CITIC Securities: Current Global AI Computing Infrastructure Order Visibility Remains High, Maintaining Optimism on AI Industry Chain 中信建投：当前全球AI算力基建的订单能见度仍比较高，依旧看好AI产业链

In an interesting contrast to this macro optimism is the awkward term found in the Huanghe Industrial announcement: "Agentic AI." This seems like a microcosm of the current AI industry: concepts always outpace implementation, and vocabulary always grows faster than problem-solving capabilities. Agentic AI (intelligent agent AI) is widely recognized as the next frontier, but Huanghe Industrial, a company known for industrial and tech investments, is delving into "potential investments" so quickly? The disclaimers in the announcement, such as "no agreement has yet been entered into" and "may or may not materialize," read like a standard, cautious attempt to ride the trend in capital markets. This isn’t necessarily a bad thing, but it raises the question: is Agentic AI just another "hype cycle" poised to be over-financialized and conceptualized? When a technical concept starts appearing frequently in listed companies' announcements about "potential investments," it often signals that it has moved beyond pure technical discussion and entered the realm of capital narrative. And the hallmark of capital narratives is their ability to obscure specific challenges with vague visions.

Placing these two pieces of information together perfectly outlines the fragmented reality of the current AI industry: on one side, the macro level (represented by brokerage reports) continues its triumphant march,坚信 that computing power is hard currency and the industry chain will continue to benefit; on the other side, the micro level (represented by corporate announcements) is rapidly chasing cutting-edge concepts, filled with trial, uncertainty, and potential risks of a bubble. News on hot lists also corroborates this divide. "Chat is Dead, GPT's Biggest Revision Ever" heralds the inevitable evolution of interaction forms, while "DingTalk VP Resigns" and "After Adopting AI, the Company Stops Hiring" reveal the real impact of AI on existing organizational structures and workplace ecosystems. An AI mathematician "posing problems in the morning and handing in proofs in the afternoon" is stunning, but does this prove AI's potential for general intelligence, or is it limited to excelling in closed, rule-bound mathematical domains? Progress and impact, vision and growing pains are forever intertwined.

Returning to the judgment of "high order visibility." It is not wrong in itself, but overemphasizing it causes one to overlook a more critical question: How efficiently can these orders be converted into profits and moats? Chip manufacturers are making a fortune, but what about downstream application vendors? How many are still running at a continuous loss? The boom in computing power infrastructure ultimately needs to be paid for by the commercial success of the applications built on top. If the application layer cannot sustain a viable profitability model for a long time, the current high "visibility" could evolve into future mountains of inventory and overcapacity. We have seen too many cases of technical infrastructure outpacing application adoption, from fiber networks to renewable energy capacity. History does not repeat itself simply, but it often rhymes.

Therefore, regarding CICC's "optimism," my view is: agree with caution. The trend of AI is irreversible, and the demand for computing power will continue to grow for a considerable period. But being optimistic about the "entire industry chain" is a correct but overly broad platitude. The real opportunities lie in segments that deeply integrate AI capabilities with specific industry knowledge to solve real-world problems, rather than broadly investing in "any concept related to AI." For moves like those of Huanghe Industrial, my view is: closely monitor its specific technical routes and business models, but maintain a healthy dose of skepticism toward any unimplemented "Agentic AI" investment story. The real intelligent agent revolution will not be born from an investment letter of intent announcement, but perhaps in a garage team's complete restructuring of workflows, or an engineer's relentless pursuit of agent reliability.

The AI industry is undergoing a difficult transition from a "phase of technological awe" to a "phase of value realization." During this process, grand narratives remain appealing, but what will truly determine victory or defeat are the tedious, specific details concerning efficiency, costs, and implementation outcomes. High visibility is a good thing, but don't forget, when the night is too deep, the distant lights can also easily become a mirage.

中信建投的分析师们又一次把“看好AI产业链”的结论摆在了台面上，理由是“订单能见度高”。这话听着耳熟，像极了2023年喊“AI是工业革命”、2024年说“算力是新石油”的同一批嗓音。订单能见度高当然不是坏事，但它究竟是真实繁荣的映射，还是下游客户在AI热潮下的恐慌性下单？当所有云厂商、车企、甚至传统药企都在抢购GPU时，这种“能见度”有多少是基于真实业务需求，又有多少是生怕落后于同行的焦虑驱动？真正的产业革命，驱动因素从来不是“怕错过”，而是“算得过来账”。目前除了头部互联网公司的核心业务（推荐、搜索、广告）确实被AI显著提效外，大量中小企业的AI应用，要么停留在成本高昂的“PPT场景”，要么就是调用API做些边缘创新。订单很热闹，但利润和真正的生产关系变革，还躲在云雾之中。

与这种宏观的乐观形成有趣对照的，是黄河实业公告里那个拗口的词：“自主智能式人工智慧（Agentic AI）”。这像极了当前AI产业的一个缩影：概念永远跑在落地前面，词汇量永远比解决问题的能力增长得快。Agentic AI（智能体AI）是公认的下一个前沿，但黄河实业，一家以实业和科技投资闻名的公司，这么快就深入讨论“潜在投资”了？公告里“尚未订立任何协议”、“可能会或不会落实”的免责声明，读起来像极了资本市场对热点标准的、谨慎的蹭热操作。这未必是坏事，但足以让人嘀咕：Agentic AI是不是又一个即将被过度金融化、概念化的“风口”？当一个技术概念开始频繁出现在上市公司的“潜在投资”公告里时，往往意味着它已经越过了纯粹的技术探讨阶段，进入了资本叙事的范畴。而资本叙事的特点，就是善于用模糊的愿景掩盖具体的挑战。

这两条资讯放在一起，恰好勾勒出当下AI产业的分裂现实：一边是宏观层面（以券商报告为代表）的持续高歌猛进，坚信算力是硬通货，产业链将持续受益；另一边是微观层面（以公司公告为代表）对前沿概念的快速追逐，充满了试探、不确定性和潜在的泡沫风险。热榜上的新闻也佐证了这种分裂。“Chat已死，GPT史上最大改版”宣告了交互形态的必然进化，但“钉钉副总裁离职”、“用上AI后，公司停止招人了”则揭示了AI对现有组织结构和职场生态的真实冲击。AI数学家“上午出题，下午交证明”震撼人心，但这究竟是证明了AI的通用智能潜力，还是仅限于在封闭、规则明确的数学领域大放异彩？进步与冲击、愿景与阵痛始终交织。

回到“订单能见度高”这个判断。它本身没有错，但过于强调这点，会让人忽略更关键的问题：这些订单转化为利润和护城河的效率如何？芯片厂商赚得盆满钵满，但下游的应用厂商呢？有多少在持续亏损？算力基建的繁荣，最终需要由上层应用的商业成功来买单。如果应用层迟迟无法跑通可持续的盈利模型，那么当前高企的“能见度”就可能演变成未来堆积如山的库存和过剩产能。我们见过太多次技术基建超前于应用普及的案例，从光纤网络到新能源产能。历史不会简单重复，但总是押着相似的韵脚。

所以，对于中信建投的“看好”，我的看法是：谨慎地同意。AI的趋势不可逆转，算力需求在相当长一段时间内仍会增长。但“看好”整个产业链，是一句正确但过于宽泛的废话。真正的机会在于那些能将AI能力与具体产业知识深度结合、解决真实世界问题的环节，而不是泛泛地投资“所有和AI沾边的概念”。对于黄河实业此类动作，我的看法是：密切关注其具体技术路线和商业模式，但对任何尚未落地的“Agentic AI”投资故事保持七分怀疑。真正的智能体革命，不会诞生于一份投资意向公告里，而可能诞生于某个车库团队对工作流程的彻底重构，或者某位工程师对Agent可靠性的死磕。

AI产业正在经历从“技术仰望期”到“价值兑现期”的艰难过渡。在这个过程中，宏大叙事依然动听，但真正决定胜负的，将是那些枯燥的、具体的、关于效率、成本和落地效果的细节。能见度高是好事，但别忘了，夜太深时，远处的灯火也容易成为海市蜃楼。