Podcast Atlas
concept Updated 2026-08-24 Topics: Technology

Embodied AI Value Chain

宇树上市暴涨，但人形机器人的钱到底从哪里赚？｜S10E26 adds a capital-market and platform-value layer through Unitree Robotics. The episode says cheaper Unitree robot platforms can help universities and developers experiment, but Unitree IPO Valuation / 宇树上市估值 still depends on whether body hardware, robot brains, data/simulation, and scenario owners turn into durable shareholder value rather than only ecosystem value.

Embodied AI value chain is the source’s frame for why robot companies cannot be judged only by algorithm papers or demo videos. In 132. 对星海图创始人高继扬的3小时访谈：鲶鱼、曾国藩、Waymo与Momenta的两面、一只狼与许华哲的离开, Gao Jiyang says the value chain includes whole machines, supply chain, data, AI infrastructure, algorithms, models, distribution, and customer value.

134. 【数据的综述】和谢晨聊，新时代的石油、历史、版图、数据金字塔、定价与Recipe adds 谢晨’s ecosystem split. He expects robot-body companies such as Unitree Robotics and Zhiyuan Robotics, model-brain companies, data/simulation companies such as 光轮智能, and scenario owners to cooperate rather than one company cleanly owning the whole stack.

143. 对何小鹏的第二次访谈：更大赌注、人形机器人Iron诞生、那场意外、技术剧变下CEO、GX和缝合怪 adds XPeng / 小鹏汽车’s integrated hard-tech version. The source puts cars, XPeng Iron, XPeng GX, autonomous driving, robot hardware, data governance, compute, manufacturing, and AI Organization Design into one Physical AI value-chain argument.

170: 【具身季报 26Q2】世界模型大风不停，和不想被贴标签的人 adds a cross-layer competition view. Chen Zhe Peter asks whether durable value will sit with robot-body companies, model companies, full-stack firms, dexterous-hand suppliers, data companies, or general foundation-model labs. The episode’s examples include Honor in hardware stress tests, Figure AI and Xingdong Era in logistics sorting, 5G Robotics and Genesis Robotics in hands and manipulation, and Cosmos 3, Physical Intelligence, Generalist, OpenAI, and Google DeepMind in robot brains.

166: 许华哲再次具身创业：不想错过最大的西瓜 adds Xu Huazhe’s warning that the value chain can still aim at the wrong prize if it optimizes around shipments, production landing, or data sale too early. In his Poke Robotics route, the value-chain question is whether the company controls enough body, model, data, product, and user feedback to define Physical AGI rather than becoming a hardware supplier or scene integrator.

E244｜端到端vs上下分层：机器人路径之争，正在转向？ adds 速度科技 / Sudu Technology’s platform-route version. Han Zheng / 韩正 argues that a robotics company may need to own hardware, low-level manipulation models, Sim2Real, developer APIs, and tools, while letting outside developers assemble many vertical long-horizon applications.

Momenta IPO后再访曹旭东：就是想做没有尽头的AI adds Momenta’s autonomous-driving-to-robotics version. Cao Xudong says Momenta will start from target scenes and the robot brain, design the body when the product requires it, and avoid duplicating value-chain layers that strong partners already cover. The source also shows a platform-partner split in Robotaxi: Momenta supplies ASG capability while players such as Didi, AutoNavi / Gaode, and T3 Chuxing may own more of the user and fleet layer.

173: 对话姚颂：深鉴、东方空间、再出发，「天才少年」十年后 adds Striding AI / 正行创新’s more vertically integrated version through Yao Song / 姚颂. Yao argues that physical intelligence must include data, compute, model, software, hardware, solution, scenario, and remote-system layers together. In his view, early companies may need stronger full-stack ownership because standard robot interfaces, suppliers, and deployment recipes have not yet stabilized.

Key Claims

  • Algorithm innovation is important, but it becomes commercially meaningful only when connected to the rest of the robot product and deployment system.
  • Robotics companies face longer chains and cycles than many model or AI-application companies because they also handle hardware, supply chain, offline customers, and maintenance-like realities.
  • A company can be technologically ambitious and still be forced into pragmatic ROI thinking every day.
  • The value-chain view explains why Xu Huazhe’s departure does not, by itself, imply that Xinghaitu is abandoning algorithm innovation.
  • Data and simulation can become a separate value-chain layer because Embodied Data Pyramid, Robotics Simulation Evaluation, and Data Recipe Co-Creation require specialized infrastructure.
  • An automotive company may try to internalize more of the value chain when vehicle safety, humanoid control, data, and manufacturing are treated as one strategic system.
  • Dexterous hands can become their own value-chain layer because hardware standards, degrees of freedom, sensing, retargeting, and supply reliability shape both data and model progress.
  • Full-stack manipulation platforms can occupy a different value-chain position from vertical scene integrators: they provide stable short skills, hardware, APIs, and development tools rather than personally building every deployment scenario.
  • Autonomous-driving companies can enter the embodied-AI value chain through model/data advantages first, then selectively add body design, manufacturing partnerships, or platform cooperation where needed.
  • A physical-intelligence startup may need temporary vertical integration when the ecosystem cannot yet supply reliable standard modules, interfaces, or data loops.
  • Lowering hardware platform cost can expand the embodied-AI ecosystem, but the value chain still has to show where repeat demand, profit, and public-market value are captured.

Connections