster while preserving access to solver-grade reference methods. SE002, SE005, SE011
CE006 PhysicsX publicly lists integrations with ANSYS, CATIA, Siemens NX, OpenFOAM, and Siemens STAR-CCM+. SE002
CE007 PhysicsX publicly supports AWS and Azure natively and also says the platform can deploy to hybrid, on-prem, and air-gapped environments. SE002, SE005, SE012, SE014, SE022, SE023
CE008 PhysicsX says each customer’s data is used only to train models specific to that customer environment and is not used to train models for other clients. SE002
CE009 AI Workbench is described as a unified environment to develop, train, fine-tune, and deploy physics AI models with both low-code interfaces and full programmatic access. SE005
CE010 Engineering Applications are described as web interfaces, APIs for toolchain integration, and edge deployments for specialized manufacturing or operational environments. SE005
CE011 Simulation Workbench is described as a unified data foundation that automates simulation workflows and merges simulation, experimental, and operational data into a traceable system of record. SE005, SE009
CE012 PhysicsX frames simulation automation as a modular orchestration problem spanning geometry, mesh, setup, solve, post-process, and KPI extraction with dependency handling. SE009
CE013 PhysicsX says its simulation data backbone links geometry, mesh, configuration, boundary conditions, parameters, and outputs into a searchable simulation database. SE009
CE014 PhysicsX says its models include uncertainty quantification and active learning loops that trigger new targeted data generation when uncertainty is high. SE002, SE005, SE008, SE011
CE015 PhysicsX says its Delivery team works alongside customer teams to integrate the platform into existing engineering workflows and to encode reusable patterns back into the product. SE005, SE016, SE031
CE016 PhysicsX says its simulation-engineering organization includes 35 simulation engineers spanning domains such as CFD, FEA, electromagnetics, chemistry, and material science. SE016
CE017 PhysicsX explicitly says simulations are not ground truth and that engineers remain responsible for judging when results are meaningful. SE016
CE018 PhysicsX repeatedly claims that workflows that once took hours or days can be compressed to seconds through AI inference. SE002, SE005, SE022, SE023
CE019 LGM-Aero is PhysicsX’s public Large Geometry Model for aerospace engineering and the company says it has about 100 million parameters and has seen 25 million diverse 3D shapes. SE006, SE015, SE024, SE027
CE020 PhysicsX says its geometry foundation model maps any geometry into a 512-dimensional latent code that can support downstream prediction and optimization. SE006, SE018
CE021 PhysicsX says LGM-Aero training used AWS Batch, S3, DynamoDB, FSx for Lustre, EFA, and multi-GPU clusters including 128 H100s followed by 64 A100s. SE006, SE023
CE022 PhysicsX and independent coverage say LGM-Aero was built on more than 25 million meshes and tens of thousands of CFD and FEA simulations generated with Siemens tools. SE015, SE020, SE024, SE027
CE023 PhysicsX says LGM-Aero generalizes across a broad set of aeroelastic applications and can perform zero-shot inference of aero performance, flight stability, and structural stress. SE015, SE020, SE024, SE027
CE024 Ai.rplane is a public reference application on airplane.physicsx.ai that PhysicsX presents as a showcase for LGM-Aero. SE015, SE027, SE032
CE025 PhysicsX says Ai.rplane can generate novel aircraft designs and instantly predict lift, drag, stability, structural stress, and related performance attributes. SE015, SE024, SE026, SE027
CE026 Siemens and PhysicsX describe their collaboration as combining high-fidelity CAE data and robust CAE-AI integrations to build deep physics simulation workflows. SE011, SE020
CE027 The Simcenter X collaboration extends the product into SaaS-based HPC and remote-desktop CFD environments for AI-assisted design and optimization. SE011
CE028 PhysicsX says its Microsoft collaboration integrates the platform with Microsoft Discovery and makes a private release available through the Azure Marketplace. SE012, SE022
CE029 PhysicsX says its Deutsche Telekom and T-Systems deployment puts the platform on sovereign European AI infrastructure powered by NVIDIA accelerated computing. SE014, SE030
CE030 PhysicsX says its CoreWeave partnership gives customers high-throughput GPU infrastructure to train private Large Physics Models and PhysicsX’s own frontier pretrained models. SE013, SE029
CE031 PhysicsX says private foundation models can be built by fine-tuning pretrained Large Physics Models on proprietary customer data. SE005, SE013, SE014
CE032 PhysicsX says its platform can combine proprietary models with third-party model families such as NVIDIA PhysicsNeMo and Apollo. SE005, SE031
CE033 Siemens’ launch of Simcenter PhysicsAI in 2026 shows that incumbent CAE vendors are productizing AI surrogate tooling directly inside simulation environments. SE021
CE034 PhysicsX publicly announced ISO 27001 certification and says the certification covers financial data, intellectual property, employee information, and third-party data. SE010
CE035 PhysicsX’s privacy notice says the company’s privacy policy complies with the UK Data Protection Act 2018 and the EU GDPR. SE019
CE036 The privacy notice says PhysicsX limits access to personal data to employees and contractors with a business need to know and uses ICO-approved safeguards for international transfers. SE019
CE037 PhysicsX says the platform can run hosted, in customer clouds, or fully air-gapped and that it holds both SOC 2 and ISO 27001 certifications. SE005
CE038 PhysicsX’s own materials present validation as a continuing workflow that pairs AI with high-fidelity simulation, real-world data, and engineer judgment rather than autonomous model-only release. SE005, SE011, SE016, SE031
CE039 PhysicsX’s FNO explainer says vanilla FNO is a poor fit for shock-dominated flows, strict boundary-condition enforcement, irregular geometries, highly localized fine-scale features, very scarce training data, and problems above three dimensions. SE007
CE040 PhysicsX’s geometry-model explainer says LGM deployment may require domain-specific fine-tuning when geometries sit far outside the training distribution or when subtle variations must be captured precisely. SE018
CE041 PhysicsX’s geometry-model explainer says its latent representation is lossy and relies on strong prior assumptions about latent-space structure. SE018
CE042 PhysicsX’s December 2024 deep-dive said a full technical paper with LGM-Aero benchmarks was still forthcoming. SE006
CE043 The public materials reviewed here do not expose uptime metrics, incident history, or enterprise SLA terms for the platform. SE002, SE005, SE010, SE019
CE044 The public materials reviewed here do not provide a detailed public enterprise model-governance or customer-data-retention whitepaper beyond privacy-policy and certification-level statements. SE002, SE010, SE019
CE045 The public materials reviewed here do not describe a public export-control or defense-data handling program despite Aerospace & Defense being a target vertical. SE003, SE014, SE019
CE046 PhysicsX says Delivery and Simulation Engineering encode repeatable data pipelines, orchestration strategies, and application interfaces back into productized templates. SE005, SE016
CE047 CDFAM’s 2026 interview quotes PhysicsX saying Large Physics Models are already deployed in mission-critical programs and integrated into existing toolchains rather than waiting in a research queue. SE031
CE048 Microsoft’s 2026 feature says PhysicsX has applied its platform to Microsoft Surface cooling-fan design and to copper-extraction optimization, supporting manufacturing and operations use cases beyond concept design. SE022
CE049 Because Siemens is embedding its own PhysicsAI add-on into STAR-CCM+, part of PhysicsX’s workflow acceleration thesis could migrate into incumbent CAE suites rather than remain unique to an overlay platform. SE011, SE021
CE050 AWS says the PhysicsX platform runs on services including EKS for platform workloads and AWS Batch for model training, which supports the claim that the product is already engineered for cloud-scale operation. SE023
CU001 PhysicsX said in June 2026 that its platform was deployed across aerospace and defense, semiconductors, industrial machinery, automotive, energy, and materials. SU001
CU002 PhysicsX said its customer count more than doubled over the prior year by June 2026. SU001, SU026
CU003 PhysicsX said recognized revenue doubled year over year by June 2026. SU001, SU026
CU004 PhysicsX said booked revenue tripled by June 2026. SU001, SU026
CU005 Bloomberg syndication said PhysicsX customers include Applied Materials, Siemens, and Stellantis. SU013, SU037
CU006 Bloomberg syndication said semiconductors were expected to become PhysicsX's largest segment by the end of Q2 2026. SU013, SU037
CU007 Bloomberg syndication said PhysicsX had a roughly six-month backlog of customer demand in June 2026. SU013, SU037
CU008 Bloomberg syndication said PhysicsX expected revenue close to $50 million in 2026 and aimed to more than double that figure in 2027. SU013, SU037
CU009 Sifted reported that Corbo said PhysicsX was supply-side limited and was moderating rollout to existing customers because of strong demand. SU014
CU010 Siemens CTO Peter Koerte said Siemens' strategic investment in PhysicsX built on successful collaboration in AI-based deep-physics simulations. SU002
CU011 PhysicsX and Siemens announced a March 2026 collaboration on power-distribution systems for next-generation AI data centers. SU004, SU020
CU012 PhysicsX said the Siemens data-center workflow compressed analyses that previously took days or more than 24 hours into seconds. SU004, SU020
CU013 Siemens said PhysicsX lets engineers predict thermal behavior in complex busway systems in real time and iterate faster. SU020, SU025
CU014 PhysicsX and Deutsche Telekom announced a three-year strategic partnership under which T-Systems would onboard industrial customers and mobilize forward-deployed squads at customer sites. SU012, SU007
CU015 PhysicsX was a launch partner on Deutsche Telekom's Industrial AI Cloud and later said its platform was live on that cloud. SU018, SU007
CU016 Deutsche Telekom said several companies were already using its AI factory capacity, including PhysicsX, and that the facility was already operating at more than one-third of capacity with existing customers. SU019
CU017 Siemens said customers such as Mercedes-Benz and BMW Group could use the Industrial AI Cloud for AI-powered digital twins, indicating that the broader PhysicsX ecosystem is oriented toward large OEM procurement rather than self-serve usage. SU018
CU018 PhysicsX said its platform connects to Siemens Teamcenter and interoperates with Simcenter without disrupting established workflows. SU008
CU019 PhysicsX said the CoreWeave partnership lets enterprise customers train private Large Physics Models on proprietary data and deploy them within secure enterprise environments. SU006, SU023
CU020 CoreWeave said PhysicsX was already delivering production-grade physical AI on its cloud across aerospace, automotive, semiconductors, materials, and energy. SU023
CU021 PhysicsX became GB1's official AI Engineering Platform partner for the 38th America's Cup campaign. SU005, SU035, SU036
CU022 PhysicsX said it works alongside GB1 engineers in Portsmouth as an embedded partner integrated into active development programs. SU005, SU036
CU023 GB1's Head of Design said PhysicsX gave designers higher-fidelity models, lower data costs, quicker turnaround, and faster design iteration. SU005, SU035, SU036
CU024 Microsoft said PhysicsX significantly reduced the time needed to develop new equipment prototypes in semiconductor manufacturing. SU021
CU025 Microsoft said PhysicsX improved thermal behavior in Microsoft Surface devices by enabling engineers to test many more cooling-fan design variations. SU021
CU026 Microsoft said PhysicsX was working with an unnamed global leader to improve the efficiency of copper extraction in mining and metals. SU021
CU027 Microsoft said PhysicsX built its engineering stack on Azure with high-performance infrastructure and security suited to mission-critical engineering environments. SU021, SU022
CU028 PhysicsX said it was one of 13 companies selected from more than 500 applications for Microsoft's Agentic Launchpad and would receive go-to-market acceleration across Microsoft's ecosystem. SU009
CU029 PhysicsX's automotive technical note said many strong public benchmark results come from in-distribution evaluation that does not match how industry actually uses these tools. SU010
CU030 PhysicsX's automotive technical note said state-of-the-art architectures still generalize poorly to out-of-distribution design and cross-simulator settings. SU010
CU031 PhysicsX said a true automotive Large Physics Model must generalize to whatever car design a customer brings rather than to small variations of known training designs. SU010
CU032 PhysicsX's careers board showed 34 open roles in June 2026 across delivery, product, and research, including forward-deployed and security roles in London, New York, San Francisco, and Singapore. SU034
CU033 Observer reported that an unnamed aerospace client used PhysicsX to cut jet-engine turbine-blade scrap rates by 70 percent. SU029
CU034 Observer reported that another PhysicsX program improved artificial-heart efficiency and reduced blood damage by 42 percent, but the customer was unnamed and outside the core industrial vertical set for this chapter. SU029
CU035 Independent June 2026 coverage from TNGlobal, Tech.eu, Digital Engineering 24/7, and NGP corroborated the Series C, valuation, and continued industrial-adoption narrative. SU015, SU016, SU026, SU027
CU036 EU-Startups and MarketScreener corroborated that Siemens and Applied Materials were already in the 2025 financing syndicate before Bloomberg later named them as customers. SU028, SU030
CU037 PhysicsX's Deutsche Telekom and CoreWeave announcements both positioned compute and deployment infrastructure as a route to faster industrial onboarding rather than as proof of end-customer spend by themselves. SU012, SU007, SU006
CU038 None of the reviewed public sources disclosed PhysicsX's NRR, GRR, churn rate, contract length, or renewal cadence. SU001, SU013, SU014, SU021, SU031
CU039 Public customer proof is strongest for Siemens, GB1, and Microsoft Surface because those cases include specific workloads or counterparty quotes, while Applied Materials and Stellantis are named without public workload detail. SU013, SU037, SU004, SU020, SU005, SU035, SU021
CU040 Public evidence supports semiconductor, automotive, data-center, and high-performance marine use cases, but no reviewed source publicly names a defense end customer. SU001, SU013, SU021, SU005
CU041 Deutsche Telekom and T-Systems are evidenced as channel and deployment partners for European industrial companies rather than as clean proof of recurring end-customer revenue. SU012, SU007, SU019
CU042 CoreWeave and Microsoft are better evidenced as platform enablers than as independent end customers, except that Microsoft publicly described internal use on Surface devices. SU006, SU023, SU021, SU022
CU043 The strongest customer-proof artifacts cluster in March through June 2026, making adoption momentum current, but retention durability evidence remains absent. SU001, SU004, SU005, SU006, SU007, SU013, SU021
CU044 PhysicsX's investor list overlaps with named customers and platform partners such as Siemens, Applied Materials, and NVIDIA, which increases strategic alignment but weakens testimonial independence. SU001, SU002, SU028, SU030
CU045 Public evidence suggests procurement friction centers on integration into mission-critical workflows, secure enterprise deployment, and implementation capacity rather than on lack of demand. SU014, SU021, SU022, SU012, SU006, SU032, SU031
CU046 Only three publicly named paying customers were independently identified in the reviewed corpus by June 2026: Applied Materials, Siemens, and Stellantis. SU013, SU037
CU047 Because the public record names only three customers against a roughly $50 million 2026 revenue outlook, concentration risk could be material even though exact top-account share is undisclosed. SU013, SU037
CU048 Public evidence separates active deployments from ecosystem announcements: Siemens, GB1, and Microsoft Surface show concrete workloads, while Deutsche Telekom, CoreWeave, and Agentic Launchpad are enablement partnerships. SU004, SU020, SU005, SU035, SU021, SU012, SU006, SU009
CR001 PhysicsX publicly targets advanced-industrial sectors including aerospace & defense, semiconductors, automotive, materials, energy, and data-center infrastructure. SR004, SR008, SR010, SR016
CR002 PhysicsX positions its platform across the full engineering lifecycle from design through manufacturing and operations, with integrations into incumbent CAE toolchains. SR001, SR003
CR003 PhysicsX says platform outputs are approximations and pairs them with uncertainty quantification, confidence context, and active learning around high-uncertainty areas. SR003, SR013
CR004 PhysicsX technical material says standard FNO approaches are poor fits for shocks, discontinuities, irregular geometry, scarce data, some high-dimensional problems, and strict boundary-condition requirements. SR012
CR005 PhysicsX says it is purpose-built for complex, mission-critical engineering challenges, increasing the importance of liability allocation when model limits are exceeded. SR003, SR026
CR006 Because PhysicsX names aerospace & defense, semiconductors, and energy as target sectors, export-control and sanctions screening is relevant to its go-to-market even without a disclosed enforcement event. SR004, SR008, SR010, SR030, SR031
CR007 BIS says exporters must determine what is subject to the EAR and whether a license or exception is required before export. SR030
CR008 OFAC maintained active sanctions programs across multiple jurisdictions, including Russia-, Iran-, North Korea-, Belarus-, and cyber-related programs, on the June 2026 page reviewed for this chapter. SR031
CR009 PhysicsX's sovereign-European-compute story depends on Deutsche Telekom, T-Systems, and NVIDIA-backed infrastructure, making geopolitical and compliance execution part of the growth thesis. SR018, SR019, SR029
CR010 PhysicsX's public legal pack on the website is a privacy notice whose text says the version was last updated in October 2022. SR007
CR011 The public privacy notice still lists the former Shoreditch High Street address, while Companies House now shows the registered office as Victoria House, 1 Leonard Circus, London. SR007, SR024, SR025
CR012 PhysicsX's privacy notice says the company relies on GDPR and the UK Data Protection Act 2018 and directs complainants to the ICO. SR007, SR033
CR013 PhysicsX says it obtained ISO 27001 certification and that the certification covers financial data, intellectual property, employee information, and third-party data. SR011
CR014 The extracted PhysicsX public site links surfaced a privacy notice but did not surface a public trust portal, DPA page, security page, or public terms page. SR001, SR002, SR003, SR005, SR006, SR007
CR015 Companies House shows an active private company with share allotment, accounts, and governance filings, but the retained registry materials do not disclose product-liability or litigation outcomes. SR024, SR025
CR016 PhysicsX says its models can predict physical behavior in seconds rather than hours or days, enabling engineering teams to evaluate far more design variants. SR008, SR026
CR017 PhysicsX's FNO article says neural operators can evaluate a new PDE instance in milliseconds rather than re-solving it from scratch. SR012
CR018 The same FNO article says standard FNO does not guarantee strict boundary-condition satisfaction and often needs explicit BC channels or geometry-aware variants. SR012
CR019 PhysicsX's technical writing says very high-frequency structure uncorrelated with retained low-frequency modes cannot be recovered and that shock-dominated flows are weak fits for vanilla FNO. SR012
CR020 PhysicsX's uncertainty article says downstream decisions require credible intervals, confidence bounds, or similar uncertainty measures rather than bare point estimates. SR013
CR021 The same uncertainty article says calibration is necessary because model confidence can diverge from true prediction error. SR013
CR022 PhysicsX says customer relationships begin with jointly scoped pilots that usually last one to three months and require falsifiable success criteria. SR014
CR023 PhysicsX says forward-deployed engineers embed directly inside customer programs rather than simply shipping software and stepping back. SR002, SR014
CR024 PhysicsX says each customer's data trains only customer-specific models and is not used to train models for other clients. SR003
CR025 PhysicsX says the platform supports AWS and Azure natively and can also deploy to hybrid or on-prem environments to support high-classification requirements. SR003
CR026 Microsoft Stories says PhysicsX built its engineering stack on Microsoft Azure with high-performance computing and security for mission-critical engineering environments. SR026
CR027 PhysicsX joined the 2024 AWS Generative AI Accelerator to leverage AWS infrastructure and expertise for training Large Physics Models at scale. SR020
CR028 PhysicsX's Microsoft collaboration made it a featured launch partner for Microsoft Discovery and a private Azure Marketplace release. SR017
CR029 PhysicsX says it will train its largest foundation models on Deutsche Telekom's Industrial AI Cloud and use T-Systems for customer onboarding in Europe. SR018, SR019
CR030 PhysicsX announced in February 2026 that its platform was live on Deutsche Telekom's Industrial AI Cloud powered by NVIDIA accelerated computing. SR019
CR031 PhysicsX says frontier Large Physics Models require sustained high-throughput GPU compute across massively distributed workloads. SR015
CR032 The CoreWeave partnership offers customers secure enterprise environments for training private domain-specific Large Physics Models and deploying them in production. SR015
CR033 PhysicsX's Siemens collaboration extends the company into AI data-center power infrastructure and NVIDIA Omniverse-linked engineering workflows. SR016
CR034 Microsoft Stories says PhysicsX has applied its models to Microsoft Surface cooling and to mining-and-metals optimization workflows. SR026
CR035 PhysicsX publicly moved from a $135 million Series B in June 2025 to more than $155 million after the November 2025 extension and then to a $300 million Series C at about a $2.4 billion valuation in June 2026. SR008, SR009, SR010, SR027, SR029
CR036 tech.eu reported that PhysicsX's June 2026 valuation more than doubled from the prior roughly $1 billion level within about twelve months and that total funding reached around $500 million. SR027, SR029
CR037 PhysicsX says recognized revenue doubled year over year, booked revenue tripled, and customer count more than doubled in the year before the Series C. SR008
CR038 PhysicsX said in June 2025 that revenue had more than quadrupled over the prior two years and that headcount had grown to more than 150. SR010
CR039 PhysicsX said in June 2026 that the team had grown to more than 300 people, doubling in size over the prior twelve months. SR008
CR040 tech.eu reported that PhysicsX employed around 350 people by June 2026. SR027
CR041 Public fundraising and partnership pages repeatedly say new capital will fund global expansion, larger foundation models, and frontier research, implying continuing capital intensity after the Series C. SR008, SR015, SR018, SR019, SR021
CR042 PhysicsX's own pilot guidance says commercialization succeeds only if hard engineering pilots convert into useful, falsifiable, and eventually independent customer workflows. SR014
CR043 PhysicsX's careers page says the company provides relocation support and, in the UK, visa sponsorship where applicable for new hires. SR005
CR044 PhysicsX added Chris Wigley as COO in November 2025 and shifted former COO Alexander Dreismann into a chief-strategy role as the company scaled globally. SR021
CR045 PhysicsX's public team pages show a bench spanning research, product, delivery, simulation, strategy, and North American leadership, but the mix highlights continued dependence on scarce multidisciplinary talent. SR002, SR005, SR021
CR046 Companies House filing history shows board and governance evolution in 2025 and 2026, including Laura Connell's appointment and James Parker Baum's details change. SR025
CR047 ISO 27001, customer-specific model training, uncertainty quantification, and hybrid deployment are real public mitigants, but none removes partner, liability, or commercialization risk by itself. SR003, SR011, SR013, SR015, SR018
CR048 The clearest thesis-break signals are a visible model failure in production, a material cloud or GPU disruption, or proof that pilot-heavy delivery is not converting into scalable software economics. SR012, SR014, SR015, SR017, SR018, SR019
CR049 Public evidence supports live privacy, security, and compliance activity, but not public proof of sector-specific approvals, export-license history, or enterprise liability terms. SR007, SR011, SR024, SR025, SR033
CR050 PhysicsX repeatedly frames advanced-industry engineering as constrained by resource and skill bottlenecks and says its teams embed in live programs, making delivery throughput and scarce engineering talent central execution constraints. SR002, SR005, SR010, SR021
CR051 Across the public fundraising, registry, and news materials reviewed for this chapter, gross margin, burn, runway, renewal quality, and customer concentration remain undisclosed. SR008, SR009, SR010, SR024, SR025, SR027
CR052 The retained public materials did not surface a public status page, uptime history, incident archive, or product SLA that would let an outside investor verify resilience directly. SR001, SR003, SR011
CV001 PhysicsX operates as PHYSICSX LIMITED, an active private limited company incorporated on 1 August 2019 with registered office at Victoria House, 1 Leonard Circus, London EC2A 4DQ. SV001, SV011
CV002 PhysicsX publicly lists offices in London and New York, consistent with the UK registered office shown in Companies House records. SV001, SV002, SV011
CV003 Companies House records show the company previously used the name MOTODYNAMICS LTD before changing to PhysicsX. SV011
CV004 PhysicsX announced an oversubscribed $300 million Series C financing at an approximately $2.4 billion valuation on 8 June 2026. SV005, SV013
CV005 The Series C round was led by Temasek, with M&G Investments and Intrepid Growth Partners joining existing investors including Applied Materials, Atomico, General Catalyst, July Fund, NGP, NVIDIA, Radius, and Siemens. SV005, SV013
CV006 PhysicsX said Temasek first invested in 2025 before leading the 2026 Series C. SV005
CV007 Pulse 2.0 reported that PhysicsX's Series B extension pushed total Series B funding above $155 million and valuation toward $1 billion. SV014
CV008 The public valuation narrative moved from roughly $1 billion in the 2025 Series B extension to approximately $2.4 billion in the 2026 Series C, implying about a 2.4x step-up in roughly a year. SV005, SV014
CV009 PhysicsX said recognized revenue doubled year over year ahead of the Series C. SV005, SV013
CV010 PhysicsX said booked revenue tripled year over year ahead of the Series C. SV005, SV013
CV011 PhysicsX said customer count more than doubled over the prior year. SV005, SV013
CV012 PhysicsX said it had grown to more than 300 people and doubled headcount over the prior twelve months. SV005, SV010, SV013
CV013 PhysicsX describes its product as an AI-native engineering platform spanning simulation management, AI workbench tooling, and engineering applications across the product lifecycle. SV003, SV006
CV014 The company says the platform is enterprise-ready with multicloud, hybrid, and on-prem deployment options plus security-first controls. SV003
CV015 PhysicsX says its platform integrates with ANSYS, CATIA, Siemens NX, OpenFOAM, and Siemens Star CCM+ workflows. SV003
CV016 PhysicsX says each customer's data is used only to train models for that customer and not reused across other clients. SV003
CV017 PhysicsX publicly highlights deployment relevance across aerospace and defense, semiconductors, materials, automotive, and energy. SV004, SV005
CV018 PhysicsX announced a March 2026 collaboration with Siemens Smart Infrastructure on power-distribution design for next-generation AI data centers powered by NVIDIA. SV007
CV019 In that Siemens collaboration, PhysicsX said analyses that previously took days of simulation can be performed in seconds. SV007
CV020 PhysicsX announced a separate March 2026 collaboration with NVIDIA around open standards for physics AI architecture, including PhysicsNeMo and its own Opora framework. SV009
CV021 PhysicsX frames Large Physics Models as a central roadmap ambition for its future platform expansion. SV005, SV009
CV022 PhysicsX's own technical note says vanilla Fourier Neural Operators struggle with shocks, sharp discontinuities, strict boundary conditions, irregular geometry, and higher-dimensional problems. SV008
CV023 Because those limitations are material, commercial success depends on deploying PhysicsX into workloads where solver-approximation assumptions remain valid and uncertainty can be managed. SV003, SV008
CV024 PhysicsX's careers page shows distinct research, product, delivery, and operations teams and says most staff are in the office two to three days per week. SV010
CV025 Companies House shows that PhysicsX filed group accounts made up to 31 December 2024 in October 2025. SV011, SV012
CV026 Companies House filing history shows statements of capital following share allotments in February and April 2026, consistent with fresh equity issuance around the latest financing period. SV011, SV012
CV027 Using CompaniesMarketCap for market value and Macrotrends for trailing revenue, Ansys was roughly $32.90 billion on about $2.468 billion of trailing revenue, implying around 13.3x market-cap-to-revenue. SV015, SV016
CV028 Using CompaniesMarketCap for market value and Macrotrends for trailing revenue, Cadence was roughly $106.84 billion on about $5.213 billion of trailing revenue, implying around 20.5x market-cap-to-revenue. SV018, SV019
CV029 Using CompaniesMarketCap for market value and Macrotrends for trailing revenue, PTC was roughly $13.25 billion on about $2.739 billion of trailing revenue, implying around 4.8x market-cap-to-revenue. SV021, SV022
CV030 Using CompaniesMarketCap for market value and Macrotrends for trailing revenue, Autodesk was roughly $40.92 billion on about $6.888 billion of trailing revenue, implying around 5.9x market-cap-to-revenue. SV024, SV025
CV031 Across the four sampled public comparables, rough market-cap-to-revenue multiples span about 4.8x to 20.5x with a median near 9.6x. SV015, SV016, SV018, SV019, SV021, SV022, SV024, SV025
CV032 At PhysicsX's current $2.4 billion valuation, the implied valuation multiple is about 48x at $50 million of recurring revenue, 24x at $100 million, 16x at $150 million, and 12x at $200 million. SV005, SV015, SV016, SV018, SV019, SV021, SV022, SV024, SV025
CV033 To look merely in-family with the top end of the sampled public band, PhysicsX likely needs at least roughly $120 million of annualized recurring revenue, and closer to roughly $250 million to sit near the sampled median. SV005, SV015, SV016, SV018, SV019, SV021, SV022, SV024, SV025
CV034 Public sources disclose only growth rates, headcount, and high-level partner narratives, not the absolute revenue base, gross margin, retention, software-versus-services mix, or preference stack required to fully underwrite the $2.4 billion price. SV005, SV012, SV013
CV035 The bull thesis is that PhysicsX converts its current growth and partner momentum into productized recurring revenue that scales like premium industrial software rather than bespoke engineering services. SV003, SV005, SV007, SV009
CV036 The base case is that PhysicsX is strategically important and growing fast, but still too opaque on economics to justify a clean buy recommendation at $2.4 billion. SV005, SV012, SV013
CV037 The bear thesis is that deployments remain services-heavy or technically narrow, causing the business to re-rate toward mature industrial-software valuation bands if repeatability disappoints. SV008, SV015, SV016, SV018, SV019, SV021, SV022, SV024, SV025
CV038 Because the round moved from roughly a $1 billion valuation narrative in 2025 to about $2.4 billion in 2026, the latest price already embeds expectations for sustained hypergrowth rather than ordinary industrial-software execution. SV005, SV014
CV039 The public-evidence recommendation is research-more rather than buy at the current valuation. SV005, SV012, SV013, SV015, SV016, SV018, SV019, SV021, SV022, SV024, SV025
CV040 Confidence in that recommendation is medium because the product and partner signals are credible but the critical financial inputs remain private. SV005, SV007, SV009, SV012, SV013
CV041 Risk rating is high because both execution risk and valuation risk hinge on undisclosed metrics and hard industrial deployment proof. SV005, SV008, SV012, SV013
CV042 Valuation stance is stretched because the price appears ahead of what public evidence can currently justify. SV005, SV012, SV013, SV015, SV016, SV018, SV019, SV021, SV022, SV024, SV025
CV043 Price discipline is to re-engage only if private diligence shows more than roughly $100 million of recurring software revenue with strong retention and margin quality, or if the entry valuation resets meaningfully lower. SV005, SV012, SV013, SV015, SV016, SV018, SV019, SV021, SV022, SV024, SV025
CV044 The highest-priority diligence request is a cohort-level revenue and retention pack that reconciles booked growth, recognized growth, gross margin, and services intensity by vertical. SV005, SV012, SV013
CV045 The second gating diligence request is the full cap table with liquidation preferences, anti-dilution terms, and option-pool math because return outcomes cannot be trusted without it. SV012
CV046 One thesis-break trigger is failure to convert current lighthouse programs into repeat multi-site or multi-program deployments over the next 12 to 18 months. SV005, SV007, SV009
CV047 Another thesis-break trigger is evidence that PhysicsX's commercial wins depend mainly on narrow workloads where neural-operator limitations constrain repeatability. SV007, SV008
CV048 A third thesis-break trigger is further public multiple compression in engineering and industrial software before PhysicsX discloses enough revenue quality to offset it. SV016, SV019, SV022, SV025
CV049 An illustrative bear-case valuation range is about $1.2 billion to $1.8 billion if PhysicsX proves subscale or services-heavy and gets anchored toward lower public industrial-software multiples. SV015, SV016, SV021, SV022, SV024, SV025
CV050 An illustrative base-case valuation range is about $2.0 billion to $3.0 billion if recurring revenue is already above roughly $120 million but still short of clear best-in-class software proof. SV005, SV015, SV016, SV018, SV019, SV021, SV022, SV024, SV025
CV051 An illustrative bull-case valuation range is about $3.5 billion to $5.0 billion if PhysicsX reaches roughly $200 million to $250 million of recurring revenue with repeatable platform adoption and premium-software margins. SV005, SV015, SV016, SV018, SV019, SV024, SV025
来源
编号出版方标题引文
SO001 PhysicsX PhysicsX The PhysicsX platform unifies simulation, physics AI, data, and engineering applications into a single foundation.
SO002 PhysicsX PhysicsX | Careers Our London headquarters are in Shoreditch, the heart of London’s tech scene, while the New York office is nestled in Downtown Manhattan.
SO003 PhysicsX PhysicsX Announces $300M Series C to Accelerate Physics AI for Industrial Engineering PhysicsX has doubled year-over-year recognized revenue, tripled booked revenue, while more than doubling its customer count over the past year.
SO004 PhysicsX PhysicsX | About Jacomo Corbo — CEO & Co-Founder ... Robin Tuluie, PhD — Co-Founder & Chairman ... Nicolas Haag — Co-Founder & Director of Simulation Engineering.
SO005 PhysicsX Jacomo Corbo Joins PhysicsX as Co-CEO PhysicsX launched in 2020 and is headquartered in the United Kingdom, with offices in Oxfordshire and London.
SO006 Companies House PHYSICSX LIMITED overview - Find and update company information Registered office address: Victoria House, 1, Leonard Circus, London, England, EC2A 4DQ ... Incorporated on 1 August 2019.
SO007 Companies House PHYSICSX LIMITED people - Find and update company information BAUM, James Parker ... Appointed on 20 October 2023 ... CORBO, Jacomo ... Appointed on 1 January 2023 ... TULUIE, Robin, Dr. ... Appointed on 1 August 2019.
SO008 PhysicsX PhysicsX Raises $32M to Give Engineers AI Superpowers to Accelerate Design of Advanced Technologies PhysicsX, a startup bringing the power of generative AI to enable breakthrough engineering ... raised $32M in a Series A round led by General Catalyst.
SO009 PhysicsX PhysicsX Raises $135M Series B to Usher in a New Era of AI-Native Engineering and Manufacturing Since its Series A in November 2023, PhysicsX has scaled rapidly, growing to a team of over 150 and more than quadrupling revenue over the last two years.
SO010 PhysicsX PhysicsX and Siemens Collaborate to Advance Data Center Power Infrastructure with Physics AI, Powered by NVIDIA PhysicsX has partnered with Siemens to develop an AI-accelerated modeling and optimization framework built on top of Siemens’ existing multi-physics simulation workflows.
SO011 PhysicsX PhysicsX Announces Advancement to Open Standards for Physics AI, Powered by NVIDIA We’re entering the era of Large Physics Models.
SO012 PhysicsX PhysicsX Joins Forces with GB1 to Power Britain’s America’s Cup Challenge PhysicsX will deploy its AI-native engineering platform ... helping the team learn faster, explore a broader design space, and enhance performance.
SO013 PhysicsX PhysicsX and CoreWeave Partner to Deliver High-Performance Physics AI for Advanced Industrials Customers will be able to train their private, domain-specific Large Physics Models (LPMs) on proprietary data, and deploy them within secure enterprise environments.
SO014 TechCrunch PhysicsX emerges from stealth with $32M for AI to power engineering simulations Tuluie has already had two different lives as a theoretical physicist ... first at Renault and then Mercedes ... Corbo ... founded and headed up QuantumBlack, the AI labs at McKinsey.
SO015 Microsoft UK Stories How PhysicsX is transforming engineering with physics AI PhysicsX has also applied the same approach to improve thermal behaviour in Microsoft Surface devices, enabling engineers to test many more design variations of the cooling fan.
SO016 NGP NGP invests in PhysicsX London-based PhysicsX has come up with an AI platform to create and run simulations for engineers working on project areas like automotive, aerospace and materials science manufacturing.
SO017 NGP PhysicsX Announces $300M Series C to Accelerate Physics AI for Industrial Engineering - NGP PhysicsX has doubled year-over-year recognized revenue, tripled booked revenue, while more than doubling its customer count over the past year.
SO018 tech.eu PhysicsX raises $300M at $2.4BN valuation The startup has raised around $500m in total ... PhysicsX, which employs around 350 people.
SO019 The Next Web PhysicsX hits $2.4bn valuation as Temasek leads $300m round for the AI startup that cuts simulation times from days to seconds The Series C will fund expansion in the US and a new office in Singapore ... PhysicsX has grown from 150 to 350 employees over the past year.
SO020 SiliconANGLE PhysicsX reels in $300M to speed up hardware design with AI PhysicsX reels in $300M to speed up hardware design with AI.
SO021 Digital Engineering 24/7 PhysicsX Raises $300M in Series C Investments to Advance Physics AI PhysicsX has doubled year-over-year recognized revenue, tripled booked revenue, while more than doubling its customer count over the past year.
SO022 Sifted PhysicsX hits $2.4bn valuation after $300m Temasek-led funding round Founded in 2019 by former Formula 1 engineers Jacomo Corbo and Robin Tuluie ... PhysicsX has grown rapidly over the past year, increasing headcount from 150 to 350 employees.
SO023 New Market Pitch Is PhysicsX really worth $2.4B? The clean judgment is that PhysicsX’s $2.4 billion valuation is aggressive rather than irrational.
SO024 BestStartup.co.uk PhysicsX Funding 2026: Record $300M Series C Stuns UK Deeptech PhysicsX was founded in 2019 by Jacomo Corbo and Robin Tuluie, both former Formula 1 engineers.
SO025 PhysicsX PhysicsX | Platform We currently provide integrations for many industry-standard CAE tools like ANSYS, CATIA, and Siemens NX for CAD and OpenFOAM, Siemens Star CCM+ for simulation.
SO026 PhysicsX PhysicsX | Industries As chips get smaller and more complex, traditional simulation methods struggle to keep pace.
SM001 PhysicsX PhysicsX homepage
SM002 PhysicsX PhysicsX Announces $300M Series C to Accelerate Physics AI for Industrial Engineering
SM003 Siemens Digital Industries Software Siemens & PhysicsX collaborate to build AI-based deep physics
SM004 Microsoft UK Stories How PhysicsX is transforming engineering with physics AI
SM005 AWS Startups PhysicsX: Accelerating engineering innovation with generative AI
SM006 PR Newswire PhysicsX introduces free-to-use AI for advanced engineering to transform aerospace development
SM007 Securities and Exchange Commission ANSYS, Inc. Annual Report on Form 10-K
SM008 Securities and Exchange Commission Synopsys, Inc. Annual Report on Form 10-K
SM009 Securities and Exchange Commission Cadence Design Systems, Inc. Annual Report on Form 10-K
SM010 Securities and Exchange Commission Altair Engineering Inc. Annual Report on Form 10-K
SM011 Siemens AG Siemens Report 2025
SM012 Siemens / Altair Altair HyperWorks 2026
SM013 Mordor Intelligence Simulation Software Market Size, Growth Trends, Outlook 2031
SM014 Grand View Research Simulation Software Market Size | Industry Report, 2030
SM015 Business Research Insights CAE Simulation Software Market Size | Research Report, 2026 To 2035
SM016 Global Growth Insights CAE Market Size, Growth | Report [2026-2035]
SM017 Cambashi Mechanical CAE and Simulation: Review of 2025 and Opportunities in 2026
SM018 SimScale The State of Engineering AI 2026 (report PDF)
SM019 SimScale The State of Engineering AI 2026
SM020 Boston Consulting Group AI-First Companies Win the Future: Aerospace and Defense
SM021 GovInfo / Federal Register Framework for Artificial Intelligence Diffusion
SM022 International Energy Agency Data centre electricity use surged in 2025, even with tightening bottlenecks driving a scramble for solutions
SM023 SEMI Global Fab Equipment Investment Expected to Reach $110 Billion in 2025
SM024 SEMI SEMI Projects Double-Digit Growth in Global 300mm Fab Equipment Spending for 2026 and 2027
SM025 TGM AI in Engineering 2026: How Simulation, Digital Twins & Surrogate Models Are Redefining CAE
SM026 ENGtechnica Engineering Simulation Trends for 2026
SM027 Ansys Ansys | Engineering Simulation Software
SM028 Ansys High-Tech: Electronic Product Design Engineering
SP001 PhysicsX PhysicsX Announces $300M Series C to Accelerate Physics AI for Industrial Engineering The financing comes amid rapid growth. PhysicsX has doubled year-over-year recognized revenue, tripled booked revenue, while more than doubling its customer count over the past year.
SP002 PhysicsX PhysicsX | Platform Simulation Workbench: Unified system for simulation management and orchestration. AI Workbench: Environment for the development and deployment of DPMs. Engineering Applications: Workflows for engineers to seamlessly harness AI.
SP003 PhysicsX Engineering in the Age of Physics AI: The Platform Driving the Shift The platform also supports active learning — a live feedback loop that enables continuous retraining when areas of high uncertainty are encountered.
SP004 PhysicsX AI-Driven Brake Cooling Optimization: Reducing Development Time and Cost The first platform design only needed 50 CFD runs, while the second platform required just 25 runs.
SP005 PhysicsX PhysicsX Joins Forces with GB1 to Power Britain’s America’s Cup Challenge PhysicsX will work as an embedded partner within the GB1 design team in Portsmouth.
SP006 PhysicsX From Compute to Capability: PhysicsX Goes Live on the European Industrial AI Cloud PhysicsX goes live on the European Industrial AI Cloud.
SP007 PhysicsX PhysicsX Extends Collaboration with Siemens to Simcenter™ X, Enabling Seamless AI-Powered CFD Workflows PhysicsX extends collaboration with Siemens to Simcenter™ X, enabling seamless AI-powered CFD workflows.
SP008 PhysicsX PhysicsX Announces Advancement to Open Standards for Physics AI, Powered by NVIDIA Looking ahead, PhysicsX has an ambition to make these standards broadly available through open-source, ecosystem-oriented tooling.
SP009 Ansys 2026 R1: Ansys GeomAI Software and a Reimagined Ansys SimAI Portfolio This release introduces the Ansys GeomAI AI platform for geometry ... alongside a restructured Ansys SimAI platform with SimAI Pro and SimAI Premium.
SP010 Synopsys Synopsys Launches Ansys 2026 R1 to Re-Engineer Engineering with Joint Solutions and AI-Powered Products Synopsys Launches Ansys 2026 R1 to Re-Engineer Engineering with Joint Solutions and AI-Powered Products.
SP011 ANSYS Investor Relations Synopsys Completes Acquisition of Ansys Synopsys completed its acquisition of Ansys, creating the leader in engineering solutions from silicon to systems.
SP012 Siemens Simcenter simulation software Accelerate innovation with Simcenter X — flexible, cloud-powered multi-domain simulation.
SP013 Siemens Simcenter X Advanced One named-user license and floating token pool can span CFD, mechanical, systems, and MDAO workflows.
SP014 Siemens Simcenter X Tokens Tokens are floating, shareable, enabling cost-effective, scalable multi-domain simulation.
SP015 Siemens Siemens Energy uses Simcenter and artificial intelligence to improve efficiency, sustainability and cost-effectiveness With AI ... they are able to complete the equivalent of those 2,000 runs in minutes.
SP016 Siemens Altair is now part of Siemens Siemens acquires Altair to create the world’s most complete AI-powered portfolio of industrial software.
SP017 Siemens Altair Units Altair Units is a patented, units-based subscription model that provides access to more than 180 products.
SP018 Cadence Fidelity CFD Platform The Fidelity CFD Platform provides more than 10X acceleration across all steps of the simulation process.
SP019 Rescale Rescale Secures $115 Million to Accelerate Innovation with AI-Driven Digital Engineering Hundreds of enterprise customers spend more than $1 billion annually in HPC infrastructure through Rescale.
SP020 Rescale Rescale Introduces Agentic Digital Engineering Rescale introduces agentic digital engineering with simulation-native AI agents automating validation, troubleshooting, report generation, and hardware selection.
SP021 Rescale Cloud Computing Security | Rescale Rescale adheres to security protocols and compliance frameworks like FedRAMP, SOC 2, and ISO 27001.
SP022 Monolith AI Software: Engineering Product Development | Monolith See what self-learning models trained from your test data can do for you.
SP023 Monolith NAFEMS Webinar: PINNs for Manufacturability | Monolith NAFEMS webinar: PINNs for manufacturability.
SP024 nTop About Us Our technology collapses months of iteration into hours, letting teams explore thousands of variants instead of settling for the first option.
SP025 Akselos Structural Performance Management: the missing operating layer At Pearl GTL, a WEF Lighthouse site, SPM contributed to six additional years of critical asset life and a 64% reduction in annualised CAPEX.
SP026 BeyondMath BeyondMath completes $18.5 million Seed Round to scale world’s largest foundational physics AI model BeyondMath completes $18.5 million Seed Round to scale world’s largest foundational physics AI model.
SP027 BeyondMath Mastering the Aeromap: 1000x Faster Performance Discovery Mastering the Aeromap: 1000x Faster Performance Discovery.
SP028 OpenFOAM Foundation Supporting OpenFOAM Maintenance | OpenFOAM OpenFOAM is free to use, avoiding the high licence costs of proprietary CFD software (estimated at $50,000 per user per year and rising rapidly for parallel simulations).
SP029 SU2 Foundation SU2 Foundation Making multiphysics analysis and design optimization software free and publicly available.
SP030 SU2 Foundation / community GitHub - su2code/SU2: SU2: An Open-Source Suite for Multiphysics Simulation and Design SU2: An Open-Source Suite for Multiphysics Simulation and Design.
SP031 NVIDIA NVIDIA PhysicsNeMo NVIDIA PhysicsNeMo is an open-source Python framework for building, training, and fine-tuning physics AI models at scale.
SP032 NVIDIA NVIDIA PhysicsNeMo Open-source deep-learning framework for building, training, fine-tuning and inferring Physics AI models.
SP033 Cambridge Innovation Capital BeyondMath completes $18.5 million Seed Round to scale world’s largest foundational physics AI model The company plans to use the funding to accelerate commercial deployment and double team size within the next year.
SI001 PhysicsX PhysicsX
SI002 PhysicsX PhysicsX | About
SI003 PhysicsX PhysicsX | Platform Our Delivery team works in close synergy with customers to tailor model architectures, simulation pipelines, and optimization loops to their unique context.
SI004 PhysicsX PhysicsX | Careers
SI005 PhysicsX PhysicsX | Newsroom
SI006 PhysicsX PhysicsX Announces $300M Series C to Accelerate Physics AI for Industrial Engineering PhysicsX has doubled year-over-year recognized revenue, tripled booked revenue, while more than doubling its customer count over the past year. The team has grown to more than 300 people, doubling in size in the last twelve months.
SI007 PhysicsX PhysicsX Raises $135M Series B to Usher in a New Era of AI-Native Engineering and Manufacturing Since its Series A in November 2023, PhysicsX has scaled rapidly, growing to a team of over 150 and more than quadrupling revenue over the last two years.
SI008 PhysicsX PhysicsX Announces Series B Round Extension This brings PhysicsX’s total Series B funding to more than USD 155M, valuing the company at nearly USD 1B.
SI009 PhysicsX PhysicsX Teams with Deutsche Telekom and NVIDIA to Deliver Sovereign AI Infrastructure for Europe’s Advanced Industries
SI010 PhysicsX Scaling Physics AI for Automotive Aerodynamics Our Data Factory generated over 20,000 CFD simulations from over 250 baseline vehicle designs.
SI011 PhysicsX How a Fourier Neural Operator Learns to Solve PDEs — and Where It Falls Short
SI012 The Business Times Temasek-backed startup PhysicsX hits US$2.4 billion valuation to provide AI for manufacturing Corbo noted that revenue will be “close to US$50 million” for the year and the company aims to more than double the figure in 2027.
SI013 tech.eu PhysicsX raises $300M at $2.4BN valuation
SI014 The Next Web PhysicsX hits $2.4bn valuation as Temasek leads $300m round for the AI startup that cuts simulation times from days to seconds Right now, candidly, we are very supply-side limited.
SI015 BusinessCloud UK AI firm PhysicsX valued at £1.8bn by fresh funding
SI016 SiliconANGLE PhysicsX reels in $300M to speed up hardware design with AI PhysicsX claims to have tripled its booked revenues in the past year thanks to strong demand. The company’s headcount doubled to 300 employees.
SI017 Silicon Republic AI company PhysicsX raises $300m in Series C funding round for expansion
SI018 Deutsche Telekom For a sovereign Germany: Deutsche Telekom launches Industrial AI Cloud with NVIDIA Currently, more than one thousand NVIDIA DGX B200 systems and NVIDIA RTX PRO Servers – with up to 10,000 NVIDIA Blackwell GPUs – are being installed in a data center in Munich.
SI019 NVIDIA NVIDIA PhysicsNeMo
SI020 Greenhouse PhysicsX job board
SI021 Greenhouse PhysicsX EU job board
SI022 Companies House All search results - Find and update company information
SI023 Companies House PHYSICSX LIMITED filing history - Find and update company information
SI024 MarketScreener PhysicsX Ltd announced that it has received $135 million in funding from a group of investors
SI025 BusinessCloud £25m backing for engineering AI firm PhysicsX
SI026 The Next Web Want engineering superpowers? This GenAI startup is here to help
SI027 The Next Web Dr. Rob’s new AI model promises to cut aircraft design time from months to days A barebones version of the model, Ai.rplane, is also accessible free of charge.
SI028 Yahoo Finance PhysicsX raises $300 million at $2.4 billion valuation
SI029 New Market Pitch Is PhysicsX really worth $2.4B? The key unresolved question is whether PhysicsX is really a scalable software platform or a services-heavy engineering shop.
SI030 General Catalyst PhysicsX | General Catalyst Portfolio
SI031 CB Insights PhysicsX - Products, Competitors, Financials, Employees, Headquarters Locations
SI032 GuruFocus PhysicsX Raises $300 Million At $2.4 Billion Valuation
SE001 PhysicsX PhysicsX
SE002 PhysicsX PhysicsX | Platform The PhysicsX platform unifies simulation, physics AI, data, and engineering applications into a single foundation, integrating natively with the tools engineers already use.
SE003 PhysicsX PhysicsX | Industries
SE004 PhysicsX PhysicsX | Careers
SE005 PhysicsX Engineering in the Age of Physics AI: The Platform Driving the Shift It can run as a hosted service, deployed into customers’ clouds, or in a fully air-gapped environment.
SE006 PhysicsX Building Beyond Human Imagination with Foundation Models for Geometry and Physics
SE007 PhysicsX How a Fourier Neural Operator Learns to Solve PDEs — and Where It Falls Short Be cautious or avoid when: The solution contains shocks, discontinuities, or sharp boundary layers; Strict enforcement of Dirichlet, Neumann, or flux BCs is required; Your geometry is unstructured or curved.
SE008 PhysicsX Uncertainty Quantification
SE009 PhysicsX Numerical Simulation Automation: Architecting the Foundation of Intelligent Engineering
SE010 PhysicsX PhysicsX Announces ISO 27001 Certification
SE011 PhysicsX PhysicsX Extends Collaboration with Siemens to Simcenter™ X, Enabling Seamless AI-Powered CFD Workflows
SE012 PhysicsX PhysicsX Forges Strategic Collaboration with Microsoft to Accelerate Engineering Innovation with Microsoft Discovery Platform
SE013 PhysicsX PhysicsX and CoreWeave Partner to Deliver High-Performance Physics AI for Advanced Industrials
SE014 PhysicsX From Compute to Capability: PhysicsX Goes Live on the European Industrial AI Cloud
SE015 PhysicsX Introducing LGM-Aero, GenAI for Aero Engineering, & Ai.rplane Showcase Application for Aerostructures
SE016 PhysicsX Simulation Engineering at PhysicsX: The Bridge Between Physics and AI
SE017 PhysicsX On Machine Learning Methods for Physics
SE018 PhysicsX Foundational Geometry Models: How Latent Representations Transform Geometric ML
SE019 PhysicsX PhysicsX Privacy Notice
SE020 Siemens Digital Industries Software Siemens & PhysicsX collaborate to build AI-based deep physics simulation PhysicsX is building its latest pre-trained deep physics model for aerodynamics on high-fidelity simulation data generated with the Siemens Xcelerator portfolio.
SE021 Siemens Digital Industries Software Siemens introduces new Simcenter PhysicsAI add-on for AI-powered CFD design exploration
SE022 Microsoft Stories How PhysicsX is transforming engineering with physics AI
SE023 AWS Startups PhysicsX: Accelerating engineering innovation with generative AI
SE024 Engineering.com PhysicsX Launches LGM-Aero for Aerospace Engineering
SE025 Aviation Week Physics-Based AI Promises To Accelerate Aerospace Design Optimization
SE026 Digital Engineering 24/7 Free-to-Use Aerospace-Targeted AI Model Opens New Doors for Plane Optimization
SE027 PR Newswire PhysicsX introduces free-to-use 'AI for advanced engineering' to transform aerospace development
SE028 Digital Engineering 24/7 PhysicsX Unveils AI for Advanced Engineering
SE029 Trending Topics PhysicsX Taps CoreWeave's GPU Cloud to Scale Large Physics Models for Industry
SE030 Technology Magazine How Deutsche Telekom, PhysicsX & Nvidia AI Fuels Industry
SE031 CDFAM From Surrogates to Large Physics Models: Making AI-Native Engineering Work in Production
SE032 PhysicsX Ai.rplane by PhysicsX
SU001 PhysicsX PhysicsX Announces $300M Series C to Accelerate Physics AI for Industrial Engineering PhysicsX has doubled year-over-year recognized revenue, tripled booked revenue, while more than doubling its customer count over the past year.
SU002 PhysicsX PhysicsX Raises $135M Series B to Usher in a New Era of AI-Native Engineering and Manufacturing Our strategic investment in PhysicsX builds on our successful collaboration in developing AI-based deep physics simulations.
SU003 PhysicsX PhysicsX Announces Series B Round Extension
SU004 PhysicsX PhysicsX and Siemens Collaborate to Advance Data Center Power Infrastructure with Physics AI, Powered by NVIDIA With physics AI, analyses that previously required days of simulation time can now be performed in seconds.
SU005 PhysicsX PhysicsX Joins Forces with GB1 to Power Britain’s America’s Cup Challenge Partnering with PhysicsX allows our designers to build higher fidelity models, with lower data costs and quicker turn around.
SU006 PhysicsX PhysicsX and CoreWeave Partner to Deliver High-Performance Physics AI for Advanced Industrials Customers will be able to train their private, domain-specific Large Physics Models on proprietary data, and deploy them within secure enterprise environments.
SU007 PhysicsX From Compute to Capability: PhysicsX Goes Live on the European Industrial AI Cloud Integrated delivery through joint PhysicsX and T-Systems forward-deployed engineering teams, mobilized at customer sites.
SU008 PhysicsX PhysicsX Teams with Deutsche Telekom and NVIDIA to Deliver Sovereign AI Infrastructure for Europe’s Advanced Industries It connects to Teamcenter as the backbone for product data management, access control, and change processes, and interoperates with CAE environments, like Simcenter.
SU009 PhysicsX PhysicsX Selected for the Microsoft Agentic Launchpad Program Chosen from more than 500 applications, this cohort represents some of the most exciting AI innovation emerging across the UK and Ireland.
SU010 PhysicsX Scaling Physics AI for Automotive Aerodynamics State-of-the-art architectures still generalize poorly to OOD design and cross-simulator settings.
SU011 PhysicsX PhysicsX Announces Advancement to Open Standards for Physics AI, Powered by NVIDIA NVIDIA PhysicsNeMo and PhysicsX's Opora frameworks give engineers a direct path from physics-grounded AI models to real-time design and engineering.
SU012 PhysicsX PhysicsX and Deutsche Telekom Announce a Multi-Year Strategic Partnership to Accelerate AI-Native Engineering Across Europe PhysicsX and T-Systems will also mobilize forward-deployed engineering squads at customers.
SU013 The Business Times / Bloomberg Temasek-backed startup PhysicsX hits US$2.4 billion valuation to provide AI for manufacturing Customers include Applied Materials, Siemens and Stellantis.
SU014 Sifted PhysicsX hits $2.4bn valuation after $300m Temasek-led funding round Right now, candidly, we are very supply-side limited.
SU015 TNGlobal Temasek leads UK AI firm PhysicsX $300M Series C funding
SU016 Tech.eu PhysicsX raises $300M at $2.4BN valuation
SU017 Trending Topics PhysicsX From London Raises $300M, Valuation Surges to $2.4 Billion
SU018 Deutsche Telekom For a sovereign Germany: Deutsche Telekom launches Industrial AI Cloud with NVIDIA Customers like Mercedes-Benz and the BMW Group can thus conduct highly complex simulations using AI-powered digital twins.
SU019 Deutsche Telekom Germany's first AI factory for industry officially goes into operation in Munich The AI factory is already operating at over a third of its capacity with existing customers.
SU020 Siemens Siemens expands data center partner ecosystem to scale next-generation AI infrastructure Siemens is collaborating with PhysicsX to apply physics AI to the design and operation of data center power distribution systems.
SU021 Microsoft UK Stories How PhysicsX is transforming engineering with physics AI PhysicsX has also applied the same approach to improve thermal behaviour in Microsoft Surface devices.
SU022 Microsoft Microsoft at NVIDIA GTC: New solutions for Microsoft Foundry, Azure AI infrastructure and Physical AI Customers need purpose-built infrastructure for inference-heavy, reasoning-based workloads that can be deployed and operated consistently across global and regulated environments.
SU023 CoreWeave Physical AI | CoreWeave AI Cloud PhysicsX, an AI-native engineering company, is delivering production-grade physical AI on CoreWeave’s high-performance cloud.
SU024 CoreWeave CoreWeave to Acquire Monolith, Expanding AI Cloud Platform into Industrial Innovation
SU025 Enlit Siemens expands AI data centre play with partnership trio
SU026 Digital Engineering 24/7 PhysicsX Raises $300M in Series C Investments to Advance Physics AI
SU027 NGP PhysicsX Announces $300M Series C to Accelerate Physics AI for Industrial Engineering - NGP
SU028 EU-Startups London’s PhysicsX nears unicorn status with backing from NVIDIA’s VC Arm and €133 million Series B extension
SU029 Observer UK Startup Led by Former F1 Engineer Reinvents Manufacturing with A.I.
SU030 MarketScreener PhysicsX Ltd announced that it has received $135 million in funding from a group of investors
SU031 Startup Fortune PhysicsX shows industrial AI is winning frontier-style valuations Industrial customers will not buy vague AI promises when safety, cost and performance are on the line.
SU032 Technology Magazine How Deutsche Telekom, PhysicsX & Nvidia AI Fuels Industry
SU033 BizFortune Siemens Strengthens Data Center Ecosystem with AI and Energy Partnerships
SU034 Greenhouse / PhysicsX Jobs at PhysicsX
SU035 GB1 GB1 announces PhysicsX partnership as team returns to the water in Italy with new sailor signings PhysicsX will deploy its engineering AI platform to help accelerate engineering insight and support system-level optimisation across the campaign.
SU036 DEVELOP3D PhysicsX sets sail for Britain’s America’s Cup challenge
SU037 Yahoo Finance PhysicsX Raises $300 Million At $2.4 Billion Valuation The company already counts Applied Materials, Siemens AG, and Stellantis NV among its customers, while a roughly six-month customer backlog is pushing it to raise more cash and expand staff.
SR001 PhysicsX PhysicsX homepage
SR002 PhysicsX About PhysicsX
SR003 PhysicsX PhysicsX platform
SR004 PhysicsX PhysicsX industries
SR005 PhysicsX PhysicsX careers
SR006 PhysicsX Contact PhysicsX
SR007 PhysicsX PhysicsX Privacy Notice This version of our privacy policy was last updated in October 2022.
SR008 PhysicsX PhysicsX announces $300M Series C to accelerate physics AI for industrial engineering PhysicsX ... announced an oversubscribed $300 million Series C financing at a valuation of approximately $2.4 billion.
SR009 PhysicsX PhysicsX announces extension to Series B round
SR010 PhysicsX PhysicsX raises $135M Series B
SR011 PhysicsX PhysicsX announces ISO 27001 certification
SR012 PhysicsX How a Fourier Neural Operator learns to solve PDEs — and where it falls short Be cautious or avoid when: The solution contains shocks, discontinuities, or sharp boundary layers; Strict enforcement of Dirichlet, Neumann, or flux BCs is required.
SR013 PhysicsX Uncertainty quantification It is necessary to augment model estimates with credible intervals, confidence bounds, or other forms of uncertainty.
SR014 PhysicsX Why we do pilots and how to choose a good one PhysicsX partnerships with customers begin with a pilot, jointly scoped.
SR015 PhysicsX PhysicsX and CoreWeave partner to deliver high-performance physics AI for advanced industrials
SR016 PhysicsX PhysicsX and Siemens collaborate to advance data center power infrastructure with physics AI powered by NVIDIA
SR017 PhysicsX PhysicsX forges strategic collaboration with Microsoft to accelerate engineering innovation
SR018 PhysicsX Deutsche Telekom and PhysicsX announce a multi-year strategic partnership
SR019 PhysicsX PhysicsX goes live on the European Industrial AI Cloud
SR020 PhysicsX PhysicsX selected to participate in the 2024 AWS Generative AI Accelerator
SR021 PhysicsX PhysicsX strengthens leadership team with appointment of Chris Wigley as COO
SR022 PhysicsX Jacomo Corbo joins PhysicsX as co-CEO
SR023 Greenhouse PhysicsX job board
SR024 Companies House PHYSICSX LIMITED company overview
SR025 Companies House PHYSICSX LIMITED filing history
SR026 Microsoft Stories How PhysicsX is transforming engineering with physics AI
SR027 tech.eu PhysicsX raises $300m at $2.4bn valuation
SR028 TechCrunch PhysicsX emerges from stealth with $32M for AI to power engineering simulations
SR029 EU-Startups London’s PhysicsX nears unicorn status with backing from NVIDIA’s VC arm
SR030 Bureau of Industry and Security Exporting basics
SR031 Office of Foreign Assets Control Sanctions programs and country information
SR032 NIST AI Risk Management Framework
SR033 Information Commissioner's Office Artificial intelligence (AI) and data protection
SV001 PhysicsX PhysicsX
SV002 PhysicsX PhysicsX | About PhysicsX rebuilds the tooling from first principles.
SV003 PhysicsX PhysicsX | Platform The PhysicsX platform is cloud-agnostic.
SV004 PhysicsX PhysicsX | Industries
SV005 PhysicsX PhysicsX Announces $300M Series C to Accelerate Physics AI for Industrial Engineering PhysicsX, the physics AI company for industrials, today announced an oversubscribed $300 million Series C financing at a valuation of approximately $2.4 billion.
SV006 PhysicsX Engineering in the Age of Physics AI: The Platform Driving the Shift
SV007 PhysicsX PhysicsX and Siemens Collaborate to Advance Data Center Power Infrastructure with Physics AI, Powered by NVIDIA With physics AI, analyses that previously required days of simulation time can now be performed in seconds.
SV008 PhysicsX How a Fourier Neural Operator Learns to Solve PDEs — and Where It Falls Short The hard limit remains: if the target contains high-frequency structure uncorrelated with the retained modes, no amount of depth will recover it.
SV009 PhysicsX PhysicsX Announces Advancement to Open Standards for Physics AI, Powered by NVIDIA
SV010 PhysicsX PhysicsX | Careers
SV011 Companies House PHYSICSX LIMITED overview - Find and update company information
SV012 Companies House PHYSICSX LIMITED filing history - Find and update company information
SV013 Pulse 2.0 PhysicsX Raises $300 Million Series C To Advance Physics AI For Industrial Engineering
SV014 Pulse 2.0 PhysicsX: Series B Extended As New Investment From NVentures Pushes Valuation Toward $1 Billion
SV015 Macrotrends ANSYS Revenue 2010-2024 | ANSS
SV016 CompaniesMarketCap Ansys (ANSS) - Market capitalization
SV017 Ansys About ANSYS, Inc. | Company Information + Values
SV018 Macrotrends Cadence Design Systems Revenue 2011-2025 | CDNS
SV019 CompaniesMarketCap Cadence Design Systems (CDNS) - Market capitalization
SV020 Cadence Company
SV021 Macrotrends PTC Revenue 2012-2025 | PTC
SV022 CompaniesMarketCap PTC (PTC) - Market capitalization
SV023 PTC About PTC. | PTC
SV024 Macrotrends Autodesk Revenue 2012-2025 | ADSK
SV025 CompaniesMarketCap Autodesk (ADSK) - Market capitalization
SV026 Autodesk About Us | Autodesk Company Info, Mission, and Values
SV027 Macrotrends ANSYS Market Cap 2010-2025 | ANSS
SV028 Macrotrends Cadence Design Systems Market Cap 2011-2025 | CDNS
SV029 CompaniesMarketCap Synopsys (SNPS) - Market capitalization
SV030 Synopsys About Us | Synopsys