AI News AI资讯 13h ago Updated 2h ago 更新于 2小时前 52

Tongji PhD Uses Geometry and Physics AI to Redesign Design and Manufacturing, Cumulative Financing Exceeds 300 Million Yuan 36氪首发 | 同济博士做几何、物理AI重构设计制造,累计获超3亿元融资

SheXu Technology completed a B-round financing exceeding 100 million RMB, bringing total funding to over 300 million RMB, driven by its "Zexing AI" platform for industrial design and manufacturing. The company has upgraded its product line into specialized Agents: a Geometric AI 3D Agent for shape generation, a Manufacturing AI 2D Agent for engineering drawings, and an emerging Physical AI for precise physics simulations. Revenue is growing rapidly with a 70% year-over-year increase in H1 contra 设序科技完成B轮超亿元融资,累计获投超3亿元,资金将用于市场开拓(含出海)及核心模型技术研发。 产品“则形AI”升级为基于自研工业世界模型与LLM的智能体平台,实现从几何设计、物理求解到制造判断的闭环。 技术架构分为三步:几何AI生成3D形状、物理AI求解性能参数、制造AI评估可制造性,目前物理AI处于POC验证阶段。 公司上半年合同额同比增长70%,预计全年营收近2亿元,RAAS(Result as a Service)模式贡献约1/3营收。 创始人预测工业世界模型的“GPT时刻”将在3-5年内到来,判断标准为达到L4/L5级智能水平。

72
Hot 热度
78
Quality 质量
75
Impact 影响力

Analysis 深度分析

TL;DR

  • SheXu Technology completed a B-round financing exceeding 100 million RMB, bringing total funding to over 300 million RMB, driven by its "Zexing AI" platform for industrial design and manufacturing.
  • The company has upgraded its product line into specialized Agents: a Geometric AI 3D Agent for shape generation, a Manufacturing AI 2D Agent for engineering drawings, and an emerging Physical AI for precise physics simulations.
  • Revenue is growing rapidly with a 70% year-over-year increase in H1 contracts, aiming for nearly 200 million RMB annually, with the Result as a Service (RAAS) model contributing one-third of revenue.
  • The strategic vision involves a three-step AI workflow where users input requirements, and the system sequentially handles geometry, physical performance solving, and manufacturability assessment.

Why It Matters

This development highlights the maturation of Industrial AI from simple generative tools to comprehensive, agent-based workflows that integrate geometric design, physical simulation, and manufacturing constraints. For practitioners, it demonstrates a viable path toward automating complex engineering tasks while preserving human expertise for high-level decision-making, signaling a shift in how hardware R&D pipelines are structured.

Technical Details

  • Architecture: The core product, "Zexing AI," utilizes a self-developed Industrial World Model combined with Natural Language Large Models to create a design and R&D agent platform operating on a cloud architecture.
  • Agent Specialization: The platform features distinct agents: a 3D Agent based on Geometric AI for precise shape design, a 2D Agent based on Manufacturing AI for generating engineering drawings, and a Physical AI currently in POC stages for solving physical fields like strength, stiffness, fluid dynamics, and magnetic fields.
  • Workflow Logic: The system executes a sequential pipeline: Geometric AI designs the shape -> Physical AI calculates performance metrics -> Manufacturing AI assesses producibility, creating a closed loop for hardware engineering.
  • Business Model: Implementation includes a Result as a Service (RAAS) delivery mode, which accounts for approximately 33% of annual revenue, focusing on delivering final engineered outcomes rather than just software access.

Industry Insight

  • Human-AI Collaboration Paradigm: The industry is moving toward a model where AI handles 80% of repetitive labor, allowing engineers to focus on the remaining 20% of creative judgment and exception handling. Professionals must adapt by enhancing their ability to make professional judgments rather than relying solely on software operation skills.
  • Data Flywheel Importance: The timeline for achieving L4/L5 level intelligence in industrial models depends heavily on data accumulation through mass production. Companies should prioritize deploying solutions that generate feedback loops to accelerate model improvement.
  • Global Expansion Strategy: With domestic markets saturated or competitive, early movers like SheXu are targeting European markets first due to regulatory compliance readiness, suggesting that global expansion in industrial AI requires navigating strict data sovereignty and compliance frameworks before scaling.

TL;DR

  • 设序科技完成B轮超亿元融资,累计获投超3亿元,资金将用于市场开拓(含出海)及核心模型技术研发。
  • 产品“则形AI”升级为基于自研工业世界模型与LLM的智能体平台,实现从几何设计、物理求解到制造判断的闭环。
  • 技术架构分为三步:几何AI生成3D形状、物理AI求解性能参数、制造AI评估可制造性,目前物理AI处于POC验证阶段。
  • 公司上半年合同额同比增长70%,预计全年营收近2亿元,RAAS(Result as a Service)模式贡献约1/3营收。
  • 创始人预测工业世界模型的“GPT时刻”将在3-5年内到来,判断标准为达到L4/L5级智能水平。

为什么值得看

本文揭示了工业AI从单一工具向全流程智能体演进的典型路径,展示了“几何+物理+制造”三位一体的技术闭环如何重构硬件工程研发范式。对于关注AI落地硬科技领域的从业者而言,其RAAS商业模式及出海战略提供了工业软件SaaS化转型的重要参考。

技术解析

  • 核心架构:基于自研“工业世界模型”封装自然语言大模型,构建面向硬件工程的通用AI入口。系统通过Agent化升级,将传统3D/2D生成能力转化为具备自主执行能力的智能体。
  • 三步执行逻辑
    1. 几何AI:根据用户需求生成精确的3D几何形状。
    2. 物理AI:对生成的几何形状进行工业级精确求解,计算强度、刚度、流场、磁场等物理性能。
    3. 制造AI:评估设计方案的可制造性,连接设计与生产环节。
  • 交付模式创新:采用RAAS(Result as a Service)模式,即按结果而非软件授权收费,该模式已成为重要增长引擎,占全年营收约三分之一。
  • 数据飞轮策略:通过L2/L3级智能的大规模量产应用获取数据,反哺模型迭代,推动技术向L4/L5级高水平智能演进。

行业启示

  • 工程师角色重塑:AI不会完全取代工程师,但会取代“不会用AI的工程师”。未来核心竞争力将从软件操作转向基于经验的创造性判断、公差设定及成本性能平衡等专业决策。
  • 工业AI接受度拐点已至:客户态度从保守观望转向激进投入POC,驱动力来自基础大模型进步、实际痛点解决及已落地场景的价值验证,表明工业AI正进入规模化应用前夜。
  • 出海成为新增长极:随着国内市场竞争加剧,具备合规能力(如欧盟数据合规)和差异化技术优势的工业AI企业加速布局欧洲、日韩及东南亚市场,出海将成为头部企业的重要战略方向。

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Agent Agent Funding 融资 Product Launch 产品发布