快科技10月13日消息，据媒体报道，DeepSeek创始人梁文锋近期密集关注具身智能领域，并与多位该领域创业者交流，探讨机器人、大模型与现实世界交互等方向。知情人士透露，梁文锋正在考虑DeepSeek进入具身智能领域的可能性，可能会亲自组建团队下场研发，也可能通过投资相关企业切入。
Kuai Technology reported on October 13 that, according to media reports, DeepSeek founder Liang Wenfeng has recently been密集 focusing on the embodied intelligence field and exchanging with multiple entrepreneurs in the field, discussing directions including robots, large models, and interaction with the real world. People familiar with the matter revealed that Liang Wenfeng is considering the possibility of DeepSeek entering the embodied intelligence field, possibly personally forming a team to develop in-house, or entering through investing in related companies.
此前，梁文锋已投资宇树科技。今年8月，DeepSeek以1.41亿元战略投资入股宇树科技，双方签署《战略合作备忘录》，宣布在通用人工智能（AGI）、高性能通用机器人和AI大模型三大方向开展深度合作。根据备忘录，双方的合作涵盖三个层面：一是面向通用人工智能开展联合研发；二是在高性能通用机器人方面深度合作；三是在AI大模型方面开展深度合作，DeepSeek将为宇树科技提供具有市场竞争力的商业合作方案。
Previously, Liang Wenfeng had already invested in Unitree. In August this year, DeepSeek made a strategic investment of 141 million yuan in Unitree, and the two signed a "Strategic Cooperation Memorandum," announcing deep cooperation in three major directions: artificial general intelligence (AGI), high-performance general-purpose robots, and AI large models. According to the memorandum, cooperation covers three levels: first, joint R&D for artificial general intelligence; second, deep cooperation in high-performance general-purpose robots; third, deep cooperation in AI large models, with DeepSeek providing Unitree with commercially competitive cooperation solutions.
另据了解，DeepSeek计划领投腾讯前AI科学家胡瀚创立的机器人公司Winverse AI，美团旗下龙珠投资等机构也已作出投资承诺，该公司投后估值约为4亿美元。不过，据多位行业人士指出，Winverse AI不能算严格意义上的具身智能公司，更偏底层infra，做模型加速相关。对于DeepSeek布局具身智能一事，截至本文发稿DeepSeek方面尚未回应。
Separately, DeepSeek plans to lead the investment in Winverse AI, a robotics company founded by former Tencent AI scientist Hu Han, with Meituan's Long-Z Investments and other institutions also making investment commitments. The company's post-money valuation is about $400 million. However, multiple industry figures point out that Winverse AI cannot be considered a strictly embodied intelligence company—it leans more toward underlying infra, doing model acceleration-related work. As of this article's publication, DeepSeek has not responded to its embodied intelligence布局.
梁文锋接触具身智能这个动作，放在DeepSeek过去两年的路径里看，有其内在逻辑。DeepSeek做大模型，走的是开源加极致性价比的路线，在语言模型和推理能力上建立了自己的位置。但大模型要往通用人工智能走，光有语言和推理不够，还需要与现实世界交互的能力。具身智能是这条路上绕不开的一环。梁文锋考虑亲自组建团队下场，而不是只做财务投资，说明他对这件事的判断不只是财务层面的机会，而是战略层面的必要布局。
Liang Wenfeng's move into embodied intelligence, placed in DeepSeek's path over the past two years, has its internal logic. DeepSeek does large models, taking the route of open source plus extreme cost-effectiveness, establishing its position in language models and reasoning capability. But for large models to move toward artificial general intelligence, language and reasoning alone are not enough—capability to interact with the real world is also needed. Embodied intelligence is an unavoidable link on this path. Liang Wenfeng considering personally forming a team to enter, rather than only doing financial investment, indicates his judgment on this matter is not just a financial-level opportunity but a strategic-level necessary布局.
从投资动作看，DeepSeek已经在往这个方向铺路。8月以1.41亿元战略投资宇树科技，签署的备忘录覆盖AGI、通用机器人和AI大模型三个层面，这个合作框架的野心不小：宇树提供机器人的身体和运动能力，DeepSeek提供大模型和推理能力，双方在AGI方向上联合研发。这个组合如果能跑通，会是一个从大脑到身体的完整链条。现在DeepSeek又计划领投Winverse AI，虽然这家公司更偏底层infra、做模型加速，但它对应的是具身智能在算力和推理效率上的需求。模型加速是具身智能落地的关键环节，因为机器人本体的算力有限，模型必须在边缘端高效运行。
From the investment actions, DeepSeek has been paving the way in this direction. In August, it strategically invested 141 million yuan in Unitree, with the memorandum covering three levels—AGI, general-purpose robots, and AI large models. This cooperation framework has considerable ambition: Unitree provides the robot's body and motion capabilities, DeepSeek provides large models and reasoning capabilities, and both conduct joint R&D in the AGI direction. If this combination works, it will be a complete chain from brain to body. Now DeepSeek plans to lead investment in Winverse AI. Although this company leans more toward underlying infra and model acceleration, it corresponds to embodied intelligence's needs in compute and reasoning efficiency. Model acceleration is a key link in embodied intelligence deployment, because robot bodies have limited onboard compute, and models must run efficiently at the edge.
把这件事放在前面讨论的产业脉络里看，DeepSeek入局具身智能，是模型公司向物理世界延伸的又一例。前面讨论过鹿明机器人和面壁智能的合作，模型公司提供模型，本体公司提供身体和场景。也讨论过自变量机器人自己做全栈，从模型到数据采集方案全覆盖。DeepSeek如果亲自组建团队下场，走的是第三条路：一家已经在大模型领域建立起技术优势和成本优势的公司，把这套能力迁移到具身智能上。梁文锋投资宇树、接触具身智能创业者、计划领投Winverse AI，这三个动作连起来，说明他不是在看单点机会，而是在搭一个体系。
Placing this matter in the industrial thread discussed earlier, DeepSeek entering embodied intelligence is another example of a model company extending toward the physical world. Earlier we discussed the partnership between Luming Robotics and ModelBest—model companies provide models, body companies provide bodies and scenarios. We also discussed AgiBot doing full-stack itself, covering everything from models to data collection solutions. If DeepSeek personally forms a team to enter, it takes a third path: a company that has already established technical and cost advantages in large models migrating this capability to embodied intelligence. Liang Wenfeng investing in Unitree, contacting embodied intelligence entrepreneurs, and planning to lead investment in Winverse AI—these three actions connected indicate he is not looking at single-point opportunities but building a system.
