上纬新材,和智元机器人一起刷了把“存在感”Sinoma Technology, together with Zhipu AI, has brushed up on its "presence.


AI free token/AI免费积分

智元机器人正在猛刷存在感。

6月28日,央视世界杯直播节目里出现了一台启元Q1小型人形机器人,坐在解说嘉宾席上天真地聊足球。这是国内消费级具身智能产品第一次以赛事参与者的身份,和一场全民体育盛会绑在一起。

上纬新材,和智元机器人一起刷了把“存在感”

几乎同一时间,智元机器人宣布第15000台通用具身机器人精灵G2正式量产下线,距离上一次突破万台大关不到三个月。

加速,是机器人的主旋律。一边是走向产线的工业级机器人,正在用令人目眩的量产速度把行业带出“Demo时代”;另一边是走向客厅的个人机器人,为万千观众带来了独特的体验。

相较于在产品线上做加法,这里的动作更多突出了在产业纵深上做矩阵。工业端用实打实的产能和部署向市场证明人形机器人是一门算得过账的生意。,家庭端则用轻巧的姿态去触碰那个尚未被充分定义的“人机共存”市场。

显然,智元的野心不小。而这一次,上市公司上纬新材,也成了其叙事中重要的一部分。毕竟,启元Q1就是在2025年末由上纬新材旗下品牌上纬启元发布的。

一、量产即壁垒,数量代表能力的沉淀

人形机器人行业一直有个真实的规律:Demo强不代表交付强,交付能力才是商业化的分水岭。

智元的变化速度提供了很好的参考。2025年12月,其通用具身机器人累计下线5000台。2026年3月,这个数字变成10000台。到6月,仅三个月后就到了15000台。

上纬新材,和智元机器人一起刷了把“存在感”

从一年多时间摸到千台级,到三个月完成5000台增量,节奏的压缩意味着关键瓶颈很可能出现了突破。就行业现状推理,它也能反映出行业供给的一些变化。近期,资本市场也确实加强了对机器人产业链的关注。

首先是供应链。人形机器人的零部件复杂程度远超传统工业机器人,从谐波减速器、力矩传感器到灵巧手关节和嵌入式算力芯片,每一处都曾是制约产量的卡点。能把产能快速拉升到新的水平,除了证明智元实力的强悍,也说明零部件国产化和标准化程度已经抵达了可支撑规模制造的阶段。

智元曾披露,公司已建立了全球首个具身智能机器人标准化供应链体系,覆盖上下游全环节。这是被15000台数字验证过的产业能力。供应链成熟让机器人加速走向可复制的工业制造,更重要的是把单机成本拉低到真正能进入商业谈判的价格区间。

第二个突破在部署端。把一台机器人卖进工厂,和把它真正嵌入产线、持续产生价值,是完全不同的两件事。调度得好不好,适配得快不快,异常情况能不能快速排查,决定了客户是否愿意下第二单、第三单。

部署经验的积累,本质上是一个组织学习的过程。做过100个场景的团队和做过5个的团队,在面对新场景时的应变速度和质量完全不在一个层面上。15000台的累计交付量,对应的是一支已经积累了大规模现场部署能力的工程队伍,它是隐形的壁垒。

第三个突破更为根本,那就是数据飞轮开始转动了。每一台在真实场景中运行的机器人,都在产生运动控制、任务执行和环境交互数据。对于具身智能而言,数据不是锦上添花,而是当前阶段最紧缺的燃料。

大模型在语言文字上的突破,依赖于互联网积累了数十年的公开语料。但机器人在物理世界中的操作数据,根本不存在现成库存,必须从零开始积累。谁先完成大规模部署,谁就率先打开了数据采集的阀门。

今年2月,智元孵化了觅蜂科技,目标是构建百亿小时级的物理世界交互数据产能,这背后的逻辑就是:用工业量产摊开数据入口,用数据反哺模型进化,再用更强的模型巩固量产部署的领先优势。

上纬新材,和智元机器人一起刷了把“存在感”

这正是智元双轨战略的第一道支撑。工业端用速度和规模跑通商业闭环、锤炼供应链、沉积数据,把具身智能从实验室命题变成可量化、可复制的生意。这层基础一旦建立,它就为更远的布局腾出了空间。

二、机器人频繁亮相,智元醉翁之意不在酒

由上纬启元打造的小型人形机器人去当世界杯解说嘉宾,意义和生产环境使用的机器人做展示是不同的。

启元Q1没有强调工业性能,没有展示搬箱子或者拧螺丝,而是出现在一个需要表达、互动和性格的语境里。它的功课不是刚体力矩控制,而是如何在镜头前自然接话、给出有模有样的点评、被观众接受为一个有趣的“角色”。

这恰恰是消费级机器人要解决的核心问题——用户为什么会想要一台机器人待在家里?不是因为需要生产力,而是因为它能在某种维度上与人的生活产生有温度的连接。

上纬新材,和智元机器人一起刷了把“存在感”

从产品设计来看,启元Q1选择的是一条很务实的小型化路线:全身力控,可放进背包,面向科研教育、极客创作和家庭陪伴。

小型化不只是为了压缩成本,更重要的是调整了用户对“它能干什么”的期望值门槛。一个与真人等高的机器人站在客厅里,在技能上、视觉上,以及用户体感上,会有很多天然的不完美;而一个偏小巧的机器伙伴,用户更容易主动去探索它的能力边界。

不过,这也使得个人机器人真正的生命力必然不在硬件出厂的那一瞬间,而在于之后有没有人持续为它生产内容、开发应用、创造新玩法。

工厂场景里,作业任务是明确定义的;家庭场景里,用户自己都不一定说得清想要什么。解决这个难题的方式,是让产品成为一个开放平台。开发者可以给它编写新动作、设计新功能;用户可以把自己的创作分享在社交网络上;每一个新奇用法的诞生,都是一次免费的用户教育。

这恰恰是智元前面几年在开发者工具上持续投入所积攒的家底——AIMA开放平台、Genie Studio开发环境、AGIBOT World数据集、Motion-Between运动基座模型,这些资产最初服务于科研用户和工业合作伙伴,但一旦它们或者类似的开发配套与消费级产品结合,就可以降维成更轻量化的创作工具,让普通玩家也能参与进来。

启元Q1是对整个未来消费生态的一种试探,它牵引着开发者和普通家庭,在一个具体场景里做出了示范。当然,也让更多人再次关注到上纬新材这家已经由稚晖君担任董事长的公司——这家完全蜕变为智元机器人的“物理之壳”的公司。

三、双轨的背后:从造机器人到建“具身宇宙”

机器人产业在今年出现了明显的分化趋势,重点是产业链的分化。一批专门聚焦数据、深海或空天等特定场景、灵巧手等部件、感知等专项技术研发的“精品型”公司在崛起,此外也有一些其他产业链的公司因为材料等上游优势汇聚了新的焦点。

与之对应的是,智元的动作也是打造一张更加系统化的产业版图。目前,智元已经拥有六家独立运营的子公司,各守一块阵地:

做清洁产品的智鼎、做租赁平台的擎天租、做灵巧手的临界点、做数据采集的觅蜂、做四足机器人的酷拓,以及承载消费级产品和资本平台功能的上纬启元。每一个子公司都有独立的融资能力和运营团队,但从场景定位到技术底座,它们之间形成的是互补而非竞争的关系。

这种结构的好处在于,把需要不同组织基因的业务物理隔离——做租赁和做数据的团队不可能用同一套考核方式——同时又确保底层技术可以共享复用。

尤其值得留意的是,上纬新材在其中的角色不止于资本通道。作为有名的“前妖股”,上纬新材原主营是高性能复合材料和风电叶片用树脂,它的轻量化、高强度材料技术已经应用在启元Q1的结构件研发验证中。

上纬新材,和智元机器人一起刷了把“存在感”

消费级机器人未来面临的成本控制、重量优化、安全合规等挑战,最终都会传导到材料层面。拥有一个闭环于内的材料平台,意味着在产品定义的早期就可以从材料侧反向优化设计,这在产业协同上是非常直接的优势。

更宏观地说,智元的整套布局揭示了一个正在发生的产业趋势:人形机器人正在从做出样机的竞赛,转向做出规模并持续运营的竞赛。

在做出样机的阶段,一个聪明的算法团队加一个精巧的硬件团队就能脱颖而出。但进入规模运营阶段后,供应链管理、现场部署效率、运维体系、数据回流、开发者社区、融资平台,这些看似远离“炫技”范畴的能力,才是真正的胜负手。这个时候,资本很重要,但没有那么重要。比如智元的崛起更多是资本+团队能力的组合。如果只有“钱”,恐怕是无法畅通无阻的。

智元的“358宏图”,也就是三年十亿、五年百亿、八年千亿,实际展示了一个判断:具身智能的爆发不会像移动互联网那样单纯依赖软件迭代,而需要先靠工业场景把硬件、供应链和数据飞轮推到一个临界质量,再让消费市场接过品牌与生态的接力棒。

这是一个典型的“重工业养能力、轻消费养品牌”的双轮模型,两者不是替代关系,而是共生关系。哪个轮子先转起来并不重要,关键是两个轮子要同轴。因此,它有了一套“本体+数据+模型+平台”的全栈能力。

后续最大的看点,一方面是客观产业数据的提升,另一方面是规模大的一侧的部署数据和模型经验,要能反哺规模偏小的一侧,实现开发成本降低、体验提升。同时,积累的用户交互数据也要为人机协作和柔性作业提供反馈。只有这样,丰富的布局才不会掉进“贪多嚼不烂”的陷阱,而是互为犄角、巧夺天工。

 

 

Zhimu Robotics is making a strong presence.

On June 28, a Qiyuan Q1 small humanoid robot appeared in CCTV's World Cup live broadcast, sitting in the commentary seat and chatting about football with innocent charm. This marks the first time a domestic consumer-grade embodied intelligence product has been tied to a major sports event as a participant.

Sinoma Technology, together with Zhipu AI, has brushed up on its "presence.

Almost simultaneously, Zhimu Robotics announced the 15,000th unit of its universal embodied robot, the Elf G2, has officially rolled off the production line — marking a significant milestone reached less than three months since the last 10,000-unit breakthrough.

Acceleration is the main theme for robots. On one hand, industrial-grade robots are speeding toward production lines with dazzling output rates, pulling the industry out of the “demo era”; on the other, personal robots are entering living rooms with unique experiences for audiences.

Compared to adding more products, the focus here is on building a matrix across industry depths. The industrial side proves that humanoid robots are a profitable business with solid production and deployment data. On the family side, it’s about gently tapping into the yet-to-be-defined “human-machine coexistence” market.

Clearly, Zhimu’s ambition is not small. And this time, the listed company Shangwei New Materials is also part of its narrative — after all, the Qiyuan Q1 was launched by Shangwei Qiyuan, a subsidiary of Shangwei New Materials, at the end of 2025.

I. Mass Production is the Barrier — Quantity Represents Accumulated Capability

There’s a real rule in the humanoid robot industry: being strong in demos doesn’t mean you’re strong in delivery — delivery capability is the turning point for commercialization.

Zhimu’s changing pace offers an excellent reference. In December 2025, it had cumulatively delivered 5,000 universal embodied robots. By March 2026, that number jumped to 10,000. By June, just three months later, it hit 15,000.

Sinoma Technology, together with Zhipu AI, has brushed up on its "presence.

Moving from yearly growth to quarterly leaps suggests a major bottleneck has been broken. This likely reflects shifts in industry supply — recently, capital markets have also increased their focus on the robot产业链 thought

First, consider the supply chain. A single humanoid robot contains more than the sum of its parts — from harmonic reducers and torque sensors to dexterous hand joints and embedded compute chips, each component used to limit production volume. The ability to rapidly scale output implies that domesticization and standardization of these parts have reached a level capable of supporting mass manufacturing.

Zhimu has reported that it’s established the world’s first standardized supply chain system for embodied intelligence robots, covering the full upstream-downstream spectrum. The 15,000-unit figure validates this industrial capability. A mature supply chain not only enables replicable industrial manufacturing but also drives down per-unit costs to commercially viable levels.

Second, there’s the deployment front. Selling one robot into a factory is different from integrating it into an active production line where it consistently adds value. How well it’s scheduled, how quickly it adapts, how fast anomalies are resolved — all determine whether the customer will order a second or third unit.

Accumulating deployment experience is essentially organizational learning. A team that’s handled 100 scenarios operates on a completely different level from one that’s done 5. Delivering 15,000 units means Zhimu now has an engineering force with massive real-world deployment expertise — an invisible barrier.

Third, perhaps most fundamentally, the data flywheel has started turning. Every robot running in a real scenario generates data — on motion control, task execution, and environmental interaction. For embodied intelligence, data isn’t icing on the cake; it’s the fuel for the next stage of evolution.

While large language models leveraged decades-old internet text corpora, robot manipulation data doesn’t exist pre-collected — it must be accumulated from operational experience. Zhimu’s spin-off, Mifeng Technology (launched Feb 2026), aims to build a 100-billion-hour database of physical world interactions. This strategy aligns with the belief that large-scale deployment opens the data faucet, improving models, which in turn strengthens deployment capabilities.

Sinoma Technology, together with Zhipu AI, has brushed up on its "presence.

Sinoma Technology, together with Zhipu AI, has brushed up on its "presence.

II. Frequent Robot Appearances — But Zhimu’s Intent Goes Beyond Visibility

Sending the Qiyuan Q1 to host World Cup commentary isn’t just about showing off. It’s about redefining what a robot can be in a social context.

The Q1 doesn’t tout industrial strength or demo box-carrying feats. Instead, it’s placed in an environment that demands expression, interaction, and personality. Its job isn’t precision torque control — it’s engaging with viewers, delivering believable commentary, and being perceived as an enjoyable “character.”

This is the core problem for consumer robots: why would someone want a robot at home? Not for productivity — for emotional connection.

Design-wise, the Q1 takes a pragmatic path — small enough to fit in a backpack, versatile for research, hacking, or companionship.

Sinoma Technology, together with Zhipu AI, has brushed up on its "presence.

However, the real lifeblood of personal robots will always lie beyond the factory. It requires an ecosystem where developers create new behaviors and users share discoveries. Each time someone hacks a new trick for the Q1, it’s free marketing — and proof that the platform is alive.

Conclusion: Balancing Scale and Imagination

Zhimu’s strategy seems clear: industrial production validates the science and sustains the model; consumer engagement validates the demand and inspires the future.

The 15,000-unit milestone shows that robots are no longer “just prototypes” — they’re a scalable business. But the TV appearance reminds us that robots need to be more than machines — they need to be part of our narratives.

This is exactly the foundation that Zhaoyuan has been accumulating over the past few years in developer tools — the AIMA open platform, the Genie Studio development environment, the AGIBOT World dataset, and the Motion-Between motion basement model. These assets initially served scientific research users and industrial partners, but once they or similar development support are combined with consumer-level products, they can be downgraded into lighter creation tools, allowing ordinary players to participate.

Qiyuan Q1 is a test of the entire future consumer ecosystem, leading developers and ordinary families to demonstrate in a specific scene. Of course, it also allows more people to pay attention once again to Shanghai Weida New Material, a company where Zhichi Jun has served as chairman — this company, which has completely transformed into Zhaoyuan Robotics, serves as the “physical shell.”

Three, behind the dual tracks: from building robots to building a “embodied universe”

The robotics industry has shown significant divergence this year, particularly in the supply chain. A batch of “boutique” companies focusing exclusively on data, deep-sea or aerospace specific scenarios, dexterous hands, component perception, and specialized technology development are rising. Additionally, some other supply chain companies are attracting new focal points due to upstream advantages like materials.

Correspondingly, Zhaoyuan’s actions are also shaping a more systematic industrial landscape. Currently, Zhaoyuan has six independently operated subsidiaries, each guarding its own front:

Zhiding, which makes cleaning products; Qingtianzu, which offers a rental platform; Critical Point, which produces dexterous hands; Mifei, which collects data; Koutuo, which makes quadruped robots; and Shanghai Weida Qiyuan, which carries consumer-level products and a capital platform. Each subsidiary has independent financing capabilities and operational teams, but from scenario positioning to technical foundations, they form a complementary relationship rather than competing.

The benefit of this structure is that it physically separates businesses requiring different organizational genes — a rental team and a data team cannot use the same evaluation system — while ensuring that underlying technologies can be shared and reused.

The role of Shanghai Weida is particularly noteworthy, not just as a capital channel. As a well-known “former demon stock,” Shanghai Weida’s original main business was high-performance composite materials and wind turbine blade resins. Its lightweight, high-strength material technologies are already being used in the structural component development and validation of Qiyuan Q1.

Sinoma Technology, together with Zhipu AI, has brushed up on its "presence.

Cost control, weight optimization, and safety compliance — the challenges facing future consumer-level robots will ultimately transfer to the material level. Having an internal, closed-loop materials platform means that design can be reverse-optimized from the material side during the early stages of product definition, which is a direct advantage in industrial collaboration.

More broadly speaking, Zhaoyuan’s layout reveals a developing industrial trend: humanoid robots are transitioning from a race to build prototypes to a race to build scale and sustain operations.

In the prototype-building phase, a clever algorithm team and a precise hardware team can stand out. But in the scale operation phase, supply chain management, deployment efficiency, maintenance systems, data recirculation, developer communities, and financing platforms — these abilities, seemingly far from “showing off,” are the true determinants of success. At this point, capital is important, but not that important. Zhaoyuan’s rise is primarily a combination of capital and team capabilities; without the team, capital alone is insufficient.

Zhaoyuan’s “358 Macro Plan,” which targets 10 billion, 100 billion, and 1,000 billion in three, five, and eight years respectively, actually reflects a judgment: the explosion of embodied intelligence will not rely solely on software iteration like the mobile internet did, but will require industrial scenes to first drive hardware, supply chains, and data flywheels to a critical mass, before consumer markets take over the brand and ecosystem relay.

This is a classic “heavy industry builds capability, light consumption builds brand” dual-wheel model, where neither replaces the other but co-exists symbiotically. It doesn’t matter which wheel turns first; the key is that both wheels share the same axis. Hence, Zhaoyuan has developed a full-stack capability of “core + data + model + platform.”

The biggest upcoming highlights are the objective improvement of industrial data and how experiences from large-scale deployments can feed back into smaller-scale ones to lower development costs and improve experience. Additionally, accumulated user interaction data must inform human-machine collaboration and flexible operations. Only then will Zhaoyuan’s rich layout avoid the trap of “biting off more than it can chew” and instead form a system of mutually reinforcing advantages.

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry