文 | 新立场pro
1848年，美国加州的淘金热刚刚兴起，旧金山的一把铁锹已经卖到了10美元。
亲历这场淘金热的记者爱德华·巴芬，在当年的记录中还提到，一个淘金用的锡盘要5美元，一双普通靴子卖16美元。黄金能不能挖到还不好说，但涌入加州的人们，已经愿意为挖金子的机会付出不小的代价。
此后180年，人类“淘金”的热情从未消减，如今有相当一部分落在了具身智能行业。机器人还在学习如何进入工厂、完成工作，投资人却已开始为它们未来几年的市场出价。
9月28日，正在冲刺科创板IPO的云深处更新了招股书。2026年上半年，这家以工业机器狗起家的公司实现收入2.59亿元，同比增长117.67%。但利润并没有跟上，2025年，云深处首次实现年度盈利，归母净利润2868万元；半年后，又亏损了881万元。
亏损之外，云深处准备在未来五年，把机器人产能扩大到远超现在销量的规模。2025年，公司卖出2531台四足机器人、377台轮足机器人和1台人形机器人，合计2909台。而在此次IPO募投计划中，云深处预计第五年完全达产后，新增5.68万台年产能，包括3.88万台四足机器人、1.5万台轮足机器人和3000台人形机器人。
比云深处产能扩张更快的，是它的估值。2024年8月，云深处的融资估值约10亿元；到2025年12月，投后估值达到52.50亿元，一年多涨至原来的5.25倍。据统计，2026年上半年国内具身智能赛道发生322笔融资，总额935亿元，较上年同期增长约5倍。云深处也在这场融资热潮中加快投入，此次IPO，公司计划募集25.03亿元，接近2025年全年收入的7.4倍。其中相当一部分钱要投向具身模型研发。
这是云深处此前投入不多的领域，也是一个还很难判断何时能靠产品销售收回成本的方向。
而这种投入与实际需求之间的落差，正在成为整个具身智能行业的麻烦。一些看起来不错的销售成绩，背后甚至包括数采中心、展演项目，以及投资方和生态合作伙伴的采购。
投资人也动摇了对具身智能未来的信任。9月，路透社报道称，部分一级市场机器人项目的估值已经下调30%至50%。这让朱秋国面临一个有些尴尬的处境：云深处已经比不少同行更早赚到了钱，却还需要投入更多资金，去追赶一个尚未真正成熟的市场。
成熟业务，利润开始变薄
云深处过去几年最有说服力的成绩，是让机器狗真正进入了工业现场。
2019年起，公司与国家电网、南方电网下属单位合作，尝试将四足机器人用于电力巡检。变电站设备密集，部分区域存在台阶、电缆沟和不平整路面，传统轮式巡检机器人不容易通过，四足机器人可以跨越部分障碍，再搭载摄像头、传感器和检测设备，替人完成部分巡检工作。云深处从这些具体需求出发，逐渐把绝影系列做成面向工业客户的产品。
按照招股书披露的信息，公司的机器人已经应用于国家电网、南方电网下属超过100座变电站。到2025年，来自行业应用的收入达到2.67亿元，占主营业务收入的79.33%。根据公司援引的弗若斯特沙利文数据，云深处当年四足机器人行业应用收入排名全球第一。
有客户愿意持续采购，是云深处与一些尚未形成稳定销售的机器人公司的区别。招股书问询回复披露，按合并口径统计的非终端客户A，2023年的相关销售金额为575.62万元，2024年增加到2855.40万元，2025年进一步达到3060.98万元。
卖出去的设备更多了，绝影X却没以前那么赚钱了。2025年，云深处最重要的行业级产品绝影X系列销售681台，平均售价28.75万元，毛利率54.35%。到了2026年上半年，销量达到541台，平均售价却降至22.82万元，毛利率也下降到39.75%。按平均数据计算，每卖出一台绝影X，云深处获得的毛利从约15.63万元降至9.07万元，减少约42%。
售价下降的同时，平均单位成本并没有跟着降低，反而从13.12万元增加到13.75万元。云深处的解释是，上半年集中交付了部分价格、毛利率较低的细分型号，这些产品在绝影X系列收入中的占比提高，拉低了整体平均水平。这个解释能说明平均毛利率为什么下降，但如果未来更多客户采购的是低价型号，云深处能否靠生产规模扩大，把单位成本降下来？
毛利率下降的同时，研发费用还在往上涨。上半年研发费用达到7996万元，已经接近2025年全年的8430万元。经营活动现金流由去年全年的净流入6375万元，转为今年上半年的净流出2531万元。收入依然保持高速增长，新增利润却没能覆盖同期增加的开支。
而在客户关系里还隐藏着另一层麻烦：部分向云深处采购机器人的企业，正在布局同类产品。10月14日，《北京商报》报道，云深处核心客户A1是一家央企子公司。公司在招股书中提示，包括该客户在内的部分主要客户，正在通过自主研发、生产及外部并购布局同类型产品。
工业客户采购机器狗，往往还需要考虑行业应用方案、系统集成、后续维护和长期成本。随着机器人进入更多工作现场，客户对产品和供应链的了解也在增加。采购规模越大，部分客户就越有动力考虑自主研发，或者引入其他供应商。
云深处较早进入电力巡检市场，也因此建立了客户关系和行业经验。但客户开始布局自研产品，意味着这层优势并不是永久的。未来客户可能继续采购云深处，也可能逐步提高自研产品的使用比例，或者转向其他供应商。
云深处并不是没有新产品接棒，2025年才开始销售的山猫M系列，到了2026年上半年已经卖出482台，超过去年全年的销量。它的毛利率不降反升，达到60.04%，贡献的毛利也第一次超过绝影X。
只是，这些生意增长得再快，也还没有达到资本市场对具身智能企业的期待。云深处下一阶段准备投入的资金，远远超过了它过去几年为开发这些产品花的钱。
诗和远方，还是心比天高
资本市场对科技企业的偏好，向来是颇有章法。
既要有能够赚到钱的基本盘，又要有足以支撑增长的第二曲线，最好还能留下一片诗和远方，供投资人为尚未到来的市场慷慨解囊。
云深处倒是把这三样都给凑齐了。四足机器人撑起业绩，轮足机器人负责描绘增长，人形机器人和具身大模型则承担起更远大的使命。不同业务“各司其职”，从当下的收入到未来的想象，一份符合资本市场审美的成长叙事貌似已相当完整。
如今故事讲到第三层，生意也开始变得越来越远。
云深处靠工业级机器狗起家，在电力巡检、警务安防等领域积累了客户，2025年首次实现年度盈利。这是朱秋国经营多年换来的成绩，但工业巡检终究是一门有边界的生意，市场空间可以估算，客户预算也有迹可循，很难支撑无限延伸的增长预期。
轮足机器人倒是提供了新的产品选择，但主要应用仍集中在警务安防、电力巡检等熟悉的领域。从四足走向轮足，业务版图看上去又扩大了一圈，但原有客户能否因此贡献更多采购预算，还需要订单说话。
具身智能给朱秋国提供了更加广阔的发挥空间。人形机器人能够进入工业制造、物流、商业服务乃至家庭生活，几乎每个领域都可以写进未来的市场规划。相比变电站和工厂里那些具体的采购需求，这里的想象显然宽裕得多。
工业机器狗的价值要靠客户一台台买出来，人形机器人的价值却可以先按照未来可能卖出的数量来讨论。市场尚未成熟，反倒省去了许多现实约束。
不过，云深处的人形机器人业务，确实还没有太多现实订单带来的“烦恼”。
2024年，公司售出3台人形机器人；2025年仅卖出1台；2026年上半年增加到6台，实现收入279万元。而此次IPO，云深处已经计划在募投项目第五年完全达产后，新增3000台人形机器人年产能。
从一年卖出1台，到准备建设3000台年产能，留给市场需求成长的任务，未免有些繁重。
此次IPO，云深处拟募集25.03亿元，其中约11.69亿元投入具身算法及模型研发，另有5.54亿元用于机器人本体与解决方案研发，两项合计17.23亿元。
但在此前的审核问询回复中，云深处承认，公司尚未对具身智能大模型等“大脑”技术进行大规模研发投入，与全球领先水平相比，技术布局相对滞后，场景实训积累也较为有限。
过去靠机器狗积累的技术和订单，撑起了云深处的经营基本盘；如今最舍得花钱的方向，恰恰是此前积累尚显不足的领域。
具身大模型当然值得研究，只是技术积累有其规律，算力、人才和真实场景数据并不会因为一笔募资便自然转化为竞争优势。行业未来究竟需要怎样的模型、多少人形机器人能真正进入工作岗位，至今都缺乏充分的商业验证。
研发路线尚待验证，资本市场却已开始提前计价。2024年8月，云深处融资估值约10亿元；到2025年12月，投后估值达到52.50亿元。按照2025年3.37亿元营收计算，对应约15.6倍市销率，而当年扣非净利润只有1512万元。
估值上涨有经营增长和行业预期等多重因素，但现有利润显然还不足以独自解释这样的价格。人形机器人和具身大模型代表的远期市场，至少为投资人提供了更多讨论未来价值的空间。
朱秋国已经用机器狗证明了过去，也用轮足机器人交代了所谓“增长”，又用人形机器人安排好了未来。连尚未建立充分优势的技术方向，都在募投计划里获得了足够体面的预算。
至于这些投入何时兑现为利润，仍要等产业成熟、技术突破和客户采购共同给出答案。资本市场可以提前为五年后的市场出价，企业却不得不从眼下开始支付研发、扩产和人员成本。
云深处已经学会了按照未来的市场安排今天的投入，而具身智能赛道上，抱着同样打算的企业还有很多。
倘若每家公司都按照自己将成为行业赢家的规模提前下注，那么这个行业大概需要比预期中更多的赢家，才能让所有人的故事圆满收场。
钉子太多，锤子不够用了
朱秋国甚至不是具身智能行业最激进的一个。
2026年，国内具身智能公司的估值已经跑在商业化前面。公司越开越多，融资金额越来越大，真正能够证明机器人具备规模化生产力的企业，却还是少数。
据统计，今年8月，国内具身机器人及核心零部件领域披露56起融资。其中，小鹏机器人业务、Sharpa、帕西尼感知、中科第五纪和原力无限五家企业，合计披露融资规模约140亿元。近半数融资仍处于种子轮、天使轮和Pre-A阶段。
钱越来越集中到少数头部公司手里，一边是成立不久的企业不断拿到新融资，另一边是越来越多公司需要几亿元、几十亿元，才能继续参与下一轮竞争。
今年3月，银河通用宣布完成25亿元融资，估值超过200亿元。4月，成立于2023年的星海图披露近20亿元B+轮融资，加上2月完成的近10亿元B轮融资，两个月内便拿到近30亿元，估值也进入200亿元梯队。
但在4月接受采访时，星海图CFO罗天奇仍然认为，2026年是具身智能行业的POC（概念验证）元年，重点是跑通从0到1，而非追求量产。企业还在验证机器人能不能真正干活，资本市场已经开始为它几年后的市场地位定价。
随着企业开始争夺真实订单，具身智能行业的商业化收入本身，也越来越需要仔细辨别。
2025年，智元实现10.5亿元收入。创始人邓泰华提出，2027年营收要跨越100亿元。到了今年10月，据《财经》报道，智元又启动了10亿至15亿元的新一轮融资，最新估值约500亿元。
智元确实已经形成一定的出货规模，但它的部分投资方和生态合作伙伴，同时也是机器人的采购方。去年7月，智元中标珠海具身智能应用创新中心数采及展演机器人采购项目，金额1273.6万元。采购方珠海智汇元启，是智元与当地国资等共同成立的合资公司，智元持股30%。
数采中心需要机器人收集训练数据，机器人公司也确实需要这些数据改进模型。但这也是本轮具身智能热潮里一个容易被忽略的问题：卖出了机器人，不代表机器人已经在工厂里创造了相应的生产力。
这个问题在9月被摆到了台面上，9月10日，刚刚带领梅卡曼德完成港股上市的CEO邵天兰，公开批评部分“攒局型”具身智能企业。他质疑，一些公司依靠数采中心、租赁公司和关联交易形成不可持续的收入，却急于通过大新闻和高估值推进上市，邵天兰甚至直接点名银河通用。
行业内部已经开始质疑一些订单的商业价值，以前，投资人关心企业的机器人能完成多少动作；现在，资本开始询问这些理想主义者：是谁买了机器人，为什么买，明年还会不会继续买？
资本市场也开始表现出不同于上半年的态度，9月21日，路透社援引知情人士消息称，国内监管部门通过非正式窗口指导，放缓了部分人形机器人企业的IPO进程，一位接受路透社采访的投资人还表示，部分一级市场机器人项目的估值已经下调30%至50%。
二级市场的反应则更加直接，8月19日，宇树科技以150.80元的发行价登陆科创板，开盘价达到1100元，市值一度超过4449亿元。按照开盘价计算，中一签500股的投资者，账面浮盈达到47.46万元。
但这份狂热没能持续太久，到10月16日，宇树股价跌至426.98元，较首日开盘价下跌约61%。10月20日，股价虽然反弹至526.03元，公司仍发布了交易异常波动公告，提醒投资者注意估值风险。
公告显示，宇树当时的滚动市盈率达到388.25倍，而所属通用设备制造业最近一个月的平均滚动市盈率只有39.51倍。宇树还特别提醒投资者，人形机器人仍处于商业化早期，如果规模化落地不及预期，可能面临估值回归风险。
连已经做到近17亿元年收入、拥有成熟产品线的宇树，都无法让资本市场对它的估值形成稳定共识，更年轻的具身智能公司就更难了。
而且，融资带来的压力并不会随着估值回落而消失。企业拿到钱，就要招研究员、建设数据中心、扩充产线、训练模型。竞争对手也在做同样的事。为了不被甩下，企业需要不断投入，而投资机构不可能让所有公司都获得同样规模的下一轮融资。
当越来越多企业都按自己将成为行业头部来安排支出，最后能够支撑这些规划的真实市场需求，未必足够大。
云深处也无法置身事外。它在工业机器人市场已经站稳了一部分脚跟，也确实需要补齐具身大模型的短板。但眼下公司准备一次募集25亿元，其中17亿元以上用于研发，而2025年全年收入只有3.37亿元，人形机器人只卖出1台。
朱秋国要同时守住机器狗的生意、扩大轮足机器人市场，还要追赶具身大模型和人形机器人。每条业务线都需要投入，每个对手又都在加速。
过去，具身智能公司只需要证明自己有机会成为未来的赢家。现在，投资人开始要求它们拿出证据，证明自己不会只是下一家被淘汰的公司。
写在最后
云深处花了几年时间，把机器狗卖进了变电站、工厂和其他工业现场，找到了愿意持续采购的客户，也在2025年第一次实现年度盈利。取得这些成绩不容易，也是它今天能够走到IPO门口的重要原因。
但这门生意刚开始赚钱，云深处就需要为更大的市场做准备。这当然可能是一笔必要的提前投入，机器人技术还在快速变化，等市场完全成熟再扩产，未必来得及。但当整个行业都在加快融资、扩大团队、建设产能的时候，一个问题也越来越难回避：这些机器人最后要卖给谁？
朱秋国在浙大研究机器人多年。2017年创业时，他和李超拿出30万元积蓄，想的是先让公司活过三个月。如今，他需要向资本市场解释五年后的产能和收入。很难判断这份计划有多少来自技术判断，又有多少受到融资窗口和行业竞争的影响。
但从实验室走到IPO，有些事情已经变了。技术验证可以反复试错，融资却有窗口期。朱秋国或许比谁都清楚，让机器人真正进入产业需要时间，只是公司未必总有时间慢慢来。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
New position pro
In 1848, the gold rush in California in the United States just rose, and a shovel in San Francisco was sold for $10.
Eduard Buffin, a gold-digger journalist, also mentioned in the record of the year that a tin dish for gold was worth $5 and a pair of ordinary boots sold $16. It is hard to say if gold can be dug up, but people in California are already willing to pay a considerable price for opportunities to dig gold.
Since then, the enthusiasm for human gold mining has never diminished, and much of it is now in the smart industry. Robots are also learning how to get into factories and finish their work, and investors have started bidding for their markets in the coming years.
On 28 September, the book of admissions was updated in the depths of the clouds of the Sprinting Section. In the first half of 2026, this industrial machine-based company earned $259 million, an increase of 117.67 per cent over the same period. However, profits were not kept up, and in 2025, for the first time in the clouds, annual profits were realized, with a net profit of $2.86 million for the mother; six months later, there was a loss of $8.81 million.
Beyond the losses, the clouds are preparing to expand robotic capacity to a much larger scale than is currently sold over the next five years. In 2025, the company sold 2531 four-foot robots, 377 wheel-foot robots and one man-shaped robot, totalling 2909. In the context of this IPO campaign, the cloud is expected to reach full capacity in the fifth year, with an additional 56.8 million annual capacity, including 38.8 million four-foot robots, 15,000 wheel-foot robots and 3,000 man-made robots.
It is its valuation that expands more rapidly than deep clouds. In August 2024, financing at the depth of the cloud was valued at approximately $1 billion; by December 2025, the post-investment valuation had reached $5.25 billion, more than 5.25 times the previous year. In the first half of 2026, 322 financings were recorded for the country ' s smart track, amounting to $93.5 billion, a fivefold increase over the same period of the previous year. Deep in the clouds, investment has also accelerated in this boom, when IPO plans to raise 2.503 billion yuan, close to 7.4 times its annual income in 2025. A significant part of this money goes to the development of models.
This is an area that was previously poorly invested in deep clouds and where it is difficult to determine when costs can be recovered from the sale of products.
This gap between inputs and actual demand is becoming a problem for the entire smart industry. Some of the sales that appear to have done well include even centres, exhibition projects, and procurement by investors and ecological partners.
Investors have also shaken confidence in a smart future. In September, Reuters reported that the valuation of some level-I market robot projects had been reduced by 30 to 50 per cent. This has left Zhu in a somewhat awkward position: deep clouds that have earned money earlier than many colleagues, but still need to invest more money to catch up with a market that is not yet truly mature.
Growing-up business, profit starts thinning.
The most convincing achievement of the past few years in the clouds has been the real access to industrial sites by machine dogs.
Since 2019, the company has been working with the national grid and subdivisions of the southern grid to try to use four-foot robots for electricity inspection. The transformer plant is highly equipped, with some areas with stairwells, cleavages and uneven surfaces, and traditional wheeled survey robots are not easy to pass, and four-foot robots can cross some of the barriers and carry cameras, sensors and detection equipment to complete some of the inspections. Deep in the clouds, these specific needs have led to the gradual transformation of the insinuation series into industrial-oriented products.
According to the information disclosed in the equity book, the company ' s robots have been applied to more than 100 substations under the national grid and the southern grid. By 2025, revenues from industrial applications amounted to $267 million, or 79.33 per cent of the income from the main operation. According to the Frost Sullivan data quoted by the company, the four-foot robotic industry in the dark was ranked among the highest in the world in terms of income from its application.
Clients are willing to buy on a continuous basis, as distinct from some robotic companies that have not yet developed stable sales. In response to queries in the requisitioning letter, it was revealed that the related sales by combined calibre, non-end customer A, amounted to $5.7562 million in 2023, increased to $2855.4 million in 2024 and further reached $30.609 million in 2025.
More equipment is sold and Shadow X is less profitable than before. In 2025, the most important industry-level product in the clouds was insulated from the X series, which sold 681 items at an average cost of $287.5 million and a Maori ratio of 54.35 per cent. By the first half of 2026, sales had reached 541, while the average sale price had fallen to $228.2 million and the Māori rate to 39.75 per cent. In average terms, for every single insulation X sold, the amount of Maori acquired in the depths of the clouds fell from approximately $1563 million to $90.7 million, a reduction of about 42 per cent.
The decline in sales prices has not been accompanied by a decrease in the average unit cost, which has increased from $13.12 million to $137.5 million. Deep in the clouds, it is explained that the first half of the year concentrated on sub-models with a low rate of Māori, which increased their share of the insulated X series income and lowered the overall average. This explanation explains why the average Māori rate has fallen, but if more clients are purchasing low-price models in the future, would it be possible to reduce unit costs by increasing the scale of production in the clouds?
The decline in the Māori rate has been accompanied by an increase in R & D costs. The cost of R & D in the first half of the year amounted to $799.6 million, which is close to $84.3 million throughout 2025. Cash flows from operating activities changed from $6.375 million net to $2.531 million net outflows in the first half of the year. Income continued to grow at a high rate, while new profits did not cover increased expenditures over the same period.
There is another layer of trouble in the customer relationship: some of the companies that buy robots from deep clouds are setting up similar products. On 14 October, the Beijing Business Journal reported that core client A1 in the clouds was a subsidiary of Zhang. In its equity book, the company suggests that some of its main clients, including the client, are developing, producing and externally acquiring the same type of product.
Industrial clients procure machines and dogs, often also taking into account industry applications, system integration, follow-up maintenance and long-term costs. As robots enter more work sites, client knowledge of products and supply chains increases. The larger the procurement, the greater the incentive for some clients to consider autonomous research and development or to introduce other suppliers.
The early entry of clouds into the electricity inspection market has also led to customer relations and industry experience. However, the start of the self-study product layout of the client means that the advantage is not permanent. Future clients may continue to purchase deeper, may also gradually increase the use of self-study products or shift to other suppliers.
Deep in the clouds, not without new products, the Cat M series, which began to be sold only in 2025, has sold 482 units in the first half of 2026, exceeding the volume sold throughout last year. Its Māori rate has not declined to 60.04 per cent, and for the first time it has contributed Māori more than Insight X.
However, these businesses are growing faster than capital markets expect for smart firms. The funds to be invested in the next phase of the cloud far exceed what it has spent over the past few years to develop these products.
Poetry and far away are still higher than heaven.
The preference of capital markets for science and technology enterprises has always been well organized.
It would be preferable to have both a basic set of money and a second curve that would sustain growth, leaving a poem and a faraway place for investors to contribute generously to markets that had not yet arrived.
Deep in the clouds, they're all all set. Four-foot robots carry on with their performance, rotating robots are responsible for depicting growth, and human robots and body LLM take on a larger mission. A growth narrative consistent with the aesthetics of the capital market appears to have been quite complete, from current revenues to future imaginations, for different operations.
Now that the story's on the third floor, business is getting farther and farther.
Deep in the clouds, industrial-grade machines and dogs have built up clients in areas such as electricity inspections, police security and so forth, with annual profits realized for the first time in 2025. This has been the result of many years of Zhu ' s operation, but industrial inspection is ultimately a borderline business, market space can be estimated and client budgets can be followed, and it is difficult to sustain unlimited growth expectations.
While rotational robots provide new product options, the main applications remain concentrated in familiar areas such as police security and electrical inspection. From four to four, the business chart appears to have expanded in one circle, but the ability of the original client to contribute more to the procurement budget requires an order to speak.
Intelligentness has provided Zhuqiu with wider space to play. Human robots have access to industrial manufacturing, logistics, commercial services and even family life, and almost every field can be written into future market planning. The imagination is clearly much broader than the specific procurement needs in the transformer stations and plants.
The value of industrial machines and dogs is bought by one client, but the value of human robots can be discussed first according to the amount that may be sold in the future. The market is not yet mature and many realistic constraints have been reversed.
It is true, however, that the human robotic business at the depth of the cloud does not yet have much of the “challenges” caused by realistic orders.
In 2024, the company sold three man-shaped robots; in 2025, only one was sold; in the first half of 2026, it increased to six, with a revenue of $2.79 million. This time, IPO, in the dark, plans have been made to fully reach the fifth year of the fund-raising project, with an additional 3,000 man-shaped robots producing annually.
From selling one a year to preparing to build 3,000 annual production capacity, the task left to market demand growth is somewhat onerous.
This time, IPO is planning to raise 2.503 billion yuan in deep clouds, of which approximately 1,169 million yuan are invested in the development of physical algorithms and models, and 554 million yuan in the development of robotics and solutions, totalling 1,723 million yuan.
However, in previous audit responses, it was acknowledged that companies had not yet made large-scale R & D investments in “brain” technologies, such as smart LLM, and that the technology landscape was lagging behind global lead levels and the accumulation of hands-on training was limited.
The technology and orders accumulated in the past by robots and dogs have raised the fundamentals of business in the depths of the clouds; the most expensive areas today are precisely areas where accumulation has not been sufficient.
Of course, LLM is worth studying, except that technology accumulation is regular, and data on calculus, talent and real scenes do not automatically translate into a competitive advantage by raising funds. What models are needed for the future of the industry and how many man-made robots can actually enter the workforce are still lacking sufficient commercial certification.
Research and development routes have yet to be validated, while capital markets have begun to price in advance. In August 2024, the deep cloud financing was valued at approximately $1 billion; by December 2025, the post-investment valuation had reached $5.25 billion. Based on a collection of $337 million in 2025, the market sales rate was about 15.6 times higher, compared to $1.512 million in non-net profits.
The increase in valuations has multiple factors such as business growth and industry expectations, but it is clear that current profits are not sufficient to explain such prices alone. Human robots and the forward market represented by LLM provide at least more room for investors to discuss future values.
Ju Autumn has proven his past with machine dogs, and he has written about so-called “growth” with rotor-foot robots, and he has arranged his future with human robots. Even technical orientations that have not yet established sufficient advantages have received a sufficiently decent budget in the fund-raising programme.
The question of when these inputs will be translated into profits will still have to be answered jointly by industry maturity, technological breakthroughs and customer procurement. Capital markets can bid five years in advance, while enterprises have to start paying R & D, expansion and human costs immediately.
Deep in the clouds, today's inputs have been learned according to future market arrangements, and there are many companies with the same intentions on the smart track.
If each company bets ahead of schedule on the scale in which it will be the winner of the industry, then the industry will probably need more winners than expected to bring the story to a successful conclusion.
Too many nails, not enough hammers.
Zhu Aki is not even the most radical of the intelligent industries.
In 2026, the valuation of domestic smart companies was already ahead of commercialization. The larger the company, the larger the amount of financing, the smaller the number of businesses that can truly prove the size and productivity of robots.
According to statistics, 56 cases of financing were disclosed in August this year in the area of robotics and core parts. Of these, five businesses, namely, Pyramid Robotics, Sharpa, Pasini Sensitization, Middle Quinque and Force, disclosed a total of approximately $14 billion in financing. Nearly half of the financing is still in stages of seed wheels, angel wheels and Pre-A.
Money is increasingly concentrated in the hands of a small number of front companies, on the one hand, new financing for new businesses that have not been established, and on the other hand, a growing number of companies need hundreds and billions of dollars to continue to compete in the next round.
In March, Galaxy announced the completion of $2.5 billion in financing, valued at more than $20 billion. In April, the Star Chart, which was established in 2023, revealed nearly $2 billion in B+ finance, plus nearly $1 billion in B in February, which resulted in nearly $3 billion in two months and the valuation of the 20 billion ladder.
However, at the time of the interview in April, Star Chart CFO Rottench continued to believe that 2026 was the year of the POC (conceptual validation) in the smart industry, with the emphasis on running from 0 to 1, rather than pursuing volume production. Businesses are also testing whether robots can really work, and capital markets are already pricing their market position a few years later.
As businesses begin to compete for real orders, the commercialization revenues of the smart industry itself increasingly need to be carefully identified.
In 2025, the figure was 1.05 billion yuan. The founder, Deng Taehua, suggested that the 2027 battalion would be over $10 billion. By October of this year, according to the Book of the Book, a new round of financing of between $1 billion and $1.5 billion was launched, with the latest estimate of about $50 billion.
It is true that Ji-won has developed a certain scale of output, but it is part of its investor and ecological partners, as well as a buyer of robots. In July of last year, the Centre for Innovation in the Use of Intellectually Smart Applications in the Sea was awarded a total of $12,736,000 for the acquisition and display of robotics. The purchaser, Ji-won Hai Ji-sik, is a joint venture of Ji-won, a 30 per cent joint venture with local capital.
Digital centres need robotics to collect training data, and robotics do need them to improve models. But this is also an easy problem to ignore in this wheel of smartness: selling robots doesn't mean that robots have created corresponding productivity in factories.
This issue was put on the table in September, and on 10 September, Mercamand was just led to the completion of the CEO ' s Sunland, which is listed in Hong Kong ' s stock market, publicly criticizing some of the "surplus" smart businesses. He questioned the fact that some companies relied on centres, leasing companies and related transactions to generate unsustainable revenues, but were eager to push their listings through news and high valuations, even by direct naming Galaxy.
Within the industry, the commercial value of some orders has begun to be questioned. Before, investors were concerned about how much robots could do; now, capital is asking the idealists who bought robots, why, and will they continue next year?
Capital markets have also begun to show attitudes different from those of the first half of the year, with Reuters citing sources that on 21 September, domestic regulators slowed down the IPO process for some man-made robotic enterprises through informal window guidance, and an investor interviewed by Reuters indicated that the valuation of some level-I market robot projects had been reduced by 30 to 50 per cent.
The response was more direct in secondary markets, where on 19 August, Wootreetech landed on board at a issuing price of $150.80 for $1,100 and a market value of over $44.49 billion. On the basis of opening prices, one investor with a 500 share carrying a surplus of $4,746,000.
However, this fanaticism did not last much longer, and by 16 October the value of the Woo-tree stock fell to $426.98, a decline of about 61 per cent over the opening price of the first day. On 20 October, although the equity price rebounded to $526.03, publish issued a notice of unusual fluctuations in transactions, alerting investors to valuation risks.
According to the bulletin, the rolling market rate of Woo-tree at that time was 388.25 times, compared with the average rolling market rate of 39.51 times in the last month of the generic equipment manufacturing industry. In particular, it reminds investors that human robots are still in the early stages of commercialization and may be at risk of valuation regression if size falls short of expectations.
Even trees that have achieved nearly 1.7 billion yuan in income and have mature product lines will not allow for a stable consensus on their valuation in capital markets, and even more so for younger smart firms.
Moreover, the pressures on finance do not disappear as valuations fall. When businesses get paid, they need to hire researchers, build data centres, expand production lines, and train models. Competing people are doing the same thing. In order not to be left behind, enterprises need to invest continuously, and it is not possible for investment agencies to give all companies access to the next round of finance of the same size.
When more and more enterprises organize their expenditures as they will become the head of the industry, then the real market demand that will support these plans may not be large enough.
It's hard to get out of the clouds. It has stabilized part of the industrial robotic market, and indeed needs to be filled with LLM. However, the company is now preparing to raise 2.5 billion yuan at a time, of which more than 1.7 billion will be spent on research and development, compared to $337 million for the year 2025 and only one human robot for sale.
Zhu Autumn has to keep up the business of robots and dogs, expand the market for wheeled robots, and pursue LLM and human-shaped robots. Each line of operations requires input and each counterparty is accelerating.
In the past, smart companies had to prove that they had a chance to be future winners. Now investors are asking them to produce evidence that they will not be the next company to be eliminated.
At the end.
Deep in the clouds, it took several years to sell machines and dogs into power transformers, factories and other industrial sites, to find customers willing to buy on a continuous basis and to achieve annual profits for the first time in 2025. These achievements are not easy and are important reasons why it can reach the IPO door today.
But as the business begins to make money, deep in the clouds there is a need to prepare for larger markets. This, of course, may be a necessary early investment, and robotic technology is changing rapidly, and it may not be possible to expand production once the market is fully mature. But as the whole industry accelerates financing, expands teams and builds capacity, it becomes more and more difficult to avoid: to whom do these robots end up selling?
Chu has been studying robots in Zhejiang for years. When he and Li started a business in 2017, he and Li put out 300,000 yuan in savings in order to keep the company alive for three months. Today, he needs to explain to capital markets the capacity and income after five years. It is difficult to judge how much of the scheme comes from technical judgement and how much is affected by financing windows and industry competition.
But from the lab to the IPO, something has changed. Technical certification can be tested repeatedly, but financing has a window period. Zhu Autumn may know better than anyone that it takes time for robots to actually enter the industry, but companies may not always have time to slow down.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.