Shenzhen Embodied AI Company Astribot Completes Over 1 Billion B-Round Financing, Valuation Surpasses 10 Billion | Hard氪 Exclusive 深圳具身公司星尘智能完成超10亿B轮融资，估值破百亿｜硬氪首发

Star dust intelligence secured a billion-yuan funding across three rounds within three months, with its valuation soaring to tens of billions. Another "embodied intelligence unicorn" has been born in Shenzhen. This pace has left almost the entire competitive field far behind.

Yet the flood of capital恰恰 reveals the underlying anxiety. Players in the embodied intelligence sector have collectively realized that merely demonstrating a few impressive movements on stage can no longer deceive investors into the next funding round. The industry is forcefully transitioning from an "exhibition match" to an "obstacle course." What Star dust has secured isn't just money but an "entry ticket"—a pass to the brutal testing ground of the real physical world. Investors' interest in it likely stems not solely from founder Lai Jie's Tencent and Baidu background, but from its courage to deploy robots into the greasy, noisy, and unpredictable environments of factory workshops and hotel kitchens.

Star dust's most aggressive move has been cornering itself into "full-stack self-developed" technology. From its rope-driven actuation body designed "for AI," to its proprietary foundation model Lumo, and the DuoCore framework mimicking the human dual-system. This approach is both "geeky" and "risky." It's betting on a future where robots are not passive mechanical arms, but "quasi-living entities" with autonomous thinking and action capabilities. The rope-driven solution is key—it makes robots feel more "organic," with smoother movements and higher-quality data, which is crucial for AI learning. However, the difficulty and cost of mass-producing rope-driven robots had previously deterred all manufacturers. Star dust's bold claim of delivering thousands of units suggests either it possesses true technical prowess or it's gambling everything it has.

Its proposed "fast-slow dual-system" is intriguing. The fast system handles instinctive reactions, while the slow system manages deep thinking. This sounds plausible, but the actual coordination—whether it achieves "mind-body unity" or falls into "dissociation"—differs vastly. Star dust claims its architecture is nearly identical to that of American company Figure's Helix, which is indeed an tacitly acknowledged fact in the industry: top-tier ideas are converging globally. The difference lies in who can achieve faster and more stable engineering implementation. Star dust's DuoCore framework reportedly has already achieved large-scale deployment in retail scenarios, currently its most compelling talking point.

The most noteworthy aspect is the 89,900-yuan price tag. This is akin to dropping a depth charge into the industry's pool. Previously, research-grade robots costing hundreds of thousands or even millions of yuan severely limited application scenarios. Star dust aims to use a price point close to that of a "luxury electric vehicle" to crack open the commercial service market. Tasks like searing steaks, mixing cocktails, and sorting auto parts aren't inherently difficult in isolation, but requiring a robot to reliably complete them in unfamiliar environments and over long-sequence tasks is a challenge of an entirely different magnitude. The thousand-unit orders may sound daunting, but those from Thundercomm are for industrial and commercial service applications, and the one for the Jiangdu is for a billion-yuan innovation center—both serving major B2B and B2G clients. The true consumer market is still far from realization.

Therefore, the tens-of-billions valuation more closely resembles a heavy "performance bet." Capital is voting with its money, endorsing the potential of "Design for AI" and rope-driven technology, but the market only recognizes results. Over the next 12-18 months, Star dust must prove to everyone that its robots can not only "perform" but also be "used" reliably and cost-effectively. Otherwise, this frenzy of financing may merely become another poignant footnote in the next technological cycle. The marathon of embodied intelligence has just begun, and the leader in the first hundred meters may not necessarily cross the finish line with a smile.

星尘智能三个月内三轮融了十个亿，估值冲上百亿。深圳又一家“具身智能独角兽”诞生。这个速度，几乎把整个赛道甩在了身后。

但钱涌进来，恰恰说明了焦虑。具身智能的玩家们终于集体意识到，光在舞台上演示几个漂亮动作，已经骗不到下一轮融资了。行业正从“表演赛”强行切换到“障碍赛”。星尘拿到的不只是钱，更是一张“入场券”——一张通往真实物理世界残酷测试场的入场券。投资人看中它，恐怕不是因为来杰的腾讯、百度履历，而是它终于敢把机器人，塞进那些油腻、嘈杂、充满意外的工厂车间和酒店后厨里。

星尘最狠的一招，是把自己逼到了“全栈自研”的墙角。从“设计为AI服务”的绳驱本体，到自研的基座模型Lumo，再到那个模仿人类双系统的DuoCore框架。这套路很“极客”，也很“危险”。它赌的是一个未来：机器人不是一个被动的机械臂，而是一个有自主思考和行动能力的“准生命体”。绳驱方案是关键，它让机器人摸起来更“肉”，动作更柔顺，数据质量也更高——这对AI学习至关重要。但量产绳驱机器人的难度和成本，此前让所有厂商望而却步。星尘敢喊出千台级交付，要么是真有金刚钻，要么是赌上了全部身家。

它提出的“快慢双系统”很有意思。快系统负责本能反应，慢系统负责深度思考。这听起来像那么回事，但实际协同起来，是“灵魂出窍”还是“身心合一”，差距是天壤之别。星尘声称其架构与美国公司Figure的Helix几乎一致，这倒是行业里心照不宣的事实：顶尖思路正在全球收敛。区别在于，谁的工程化落地更快、更稳。星尘的DuoCore框架宣称已在零售场景规模化落地，这是它目前最有分量的谈资。

最值得关注的是那个8.99万的定价。这简直是在行业水池里扔了一颗深水炸弹。此前动辄数十万甚至上百万的科研级机器人，让应用场景极度受限。星尘试图用一个接近“豪华电动车”的价格，撕开商业服务市场。煎牛排、调酒、分拣汽配……这些任务单拎出来都不难，但要求机器人在陌生环境、长序列任务中稳定可靠地完成，就是另一个维度的挑战。千台级订单听着吓人，但中科创达那笔是工业及商业服务订单，江都那个是亿元级创新中心，都是To B、To G的大客户。真正的消费级市场，还远未到来。

所以，百亿元估值，更像是一张沉重的“对赌协议”。资本用钱投票，认可“Design for AI”和绳驱路线的潜力，但市场只认结果。未来12-18个月，星尘必须向所有人证明：它的机器人不只是能“做”，而是能可靠地、低成本地“用”。否则，这轮狂热的融资，很可能只是下一个技术周期里，又一个令人唏嘘的注脚。具身智能的马拉松才刚开始，百米领先的选手，未必能笑到最后。