（本文作者为 公司观察，钛媒体经授权发布）
历经一年半拉锯与多次排他延期，科创新源跨境并购预案艰难出炉。
5月28日晚间，科创新源（300731.SZ）发布公告称，公司联合全资子公司香港科创，拟以2.45亿元现金，分别收购东莞兆科、新加坡智科各50%股权。同时，公司通过表决权委托，拿下两家企业各55%表决权。交易完成后，东莞兆科将成为其控股子公司，新加坡智科成为控股孙公司。
这场耗时良久的跨境产业并购，不仅是科创新源完善热管理产业链的关键落子，更被视作公司破解业绩波动、补齐业务短板、稳固盈利增长曲线的重要布局。但跨境资产整合难度以及业绩承诺兑现压力，也为这场并购的未来走向埋下诸多悬念。
相较于常规产业并购，科创新源此次跨境收购可谓一波三折，复杂的跨境股权梳理、多方利益协调以及交易结构反复迭代，让整个进程耗时整整一年半之久。
追溯交易源头，早在2024年末科创新源便首次抛出收购计划，彼时拟收购 Thermazig Limited 所持东莞兆科51%股权。然而受跨境资产权属厘清、合作各方诉求分歧、交易架构优化调整等多重因素影响，并购进程屡次按下暂停键，排他协议多次延期，历经数轮细节磋商与条款修订，直至2026年5月完整收购预案才正式出炉。
为兼顾股权归属与经营管控权，本次交易采用股权收购+表决权委托的双层运作模式。科创新源与香港科创合计出资2.45亿元现金，收购东莞兆科、新加坡智科各半数股权，再通过表决权委托方式锁定两家企业55%表决权，以此实现对标的资产的绝对控股。
本次东莞兆科体系包含昆山兆科电子材料有限公司，新加坡智科则涵盖重组后的中国台湾兆科科技、越南Ziitek Technology等海外主体，形成了覆盖大陆、港台、越南的全球化生产与销售网络。
标的资产东莞兆科是高端电子散热及密封材料领域的优质专精企业，聚焦导热、加热、密封、EMI电磁屏蔽四大核心品类，主打产品覆盖导热界面材料、TIF900系列EMI材料、K-heat加热材料、Z-foam发泡硅胶密封材料等多个细分品类。
尤其在核心的导热界面材料领域，东莞兆科已完成高分子基、金属基、新型热界面材料的全品类布局，导热硅脂、导热垫片、导热凝胶、液态金属、碳纤维导热垫片等产品均实现量产与商业化应用。凭借成熟的技术工艺与稳定的产品品质，兆科已斩获海内外头部客户资源，建立了深度长期的合作关系，具备扎实的产业化与市场落地能力。
本次标的估值采用收益法核算，评估基准日为2025年12月31日。数据显示，东莞兆科合并口径股东权益评估值达4.9亿元，较1.76亿元账面净值增值3.14亿元，增值率178.02%，高额溢价充分印证资本市场对其技术壁垒、赛道前景的高度认可。
与此同时，交易设置了清晰的业绩承诺门槛：2026年、2027年、2028年标的扣非后归母净利润分别不低于4500 万元、5000万元、5500万元，逐年稳步抬升的业绩目标，既绑定了标的经营成长预期，也对后续整合运营提出了严苛要求。
科创新源2017年登陆A股，主营高分子材料与热管理系统产品，核心品类包括防水密封、绝缘防火材料及散热金属结构件，广泛应用于通信、新能源汽车、数据中心等领域。
近两年受益于新能源汽车、AI算力产业爆发，下游热管理需求持续扩容，公司资本市场估值大幅提升，股价自阶段低点累计涨幅超247%，目前市值成功突破110亿元，跻身百亿市值赛道，资本市场热度持续走高。
市值走高的背后，是公司2025年业绩的强势爆发。财报显示，2025年公司实现营收11.57亿元，同比增长20.75%；归母净利润3560.96万元，同比大增106.08%；扣非净利润2931.68万元，同比增长130.52%，利润增速远超营收增速，盈利弹性显著。
公司业绩增长核心依托新能源汽车热管理业务，旗下子公司瑞泰克主营动力电池液冷板产品，适配麒麟电池、神行电池等主流方案，是动力电池散热的核心配套组件，凭借技术与量产优势绑定头部客户，支撑公司营收规模持续扩张。
但双增长态势未能延续，2026年一季度公司业绩迎来反转，由盈转亏。当期公司营收2.24亿元，同比下滑11.09%；归母净利润亏损299.81万元，同比下降129.75%；扣非净利润亏损350.39万元，同比下滑146.02%。短期业绩大幅波动，直接暴露了公司业务结构单一、盈利稳定性不足的核心短板。
持续下滑的毛利率，是公司盈利疲软的核心症结。数据显示，2023年至2025年，公司毛利率从25.44%逐年降至20.94%、19.24%，2026年一季度进一步跌至16.46%。
与之形成对比的是，公司营收规模持续扩容，2023-2025年营收从5.59亿元增长至9.58亿元、11.57亿元，形成营收高增长，公司净利润体量依旧偏小，甚至偶尔亏损的尴尬情形。
为突破增长瓶颈，科创新源积极布局第二增长曲线，依托全资子公司创源智热切入数据中心液冷散热赛道，发力AI算力基础设施配套业务。
2026年一季度，创源智热多款产品完成产业化验证，顺利斩获量产订单，营收同比大幅提升。但受上游芯片产能紧张制约，公司订单交付不及预期，叠加产线处于产能爬坡、工艺优化阶段，设备折旧、新增人力等固定成本攀升，导致新业务短期持续亏损，进一步拖累公司一季度整体业绩。
当前科创新源正处于传统业务盈利弱化、新兴业务尚未盈利的转型阵痛期，增长压力凸显。在此背景下，收购兆科具备极强的产业互补价值。
不过，跨境资产整合、业务协同落地、未知的业绩承诺兑现均存在不确定性，这场高价并购能否帮助科创新源摆脱盈利困境、巩固行业竞争力，仍有待市场和时间检验。(文 | 公司观察，作者 | 周健 ，编辑 | 曹晟源)
更多精彩内容，关注钛媒体微信号（ID：taimeiti），或者下载钛媒体App
（本文作者为 Chelsea_Sun，钛媒体经授权发布）
TMTPOST — If someone asks “Do you know how different farmland rents are around the world?” Zhao Feng, president of GrainCore Dynamics, offered a set of figures: “In Central China’s Henan province, annual rent for one mu of land is roughly 500 to 800 yuan. In Hainan province, that number might jump to 2,000 to 4,000 yuan. But if you look overseas—in the U.S., leasing one mu costs only about 200 yuan. And in some extremely fertile areas in Africa, annual rent for one mu can be as low as 2 yuan. With 50 yuan, you can even buy out 99 years of usage rights.”
“And Africa has five times as much uncultivated arable land as China, enough to solve the food problem for billions of people,” said Zhao.
Zhao isn’t the typical tech-founder storyteller. He’s more of a hands-on operator who’s been through ups and downs but still holds on to a dream: he’s endured the grind of starting his own business, and he has also led major overseas industrial investments for a central state-owned enterprise. What pushed him to enter agricultural robotics wasn’t a business plan, but an absurd reality he witnessed abroad: in Africa, when they bought more than a dozen square kilometers of fertile land to run cultivation trials, they discovered that local people simply didn’t have a habit of planting crops. And training locals to become skilled farmers could take 10 to 20 years.
“Instead of teaching people how to farm, it’s better to do it in one step and let robots do the farming.” That thought ultimately led him to go all in on agricultural robots.
The global AI agriculture market was about $4.7 billion in 2024, and projections say it could swell to $46.6 billion by 2034, with a compound annual growth rate of more than 26%. But the reality behind those numbers is that agriculture may be one of the hardest embodied-intelligence tracks to standardize, and one that depends most on on-the-ground decision-making. Zhao said: “What agriculture needs isn’t machines that move according to lines of code, but intelligent agents that can read farmland, understand crops, and adapt to changes in the environment.”
In 2022, Zhao and his founding team began laying out their plans in agri-tech, carrying out early-stage technology R&D and building up capabilities. GrainCore Dynamics was officially established in November 2025. Three months later, it completed an angel round worth tens of millions of yuan, reaching a valuation of 500 million yuan.
The full transcript of the conversation with Zhao Feng follows, with minor edits:
NextFin News: In a previous interview, you mentioned that you expect global agricultural automation to reach 90% by 2030. How did you arrive at that conclusion?
Zhao Feng: First, the “automation” we’re talking about isn’t automation in the traditional sense; it also includes smart farm machinery with AI-assisted decision-making. Large-scale farms in the world’s major agricultural regions are accelerating the shift toward unmanned operations. China is also vigorously advancing the development of smart-agriculture demonstration zones. In our view, having intelligent equipment cover more than 90% of key farming tasks by 2030 is an achievable goal.
Second, there are still huge disparities in agricultural production models worldwide. U.S. agriculture is already very advanced in automation and intelligence, while parts of Africa are still at the slash-and-burn stage—some places don’t even have established planting habits, let alone an agrarian civilization. In the past, to make use of that land, you first had to train local people to become farmers and teach them how to operate machinery, which could take 10 years, or even 20.
But advances in AI and robotics have been reshaping traditional farming scenarios at a speed beyond what you can imagine. Throughout 2025, breakthroughs in the maturity and cost thresholds of large AI models, computer vision, high-precision sensing, and robotic power systems collectively pushed agricultural robots from “expensive lab showpieces” into “tools that pencil out in the field.” Now, as long as we can rapidly achieve unmanned cultivation, we can skip the step of training locals to do heavy manual labor and truly put Africa’s uncultivated arable land to use.
Agricultural production models in the U.S., China, and other economically developed regions around the world likewise need to move toward unmanned operations—because a farm labor shortage is emerging globally. In China, few young people are willing to engage in traditional agricultural production; in fact, it’s much the same for second-generation farmers in Europe and the U.S. Within the next 5—10 years, the global agricultural workforce may fall off a cliff. If we’re not prepared, we could very well miss this opportunity to overtake on the curve.
That 90% automation rate isn’t really our estimate of technological progress or market growth; given the real state of global agricultural production, it’s the level of intelligence we must reach within the next 5—10 years in order to cope with a globalized food-security crisis.
NextFin News: Compared with industrial settings, what do you think is the biggest difference in agricultural settings?
Zhao Feng: Industrial settings are relatively standardized. The purpose of building a factory is to execute in a standardized way.
Agricultural field operations, however, take place in a far more complex environment with many more variables. Crop height and density, the severity of pests and disease, terrain undulations, lighting conditions, soil moisture—every plot is different, and every season is different. To deploy robots in such a complex environment, the number of parameters that need to be tuned is, in fact, far greater than in a factory setting.
That’s why agricultural robots are considered one of the most difficult scenarios in the embodied intelligence track—they require not machines that simply move according to code, but intelligent agents that can “read” the farmland, understand crops, and adapt to changing conditions.
NextFin News: In agricultural scenarios, what are the main factors customers consider when making decisions?
Zhao Feng: Agriculture is a quintessential cost-sensitive industry—farmers need to see returns on every investment they make.
Specifically: first is the payback period, and generally they’ll only consider it if they can recoup the investment within two years; second is the certainty of operational results—can pesticide drift be controlled, can harvesting losses be reduced, can the job deliver the expected outcome? Ordinary people can see the results clearly; third is product stability and maintenance service, because farming has critical time windows, and once losses occur, they’re irreversible.
In addition, across different regions and different modes of agricultural production, farmers’ considerations can vary significantly.
In China, the single most important consideration is actually yield increase. That’s because agricultural labor is still relatively abundant domestically, so people often don’t factor labor costs into the equation. The most visible cost is land rent, so maximizing yield per unit area is what farmers care about most.
In economically developed countries overseas, it’s quite different—labor costs may be their biggest expense. And with large-scale, mechanized operations using big farm machinery already widespread, boosting yield per unit area is no longer their primary demand. For them, the core objective is comprehensive cost reduction and efficiency gains.
NextFin News: What do you see as the ultimate form of intelligence for agricultural robots?
Zhao Feng: The ultimate form of an agricultural robot isn’t a single machine, but a neural network that uses robots as the execution layer and can reach the entire end-to-end agricultural value chain. From aerial field drones and ground operation robots to underground sensing devices, all interconnected—able to perceive crop growth in real time, issue early warnings about pests and diseases, autonomously devise optimal operating strategies, and execute them with precision—truly becoming an intelligent partner in the field. Once such an autonomous network takes shape, humans will be completely liberated from agricultural production.
VC: We’ve noticed that Hexin’s robots are used in four scenarios—inspection, weeding, crop protection, and harvesting. At the moment, which category of agricultural robots accounts for the largest share of market demand?
Zhao Feng:Right now, most of the demand is in crop protection.
That’s because across the “plowing, planting, management, and harvesting” cycle, mechanization can already solve most problems in plowing, planting, and harvesting. But when it comes to “management,” once the crops have grown up, a lot of farm machinery can no longer get in. Traditional solutions have to rely on manual labor, so the labor required for crop protection is enormous.
Crop-protection robots can adapt to a wide range of operating environments. Drones, in particular, place even fewer demands on terrain and can replace large amounts of manual labor. At the same time, crop protection directly affects both yield and quality, making it a hard, pay-with-real-cash necessity that farmers are willing to spend on. That makes it the best entry point.
Next are weeding robots. Weed control is becoming the biggest pain point under policies aimed at reducing pesticide use. If physical weeding can be made precise, efficient, and free of pesticide residues, the market potential is enormous. Our weeding robot’s recognition accuracy for major weeds has been raised to 98%. Both our orchard crop-protection robot and weeding robot have been included in our product lineup for external display, and we’re pushing forward with real-world deployment.
Inspection robot dogs and indoor inspection drones are incremental markets. While they currently make up a relatively small share, demand is growing quickly in large-scale farms and facility agriculture. In particular, high value-added scenarios such as plant factories and multi-span greenhouses offer considerable room for growth going forward.
Harvesting robots are a longer-term direction. Commercial rollout in the market still faces relatively high costs and technical challenges, but we’re already seeing signs of breakthroughs in high value-added fruit and vegetable scenarios such as mushrooms and tomatoes.
VC Investor: In the agricultural drone space, DJI and XAG already command a very high market share. Where exactly is our differentiated advantage?
Zhao Feng: First, it’s a different technical route. Most mainstream pesticide-spraying drones today use airflow pressure to blow open the leaf surface so the spray can reach the lower leaves. But in reality, many pests aren’t on the leaf surface—most are on the underside of the leaf. GrainCore Dynamics has developed an in-house 50–70 kV “electrostatic adhesion” technology that gives every droplet high-voltage static charge. When those charged droplets penetrate beneath the leaf, they can adhere to the underside. This system is particularly effective for crops with large leaf areas—for example, tobacco leaves.
Second, our product philosophy is different. Mainstream products on the market today are positioned more as efficient automation tools that execute human decisions. What GrainCore Dynamics is building is “embodied intelligence for agriculture,” where AI itself does the understanding, judgment, and decision-making. That way, the drone is no longer just a tool that carries out human instructions, but an intelligent agent that can judge timing on its own, generate an operation plan on its own, and execute it autonomously.
Third, our algorithmic focus is different. Because of the difference in product positioning, mainstream products’ algorithms tend to focus more on obstacle avoidance. GrainCore Dynamics’s algorithm design puts crop growth patterns at the center, focusing more on how to understand agricultural scenarios and generate sound operation plans. In fact, we moved away from the showroom-style, pixel-level pest-and-disease recognition approach, and instead adopted coarse spectral detection plus trend prediction. Combined with crop models from the Chinese Academy of Agricultural Sciences and meteorological data, we compute the optimal operating window. The real value of AI isn’t telling farmers what disease their crops have—it’s telling them “where the problem is, and when dealing with it is most effective and cost-efficient.”
VC Investor: How do you measure the real-world effectiveness of this technology that makes droplets adhere to the underside of leaves? And how hard is it for competitors to copy?
Zhao Feng: We conducted joint experiments with the Institute of Plant Protection at the Chinese Academy of Agricultural Sciences and with China Agricultural University. In real-world operations, this system can reduce pesticide use by 30% while achieving the same level of pest control, and it also shortens operating time—one pass can complete the job, without needing a separate underside-spraying step.
Essentially, what we sell isn’t a piece of hardware, but an integrated “knowledge + algorithms + hardware” system. Behind the solution is a long-term accumulation of data assets and agronomic know-how. Competitors may find it easy to copy the hardware, but replicating the entire methodology requires at least a 3–5 year runway of accumulation—including growth data collected through extensive field trials, a patent strategy, the depth of collaboration with academicians of the Chinese Academy of Engineering and Zhejiang University, and experience covering diverse crop scenarios worldwide.
NextFin News: The mushroom and tomato harvesting scenarios Hexin has chosen are a “standard form” of agricultural industrialization. Does that mean they’re relatively simple scenarios? Will agriculture in the future all evolve into an “industrialized” form?
Zhao Feng: Mushroom and tomato harvesting are relatively standardized, but by no means simple. They’re regarded by investors as the standard form of agricultural industrialization because indoor cultivation of these two crops is already very mature: controllability is extremely high in areas like lighting, temperature, humidity, and row spacing, which reduces the complexity of robotic perception and decision-making. That’s why they’re “top-student scenarios” within agricultural industrialization.
But the picking action itself places extremely demanding requirements on force control, compliant gripping, speed, and success rate, and operational efficiency is still very low. There is still a long way to go before it can be commercialized at scale.
In our view, in areas such as vegetables, fruit, and high-end cash crops, the trend toward industrialization is irreversible, because these segments have the strongest demand for quality, yield, traceability, and sustainability—and consumers are also willing to pay a premium for standardized agricultural products. But for broad-acre crops such as grains and oilseeds, improvements in farm machinery and the penetration of intelligent technologies will move in the direction of lower cost, and may not require fully shifting to greenhouse-style, precisely controlled environments. Every agricultural scenario should evolve in the direction that best suits it, rather than forcing the same “industrialization” template onto all of them.
NextFin News: At present, how is GrainCore Dynamics allocating investment across its product lines—drones, robots, and smart terminals?
Zhao Feng: Our strategy is “heavy on drones + deep focus on robots + smart terminals in reserve.”
Drones are currently the most commercially mature segment, spanning multiple operating scenarios such as crop protection, lifting/hoisting, and inspections, and they have already been delivered and validated in real agricultural environments. Robots, meanwhile, are focused on high-barrier scenarios such as orchard crop protection, weeding, and harvesting—targeting more narrowly defined, must-have labor-replacement needs. R&D investment is on par with that for drones, forming a “two-legged” product-line layout. As for smart terminals, we’re maintaining a follow-up strategy, continually tracking the latest developments in agricultural IoT and edge computing. Investment in this area is relatively cautious, and it mainly serves as foundational capability building.
In terms of product cadence, our near-term priority is to promote a robot product portfolio and a drone lineup that cover the full end-to-end workflow of cultivation, management, and harvesting. They span five major scenarios—crop protection, inspection, picking, transportation, and weeding—running through the entire value chain.
NextFin News: How far along are Hexin’s 2026 order and fundraising plans?
Zhao Feng:So far, progress across the board has been pretty solid. On the fundraising side: we completed the first round three months ago. On the orders side, we already have RMB 200 million in orders and are delivering them step by step.
NextFin News: What is GrainCore Dynamics’s ultimate vision?
Zhao Feng:To free human hands through cutting-edge technology—we’re not just building machines; we’re enabling people to break free from repetitive labor so they can create a better life.
更多精彩内容，关注钛媒体微信号（ID：taimeiti），或者下载钛媒体App
（本文作者为 Chelsea_Sun，钛媒体经授权发布）
NextFin News — The humanoid robotics industry is undergoing a pivotal shift—from “able to walk” to “able to work.” In this transition, all eyes are fixed on one thing: the dexterous hand.
Widely recognized as the component with the highest technical barriers and the most concentrated value in the humanoid robot supply chain, it accounts for 15% to 20% of the total machine cost. The challenge of building it lies in packing, into a space smaller than an adult palm, more than 20 degrees of freedom worth of joints, miniature actuation systems, multimodal sensors, and real-time control algorithms—while also striking a balance among lightweight design, high payload, low cost, and scalable mass production. Globally, only a handful of players can meet all of these requirements at once.
As a result, the dexterous hand is seen within the industry as the “last mile” to commercializing humanoid robots—and it has become the most fiercely contested segment of the current supply chain.
Recently, Xynova, a domestic provider of full-stack general-purpose dexterous-manipulation solutions and one of the “Hangzhou New Eight Champions,” completed an A-round financing of several hundred million yuan. The round was co-led by Li Auto’s strategic investment arm, CSC Capital, and China Securities Co., Ltd. (CSC) Investment, with participation from Yangtze River Delta Digital Culture Group and Yuanjia Fund. Existing shareholders—including Caitong Capital, Xiaomi’s strategic investment arm, and CETC Fund—continued to increase their stakes. Light Source Capital served as the exclusive financial advisor. To date, the company’s cumulative funding has reached nearly RMB 1 billion.
It is understood that the proceeds from this round will be used primarily to build a highly reliable general-purpose dexterous actuation platform, strengthen capabilities for scaled mass production and delivery, and advance a full-stack closed loop spanning “hardware + algorithms + data.” According to the plan, by the end of 2026 the company will officially reach production capacity of 10,000 dexterous hands and 200,000 miniature electric cylinders per year.
Xynova was founded in late 2024. Unlike single-product hardware suppliers in the sector, the company positions itself as a full-stack solution provider for general-purpose dexterous manipulation. Architecturally, it takes the dexterous hand as the core hardware foundation, uses integrated arm-hand coordination as the execution framework, and relies on “cerebellum”-style motion-control algorithms as the intelligent hub—aiming to build a complete closed-loop control system from “perception–control–feedback,” and to provide humanoid robot OEMs, algorithm companies, research clients, and end-scenario customers with a software-and-hardware foundation for dexterous manipulation.
Founder and CEO Xia Yuxuan, born in the post-1995 generation, graduated from a top-tier international university with dual degrees in physics and computer science. He previously worked at Morgan Stanley, CDH Investments, and other institutions, and has long focused on hard tech and high-end manufacturing—spanning autonomous driving and LiDAR to chips and optical modules.
In the span of a year, the team grew from five or six people in its early startup days to more than 300. Its core members have over two decades of technical expertise in high-performance motor systems, motor control, mechatronics, and precision drive technologies. It is understood that most of these R&D and engineering talents come from front-line industry players such as DJI, Schaeffler, KUKA, Apple, FESTO, CATL, UAES (United Automotive Electronic Systems), and leading medical robotics companies. The team also brings together a deep bench of technical talent with backgrounds from universities in China and abroad, including Tsinghua University, Shanghai Jiao Tong University, Zhejiang University, Wuhan University, and the University of Pennsylvania, Columbia University, and the Technical University of Munich. Collectively, they have the engineering experience to systematically transfer capabilities in motors and motor control, mechanical transmission, structures, and algorithms into the embodied intelligence domain.
At present, dexterous-hand design largely revolves around three major technical approaches: direct-drive motors, tendon-cable drive (cable drive), and linkage drive, while some manufacturers are also exploring hybrid drive solutions.
Xynova Future has opted for a hybrid drive solution centered on cable drive. In August last year, it launched the world’s first fully in-house developed, mass-producible, high-DoF tendon-driven dexterous hand, Xynova Flex 1. Xynova Flex 1 offers 25 degrees of freedom, with a palm weighing just 380 grams, a payload capacity of over 30 kilograms, and fingertip force exceeding 20N per finger. Xynova Future said it is currently the lightest and highest-payload high-DoF dexterous hand on the market.
In mid-May, building on Flex 1’s arm-hand integration advantage, it introduced the world’s first biomimetic dexterous hand using a “tendon-cable + direct-drive motor” hybrid drive, Flex 2, with a palm weighing under 400g and 23 degrees of freedom (19 active and 4 passive). Its fingertip force exceeds 20N, peak one-handed grasp payload reaches up to 12kg, and its rated payload for continuous operation is 4kg. At the same time, Flex 2 integrates four types of sensors—vision, touch, force, and proximity—and, paired with a humanoid “cerebellum-like algorithm,” can perform more complex and intelligent hand operations such as adaptive grasping and backdrivability under load.
According to its roadmap, Xynova Future will further expand its product portfolio in 2026, with plans to roll out the second-generation Flex series dexterous hands featuring systematic upgrades to sensing capabilities, miniature actuators, and transmission performance.
Scaling up to true mass production of dexterous hands remains a global challenge to this day. The key bottlenecks include the supply-chain maturity of miniature coreless DC motors and gear reducers, the long-term durability of tendon-cable materials, the stability of multi-DoF coordinated control algorithms, and the remaining room to bring down the cost of multimodal sensors.
Looking at the industry landscape, even Tesla’s Optimus has seen dexterous-hand mass-production progress lag significantly behind the robot body itself. Most domestic dexterous-hand makers are still in a “small-batch, customized” stage, with annual deliveries typically in the hundreds of units.
Whether Xinuo Future can genuinely ramp up to a 10,000-unit-level production capacity by year-end will be the key litmus test of how real its positioning as a “full-stack solution provider” truly is. After all, in hard tech, the amount raised is only the starting line—mass-production capability and customer word of mouth are what ultimately decide the outcome.
更多精彩内容，关注钛媒体微信号（ID：taimeiti），或者下载钛媒体App
（本文作者为 公司观察，钛媒体经授权发布）
图片系AI生成
5月以来，“西北乳业第一股”庄园牧场（002910.SZ）连发“复杂信号”：
昨日晚间公司公告，持股5%以上股东、董事、公司创始人及前实控人马红富，拟在2026年6月23日至9月22日期间，通过集中竞价和大宗交易方式，减持不超过2,242,400股，占公司总股本不超过1.16%，按最新约10.06元/股估算，套现金额约在2,255万元左右。
同时，庄园牧场对马红富发起的“搬迁损失兜底承诺”诉讼也将进入二审关键阶段：甘肃省兰州市中级人民法院已送达传票，定于2026年6月16日开庭，涉案金额达2,107.38万元。
此刻的马红富，质押比例高达91.5%，还拖欠着出售控股权时的业绩补偿款1360万元。与此同时，自2021年国资接盘以来，庄园牧场连亏三年，马红富也拖欠着上千万的业绩补偿款。
上述诉讼围绕“知情权”博弈的复杂性，亦已引发市场广泛关注：招股书白纸黑字承诺牧场搬迁损失由政府补偿，不足部分由其兜底，等牧场真被关停、政府只赔了505万、公司损失超2500万时，马红富的理由是招股书的法律效力仅限于“面向证券市场、监管机构及不特定投资者的信息披露内容，目的是保障投资者知情权”，不具备强制执行的法律效力。一审法院也以招股书属“要约邀请”为由，判庄园牧场败诉。
前老板忙着解冻、减持、拖欠对赌款，上市公司忙着对簿公堂、深陷赊销与亏损，这一出西北乳业大戏即将迎来白热化。
截至28日，马红富持有庄园牧场股份2639.05万股，占总股本的13.50%。本次拟减持1.16%，股份来源为其首次公开发行股票并上市前持有的股份，拟自2026年6月23日至2026年9月22日期间，其中，集中竞价交易193.58万股，占总股本的1.00%；以大宗交易减持不超过30.66万股，占总股本的0.16%，减持原因标注为自身资金需求。
图源：公司公告
公告云淡风轻，但背后的马红富早已“火烧连营”：自2025年中开始，马红富所持的庄园牧场全部股权已被法院冻结，约占公司总股本的16.47%。马红富已向法院提出执行管辖权异议，并计划向法院申请不予执行。
5月23日公司公告，马红富有217.29万股解禁，虽然剩余的质押比例依旧高达91.5%，但这部分筹码足以让他开启减持套现。
从2026年1月首次启动减持以来，马红富已经累计减持公司股份580.69万股，合计套现约6469.82万元。若此次减持计划全部完成，2026年上半年马红富的累计套现金额将接近8700万元。
质押的源头颇复杂，根据庄园牧场此前公告，2024年6月27日至2027年6月26日，马红富因个人资金需求，向兰州银行股份有限公司雁欣支行合计质押2187万股兰州庄园股份，约占公司总股本的11.18%，质押用途为个人资金需要。
公告显示，司法冻结涉及的案件债权额及执行费用合计约2126.94万元。
而深交所2025年5月下发的监管函，揭开了马红富的另一桩窟窿。
2021年1月，庄园牧场实控人马红富和自然人胡开盛，拟将所持庄园投资100%股权转让给甘肃农垦集团，交易对价4.27亿元，成功抽身。
但易主以来，庄园牧场深陷亏损，马红富未足额支付业绩补偿款（尚欠1,360.55万元）。
图源：公司公告
在庄园牧场2025年报中，公司当期收到的业绩补偿款数额仍为0元，上期发生额670万元。
更富戏剧性的是，庄园牧场与马红富之间一起围绕IPO承诺展开的2100万元追偿诉讼，即将进入二审决战。
2017年，庄园牧场冲刺上市，招股说明书控股股东马红富承诺：若下属牧场因政府划定禁养区而遭搬迁，政府补偿不足以弥补公司损失时，差额部分由其承担。
图源：公司公告
这原本是给市场吃的一颗定心丸，但2019年，湟源县人民政府一纸限期关停通知，直接让圣源牧场清栏关停，建筑物构筑物全部报废。此后围绕补偿金额打了几年官司，最终青海省高院判决，政府只补偿505.49万元。而圣源牧场测算关闭搬迁损失高达2505.34万元，中间将近2000万元的缺口，便直指马红富当年的“兜底承诺”。
然而，兰州市城关区法院的一审判决出乎市场意料：法院认为，招股说明书系发行人向社会公众投资者披露信息、公开发行股票的法定文件，性质上属于“要约邀请”，而不是发行人与控股股东之间设立权利义务的合同。马红富当年的承诺，面向的是证券市场和不特定投资者，旨在保障投资者知情权，并非和上市公司达成合意，且双方从未就此签订任何书面补偿协议。
据此，法院驳回了庄园牧场的全部诉讼请求。庄园牧场自然不服，提起上诉。甘肃省兰州市中级人民法院已送达传票，定于2026年6月16日开庭，正好与马红富的减持计划无缝衔接。
自2021年甘肃农垦集团接盘以来，市场对“国资加持”充满期待，然而庄园牧场近几年交出的财报，让“西北乳业第一股”备受争议。
根据2025年年报和2026年一季报，庄园牧场仍陷在亏损泥潭中挣扎。2025年全年实现营业收入约9.21亿元，同比微增3.44%；归母净利润-7438.87万元，虽然亏损较上年收窄55%左右，但依旧大额亏损。
2023至2025年，公司已连续三年亏损，累计亏损金额高达数亿元。2026年一季度，营业收入约2.46亿元，同比增长16.72%，看似“开门红”，但归母净利润-3150.84万元，较去年同期亏损-2595万元反而扩大21.42%，属于“越卖越亏”。
增收不增利的背后是经营质量的全线下滑。公司资产负债率从2023年的49.99%一路攀升至2025年的55.82%，到2026年一季度末已达56.91%，财务杠杆越加越高。
应收账款层面，截至2026年一季度末，应收账款余额飙升至约9691万元，同比激增108.44%，而同期营收增速仅为16.72%。这意味公司销售增长，是靠大规模赊销和放宽账期堆出来的。回款和坏账风险暗流涌动，反映出公司虽然自称“深耕西北”，但在产业链中议价能力薄弱，面对区域性乳企的内卷挤压，不得不通过“放水”来撑住收入门面。
年报显示，庄园牧场近年在不断试图通过文创概念、差异化产品撕开突破口，“老兰州”酸奶、花椒酸奶等文创新品以及富硒鲜奶、A2奶等高端产品在不断迭代创新，但从财务表现看，这些创新尚未形成可观的盈利贡献。
毕竟，费用压缩和库存管控虽使部分成本项有所下降，但节流的空间已十分有限，而开源又受制于伊利、蒙牛等全国性品牌的强势下沉，以及西北消费升级缓慢、人口流出的结构性困局。
区域乳企的宿命，在庄园牧场身上得到了完全体现——品牌溢价难以树立，也无力支撑价格战，铺开渠道反而成了失血通道。
更让局面复杂的是，虽然国资背景的控股股东想在奶源和融资上给予稳定支持，但另一边，前实控人马红富依然身在董事会，既是大额减持的当事人，又是对赌补偿的义务人，还是诉讼的对立方。
图源：公司公告
并且由马红富从内部一路被提拔起来的亲侄子马刚，亦担任庄园牧场副总经理、董事。这种“你中有我、我中带刺”的治理结构，使得庄园牧场在战略决策和公司信心层面，都潜藏着摩擦成本与冲突风险。（文 | 公司观察，作者 | 黄田，编辑 | 曹晟源）
更多精彩内容，关注钛媒体微信号（ID：taimeiti），或者下载钛媒体App