文 | 市值观察
不久前，蓝思科技发出公告，拟以2500万美元（约合人民币1.72亿元）现金，收购萨摩亚离岸主体PMG（裴美高国际）59.78%的股权，间接控制元拾科技，进一步加码服务器机柜及液冷业务。
其实不只是蓝思，立讯精密、歌尔股份、工业富联等果链企业也在通过不同路径扩大AI相关业务。过去被市场担心过度依赖消费电子的果链企业，如今却成了AI硬件产业争夺的重要制造力量。
AI时代，能不能让这批果链上的企业再做大一次？
集体转型
曾经，果链企业能快速成长的市场前提是，消费电子（尤其是智能手机）仍处在高速渗透和频繁换机的阶段。但如今，这个前提已经变了。
据Omdia统计，2025年全球智能手机出货量约12.5亿部，同比增长2%。市场仍在增长，但已经很难仅靠行业普及，复制早年的扩张速度。
市场池子不再快速膨胀，对客户集中度较高的果链企业而言，一旦大客户调整产品节奏、压价或分散供应商，企业的业绩波动就会很大。
因此，守住在苹果体系内的份额，越来越像存量博弈。于是我们能看到，这些年但凡有能力的果链企业，也在寻找不受单一终端市场限制的新生意。
比如立讯精密，最早给苹果做的是连接器和内部线缆，如今其产品已延伸至声学整机、汽车电子以及通信和数据中心设备。连接器业务中积累的精密加工和信号传输经验，也有了更多应用场景。
汽车电子已成为立讯精密营收增长最快的板块之一。在收购整合德国莱尼后，立讯2025年汽车电子收入392.55亿元，同比增长185.34%，营收占比从5.12%拉到11.81%。
而工业富联，走的并不是从手机跨界服务器的路。
早在2018年，公司云服务设备业务收入就达到1532.24亿元，占主营业务收入的37.11%，产品包括服务器、存储设备及相关精密机构件。
因此，当AI算力需求爆发，工业富联不需要重新进入一个陌生市场，而是把已有的研发、制造和交付能力，用到需求增长更快的新一代服务器上。
据财报显示，2025年工业富联云计算板块营收6026.79亿元，同比增长88.70%；云服务商AI服务器营收同比增长超过3倍，800G以上高速交换机收入同比增长13倍。
相比之下，歌尔股份更接近沿着零组件和整机两条线，等待新终端市场放量。其声学、光学、微电子及精密结构件产品，可以用于手机，也可以用于AI眼镜、混合现实等设备。
2025年，公司智能硬件板块营收537.69亿元，同比下降6.00%，但该板块毛利率同比提升2.2个百分点至11.37%。虽然不能把整体毛利率的变化全部归因于AI眼镜，但它也说明，新产品的发展与原有业务的调整正在同时发生。
这三家企业的转型方向和路径虽然看起来各不一样，但底色是相通的。
它们长期投入研发、改进工艺、建设产线，积累下来的能力，并不只属于某一款消费电子产品。汽车、服务器、智能眼镜对制造的要求不同，但其中一部分技术与生产经验可以复用。
所以，果链企业的转型命题不是抛弃过去、改弦更张，是利用已投入多年、形成优势的那些本事，支撑更新更大的生意。
AI时代
当下AI发展带来的需求很杂，看起来处处是机会。但机会多，不等于企业一定能做大。
因此，真正该看的是两件事：这些新机会里，企业能拿到多少值钱的部分；这些部分是不是正好用得上它们原来的手艺。
比如，最先跑出规模的算力基础设施。
这个方向对果链企业的意义不是服务器市场和手机市场哪个更大，而是一个机柜里能装进去的零件，比一部手机贵太多了。
据摩根士丹利测算，英伟达GB300 NVL72一套液冷散热组件就值约5万美元，折合人民币36万元，比上一代贵两成；到下一代Vera Rubin，单柜冷却组件的价值还要往上走。手机中框、玻璃盖板那些和这比完全不是一个量级。
所以，立讯精密的液冷方案在2025年底已从前瞻布局转入规模化商用，800G硅光模块量产，AI服务器电源和液冷散热进一步放量；蓝思这边，松山湖园区完成了扩产和客户验证，AI服务器结构件和液冷模组的产能正在爬坡阶段，还处在从验证走向批量交付的过渡期。
对果链企业来说，液冷不只是又多了一个可以涉猎的品类，而是做了多年的精密加工、管路连接和散热，这次换道算力硬件上，能卖出更高的价钱。
而受到热议的AI眼镜，正从小批量尝鲜转向大规模出货。
据维深信息Wellsenn XR数据，全球AI眼镜销量从2024年的153万台跳到2025年的846万台，2026年预计达到1600万台，2030年的预测是9000万台，对应2025到2030年约60%的复合增速。
品类增长很快，但如果只做整机组装，净利率大概是3%到5%，眼镜卖得再多，落到利润里也有限。
所以歌尔股份不只图那点组装费，更多的是提供声学模组、精密结构件和光学元件，这些零件占一副AI眼镜成本近三成，毛利率比组装高得多。现在歌尔不仅是Meta Ray-Ban系列核心整机代工厂之一，还拿到了华为和小米AI眼镜的大部分代工份额。
在更受追捧的机器人这块，高盛把2030年人形机器人出货量预测上调到了89万台，2035年650万台，对应约1380亿美元的市场机会，但2026年的基准预测只有7.5万台，市场还处于早期发育阶段。
并且机器人对精密执行器、热管理和轻量化结构的要求，虽然落在果链企业的能力范围内，但能力匹配和业务兑现之间，隔着产品定型、客户认证和量产爬坡好几道关口。
于是我们能看到，蓝思科技2025年人形机器人整机交付超过3000台，四足机器人突破一万台，2026年进入了头部客户的量产爬坡，但它在总营收里的分量仍然很轻；歌尔股份在机器人上的布局也更多停留在传感器和执行器能力的延伸上，远未到规模化的阶段。
在机器人上落子的果链企业并不少，但绝大多数还处在送样、验证和产线试制的阶段，距离成为业绩支柱还有不短的距离。
所以和消费电子时代不同，AI时代给果链企业带来的最大价值不在于市场空间更大了，在于需求的层次更多了。这给了果链企业将同一套精密制造手艺反复变现的机遇。
算力硬件、AI眼镜和机器人，给果链企业提供的是几种不同的机会：有的已经形成大规模需求，有的正在验证消费市场，有的还需要等待产品和应用成熟。
而决定企业能否走得更远的关键，是如何分辨出哪些需求今天就能接住，哪些则需要提前下注。
谁能成更大巨头？
果链企业如何在AI时代长成更大的公司？答案不会只有一种。从现有的企业布局，至少可以看到三种不同的成长路径。
第一种是业务变大，从依赖单一客户，到服务更多产业。
以东山精密为例，作为苹果柔性电路板（FPC）的核心供应商，它过去最被市场担心的地方就是和苹果绑得太深，苹果的出货节奏几乎决定了它的增长节奏。
但2025年，东山精密最大客户收入占比从2023年的55.8%降至46.5%，前五大客户从73.5%降至64.4%。它把在消费电子FPC和精密制造上积累的能力，延伸到了车载FPC、结构件和车载显示，用同一套制造底座服务另一条产业线。
尽管目前东山精密的汽车相关收入规模仍远小于消费电子，但客户名单每多一个名字，它的天花板就往上抬一截，并且让更多人意识到它不只是一家果链厂商。
第二种是价值变大，即抓住产业升级，扩大原有业务的价值。
前面提到过的工业富联，走的就是这条路。它原本就具备服务器整机制造能力，AI一来，把竞争门槛从单部件抬到了系统级。
比如一台GB200 NVL72要集成72颗GPU、36颗CPU，液冷、电源、高速互联、光纤和整柜联调必须一起交付。这让单笔订单的复杂度和价值量都被抬高了，也让工业富联攒了多年的能力有了用武之地。
据悉，工业富联不仅能实现交付整柜系统，还和英伟达一起把GB200的工程化落地啃了下来，成了全球少数能同时交出“CPO交换机+液冷机柜+AI服务器”的厂商。
2025年，工业富联云计算板块的毛利率提升了0.74个百分点、AI服务器营收增长超3倍。正是这种系统级能力开始兑现的信号。所以制造企业不必先变成高毛利公司，也能通过更大规模、更复杂的交付，把原有优势做成更大的生意。
第三种是利润变大，依靠核心技术，进入利润空间更大的市场。
领益智造做了多年精密功能件和散热方案，这套能力放到AI服务器液冷、AI终端上，底层工艺相通，但毛利却不同。
2025年领益智造的营收为514.29亿元，同比增长16.20%；归母净利润22.88亿元，同比增长30.34%。利润跑得比收入快，公司自己的归因排在第一位的是AI硬件高附加值产品投入和份额提升。它的AI硬件业务收入约447.93亿元，毛利率17.21%，高于公司整体的15.80%。
值得一提的是，今年年初，领益智造通过收购立敏达，接入了英伟达GB200、GB300的液冷散热器和快接头供应链。做的还是精密功能件和散热，但却是奔着利润更高的液冷组件去的。
当同一套手艺，换一个市场，回报完全不同时，这考验的是企业能不能辨别出自己手里的技术在哪个市场里更值钱，并且愿意为了更高的利润去换一个战场。
这三条路，是三种不同的判断和取舍。选哪一条，或者在哪一条上走得最远，是果链企业接下来真正的分野。
总之，AI硬件越往深处走，对精密制造、热管理、光学和规模交付的要求就越苛刻，而这些恰好是果链企业打磨了很多年的能力。AI时代，这批企业要争的，已不只是果链中的供应份额，而是下一批世界级制造巨头的席位。
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。
Market value observations
Not long ago, Lansytech announced the acquisition of 59.78 per cent of the shares of the offshore Samoan major PMG (Pemega International) in cash of $25 million (approximately RMB 172 million), indirect control over metallurgical technology, further enhancement of server cabinets and liquid cooling operations.
In fact, not only Lance, but also fruit chain enterprises such as the sophisticated information, Gole shares, and the Industrial Riches are expanding AI-related operations through different routes. In the past, fruit chains, which were feared by the market to be over-dependent on electronics, are now an important manufacturing force for the AI hardware industry.
In the AI era, can we get this chain business to do it again?
Collective transformation
In the past, the market for fast-growing fruit chain enterprises was premised on the fact that consumption of electronics (especially smart phones) was still in the phase of high-speed penetration and frequent switchover. But now that premise has changed.
According to Omdia, global smartphone deliveries in 2025 were approximately 1.25 billion, an increase of 2 per cent over the same period. Markets are still growing, but it is difficult to replicate the pace of early expansion on the basis of the expansion of industries alone.
Market pools are no longer expanding rapidly and, for fruit chain enterprises with high customer concentration, their performance will fluctuate significantly once large customers adjust product rhythms, press prices or spread suppliers.
As a result, the share of the apple system is becoming more and more like a stock game. So we can see that all these years of capable fruit chain businesses are looking for new businesses that are not restricted by a single terminal market.
For example, with sophisticated information, apples were first made of connectors and internal cables, and their products are now extended to acoustic complexes, automobile electronics and communication and data centre equipment. The experience of precision processing and signal transmission accumulated in connector operations has also generated more applications.
Automobile electronics have become one of the fastest growing segments of the intelligence complex. Following the acquisition and consolidation of Leni, Germany, the electronics of automobiles generated $39.255 billion in 2025, an increase of 185.34 per cent over the same period, while the share of receipts rose from 5.12 per cent to 11.81 per cent.
And the Federation of Industrials is not going through cross-border servers from mobile phones.
As early as 2018, corporate cloud service equipment earned $15,324 million in operating income, or 37.11 per cent of the main camp ' s operating income, including servers, storage equipment and related sophisticated institutional components.
As a result, when the demand for al-calculations erupted, the Federation did not need to re-enter a strange market, but instead to use existing R & D, manufacturing and delivery capabilities on a new generation of servers with faster demand growth.
According to the financial report, in 2025 the IUFC cloud computing board earned $60,679 million, representing an increase of 88.70 per cent over the same period; the cloud server AI received more than three times the same amount and earned more than 800 G high-speed switches 13 times the same amount.
By contrast, Gole shares are closer to the zero-assembly and machine-building lines, waiting for new terminal market discharges. Its acoustic, optical, microelectronic and precision structural components products can be used on mobile phones, or for equipment such as AI glasses, hybrid reality etc.
In 2025, the company received $53,769 million from the smart hardware block, which represented a decrease of 6.00 per cent over the same period, but the gross domestic product of the plate increased by 2.2 percentage points to 11.37 per cent over the same period. While the change in the overall Māori ratio cannot be attributed entirely to AI glasses, it also indicates that the development of new products is occurring at the same time as the adjustment of the original business.
The transformation directions and paths of these three enterprises, although different in their appearance, are consistent in the background.
Their long-term investment in research and development, process improvement, and production line build-up capacity does not belong to a single consumer electronic product. Motor vehicles, servers, smart glasses have different manufacturing requirements, but some of these technologies and production experience can be reused.
Thus, the transformational proposition of fruit chain enterprises is not to abandon the past, to change the string, to use the skills that have been invested for many years to create advantages to support the renewal of larger businesses.
The AI Age
The demands of the current development of AI are diverse and appear to be opportunities everywhere. But there are more opportunities, and it's not the same thing that businesses can do.
So what really needs to be seen is two things: how much of these new opportunities are worth of money that companies can get; and whether they are just what they used to be.
For example, the first large-scale computing infrastructure.
This direction does not mean much more to the chain business than to the server market and the mobile phone market, but rather to the parts that can be installed in a cabinet, which is far more expensive than a mobile phone.
According to Morgan Stanley, a set of GB300 NVL72 liquid cold-dispersed components is worth approximately $50,000, equivalent to RMB 360,000, 20 per cent more than the previous generation; by the next generation of Vera Rubin, the value of a single-box cooling component will go up. Cell phone frames, glass lids, and that's not a whole scale.
As a result, by the end of 2025, the sophisticated liquid cooling programme had moved forward to scale commercialization, 800 G silica module production, further discharge of the AIS server power supply and liquid cold dissipation heat; on the Lansy side, expansion and customer validation were completed at the Lake Matsuyama Park, and the production of the AI server structure and liquid cooling module units was in the slope phase and was in the transition from validation to bulk delivery.
For fruit chain enterprises, liquid cooling is not just an additional class of items that can be hunted, but has been the subject of many years of precision processing, piped connections and heat dissipation, which in this case can be sold at higher prices on swap-calculation hardware.
The acclaimed AI glasses are shifting from small quantities to large deliveries.
According to Wellsen XR data, global sales of AI glasses jumped from 1.53 million in 2024 to 8.46 million in 2025, are projected to reach 16 million in 2026 and 90 million in 2030, corresponding to a compound rate of growth of about 60 per cent between 2025 and 2030.
The product has grown rapidly, but the net interest rate is about 3 to 5 per cent if it is fully assembled, and the price of the glasses has been too high and limited to profit.
So the Gole shares are more than just the assembly fees, they are more the provision of acoustic modules, precision structures and optical components, which make up nearly 30 per cent of the cost of an AI pair of glasses and have a much higher Māori rate than assembly. Now Gole is not only one of the whole machine factories of the Meta Ray-Ban series, but also receives most of the surrogate shares of China and Miai glasses.
In the more popular robotic sector, Goldman Sachs moved up the 2030 human robotic delivery forecast to 890,000, 6.5 million in 2035, corresponding to market opportunities of about $138 billion, but the 2026 benchmark forecast is only 75,000, and the market is in the early stages of development.
And robotic requirements for precision implementers, heat management and light quantization structures, while falling within the capacity of fruit chain enterprises, lie between capacity matching and operational realization, at several points between product stereotyping, customer certification and mass production slopes.
So we can see that Lenth Tech delivered more than 3,000 man-shaped robots in 2025, that four-foot robots went over 10,000, that in 2026 it reached the mass-carrying slope of the head client, but that it still received a very small amount of money in the general base; and that the configuration of the Gole shares on the robot is more confined to the extension of the sensor and the implementer's capabilities, far from being in the scale phase.
There are not many fruit chain enterprises that have fallen on robotics, but the vast majority are still in the delivery, validation and line trial stages, not too short away from becoming a performance pillar.
So unlike the consumer electronics age, the greatest value of the AI era for fruit chain enterprises is not the greater market space, but the greater level of demand. This gives chain enterprises the opportunity to repeat the same sophisticated manufacturing skills.
Computation hardware, AIS glasses and robots offer several different opportunities to fruit chain enterprises: some have developed large-scale demand, others are testing consumer markets and others are waiting for products and applications to mature.
The key to determining whether an enterprise can go any further is how to identify which needs can be met today and which require an advance bet.
Who's gonna be bigger?
How did a chain business grow into a bigger company in the AI era? There is no one answer. There are at least three different paths to growth that can be seen in the existing enterprise configuration.
The first is a larger business, ranging from relying on a single customer to serving more industries.
For example, as a central supplier of the Apple Soft Circuit Board (FPC), it was most feared by the market that the apples had been tied too deeply, and the speed of their export had almost determined its growth.
In 2025, however, the share of the largest client income of the top five clients fell from 55.8 per cent in 2023 to 46.5 per cent and from 73.5 per cent to 64.4 per cent. It extends the capabilities accumulated in the consumer electronics and precision manufacturing to the vehicle-borne FPC, structural components and vehicle-borne displays, with the same manufacturing base serving another industrial line.
Although far smaller than consumer electronics are currently the most sophisticated vehicle-related revenues, every additional name on the customer list lifts up its ceiling and makes more people aware that it is not just a fruit chain manufacturer.
The second is the increase in value, which captures the upgrading of industries and expands the value of old operations.
That's the way to go. It had the capacity to complete the server system, and AI had raised the competitive threshold from the single component to the system level.
For example, a GB200 NVL72 is required to integrate 72 GPUs, 36 CPUs, liquid cooling, power supply, high-speed interconnection, fibre optics and whole cabinet connection must be delivered together. This has raised the complexity and value of individual orders and enabled the Federation to build up its capacity for many years.
It has been learned that not only will the Federation be able to deliver the whole container system, but it has also gnawed the GB200 project down with In Weida, becoming a small global producer who can hand over the `CPO switch + liquid cooler + AI server' simultaneously.
In 2025, the Māori rate of the ICF cloud computing plate increased by 0.74 percentage points and the collection of AI servers increased more than threefold. It is this system-level capability that is beginning to materialize. Thus, manufacturing firms do not have to become high-māori companies first, and can do larger business with larger and more complex deliveries.
The third is to increase profits, relying on core technologies and entering larger markets in profit margins.
For many years, it has developed sophisticated functional components and dissipation programmes, which are placed on the AIS server liquid cooling, the AI terminal, and the bottom process is common, but Māori is different.
In 2025, the income from the use of intellectual engineering was 51,429 million yuan, an increase of 16.20 per cent over the same period; the net profit of the mother was 2,288 million yuan, an increase of 30.34 per cent over the same period. Profit runs faster than income, and the company ' s own attribution ranks first with higher value added inputs and shares of AI hardware. Its revenue from AI hardware operations is approximately $44,793 million, with a Māori rate of 17.21 per cent, which is higher than the company as a whole of 15.80 per cent.
It is worth mentioning that, at the beginning of this year, Ludwig acquired the Lyminda supply chain with access to the British GB200, GB300 liquid chillers and fast-links. It is also made of precision functional units and heat dispersion, but it is driven by more profitable liquid cold components.
When the same kind of craftsmanship, in exchange for a market, offers a completely different return, the test is whether the enterprise can identify the market in which the technology in its hands is more valuable and is willing to trade for higher profits.
These three paths are three different judgments and trade-offs. Which one, or which one, goes the farthest, is the real separation of the chain business.
In sum, the deeper the AI hardware moves, the more demanding it is for sophisticated manufacturing, heat management, optical and scale delivery, which happens to be the ability of a chain enterprise to grind for many years. In the AI era, this group of companies is fighting not only for the share of supply in the chain, but for the next class of world-class giants.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.