Zhongji Innolight: Production Capacity Continuously Expanding 中际旭创：公司产能在持续扩建中

This dissonance is a true reflection of the current AI frenzy. Everyone believes the future lies in uncharted territories, so they're betting heavily on "shovels"—NVIDIA's chips, Zhongji Innolight's optical modules—they're the shovels of the new era. But reality is, where is the "gold mine" these shovels are digging up? Apart from a few giants raking in profits from selling computing power (cloud services) and some efficiency tools quietly transforming certain white-collar workflows, the "killer app" that could ignite a global consumer upgrade cycle or create entirely new hardware categories remains a mirage. Hence, we see hot topics like "After using AI, the company seems poorer." This isn't a joke; it's the stark reality faced by countless enterprises, especially small and medium-sized companies, after enthusiastically embracing AI: the cruel gap between high API call fees, model fine-tuning costs, computing expenditures, and unclear or even overestimated ROI (Return on Investment).

Technology adoption is never uniform. In the upstream infrastructure layer (like Zhongji Innolight's production capacity), it manifests as exponential growth, but in the application layer, it's often a prolonged, wave-like climb. What we're experiencing now is the most agonizing phase of this application-layer ascent. Take the AI video field, for example: news that "From Kling to Gemini, AI video collectively bids farewell to 'gacha mode'" is exciting—it signals improved quality control and consistency, with AI video moving from a "blind box" toy to a reliable production tool. But what's the subtext of "Director models about to explode"? It points to higher professional thresholds and shifting cost structures. Previously, anyone could try their luck with AI "gacha"; in the future, it may require more specialized models and complex workflows to produce acceptable content, implying a different business logic and cost structure.

Even more intriguing are the messages sketching ultimate blueprints. "RSI that lets AI build itself takes off, Google pours cold water"—RSI (Recursive Self-Improvement) is a concept that both excites and terrifies on the road to AGI (Artificial General Intelligence), pointing to the singularity of an intelligence explosion. Tech giants discussing these in top academic journals and launches are undoubtedly necessary for flexing muscles and seizing technological discourse. However, on the commercial side where quietly making profits happens, AI's most tangible "self-construction" might not be that sci-fi-esque self-iteration, but something like what's hinted at in "Spring founder returns to the frontlines to build an AI framework"—AI is reshaping the methodology of building software itself. When a generation of "the last framework chosen by humans" appears, it signifies a fundamental change in the underlying developer workflow. This silent, tool-chain-level revolution may have more far-reaching impact than any flashy AI application.

Thus, the current AI industry presents a peculiar split: the primary market and tech media chase sexy narratives like AGI, self-construction, multimodality, and video generation; the secondary market and the real economy focus more on Zhongji Innolight's capacity utilization, BOE's panel price fluctuations, and the real economic pulse behind "street stall equipment surging 600%." The AI story is a paradigm revolution in the mouths of tech elites, a certainty track in investors' eyes, potentially a cost center on an ordinary company's financial statements, and perhaps just a convenient tool for generating more attractive product descriptions on the live-streaming phones of street stall entrepreneurs.

Amid the frenzy, what needs the most caution isn't the technology's bottlenecks, but the narrative bubble. When a company's production capacity expansion announcement becomes news, when the impact of memory chip price hikes is analyzed alongside the AI hot list, perhaps we should pause and reflect: AI is indeed reshaping everything, but this process is not a smooth exponential curve; it's a complex game full of structural friction, cost pass-through, and demand validation. The optical cables expanded by Zhongji Innolight must ultimately carry the data and applications flowing on the screens of countless BOEs that can generate real profit—not just fantastical demo videos and grandiose statements at launches. Between computing power and demand, bubble and pragmatism, sci-fi and the bottom line, the final outcome of this grand gamble depends not on who has more shovels, but on what can actually be dug up with them.

中际旭创又发公告了，说产能在“持续扩建”。在AI算力军备竞赛的今天，这几乎算不上新闻，更像是每天早晨的打卡签到——签到证明自己还在牌桌上，而且筹码还在不断增加。同一时间，另一边厢的京东方，却在机构调研里谨慎地谈论着存储芯片涨价可能对笔记本电脑、智能手机等终端需求带来的“一定程度的影响”。两则简讯并列，一股微妙的撕裂感扑面而来：一边是上游算力基础设施的狂热扩张，仿佛明天AI就要吞噬一切；另一边却是下游消费电子需求可能面临的现实寒流。整个科技产业链，就像一头被AI这颗强劲心脏驱动的巨兽，肌肉不断膨胀（中际旭创们的产能），但末端的血液循环（终端消费）却可能不那么顺畅。

这种撕裂，正是当前AI热潮最真实的写照。所有人都相信未来是星辰大海，所以不惜重金下注“铲子”——英伟达的芯片、中际旭创的光模块，它们是新时代的铲子。但现实是，用这些铲子挖出来的“金矿”在哪里？除了少数几家巨头靠卖算力（云服务）赚得盆满钵满，以及一些效率工具悄悄改变了部分白领的工作流程，那个能够引爆全球消费者新一轮换机潮、创造全新硬件品类的“杀手级应用”，至今仍像海市蜃楼。于是，我们看到了“用了AI之后，公司好像更穷了”这种热榜话题。这不是段子，而是无数企业，特别是中小型公司，在激情拥抱AI后面对的真实账单：高昂的API调用费用、模型微调成本、算力开支，与尚不清晰、甚至被高估的ROI（投资回报率）之间的残酷落差。

技术的渗透从来不是匀速的。它在上游基础设施层（如中际旭创的产能）表现为指数级增长，在应用层却常常是漫长的、波浪式的爬坡。我们正在经历的，正是应用层爬坡最煎熬的阶段。看看AI视频领域，“从可灵到Gemini，AI视频集体告别‘抽卡模式’”的消息令人兴奋，这意味着质量的可控性和一致性在提升，AI视频正从“开盲盒”的玩具，向可靠的生产工具迈进。但“导演模型要火”的潜台词是什么？是专业门槛的提高和成本结构的改变。以前是个人就能用AI“抽卡”试试运气，未来可能得借助更专业的模型、更复杂的流程才能产出合格内容，这背后是另一套商业逻辑和成本。

更值得玩味的是那些描绘终极蓝图的消息。“让AI自我构建的RSI火了，Google泼冷水”，RSI（递归自我改进）是通往AGI（通用人工智能）道路上令人既兴奋又恐惧的概念，它指向的是智能爆炸的奇点。科技巨头们在顶级学术期刊和发布会上讨论这些，无疑是展示肌肉、抢占技术话语权的必要手段。然而，在闷声发大财的商业层面，AI最切实的“自我构建”，或许并非那种科幻般的自我迭代，而是像“Spring 创始人重回一线做 AI 框架”所暗示的——AI正在重塑构建软件本身的方法论。当一代“人类亲自选择的最后一代框架”出现，意味着开发者工作流底层的彻底变革。这种静默的、工具链层面的革命，其影响力可能比任何一个炫酷的AI应用都更深远。

所以，当下AI行业呈现出一种奇特的分裂景象：一级市场和科技媒体追逐着AGI、自我构建、多模态、视频生成等性感叙事；二级市场和实体经济则更关注中际旭创的产能利用率、京东方面板价格波动、以及“地摊设备暴涨600%”背后的真实经济脉搏。AI的故事，在技术精英口中是范式革命，在投资者眼中是确定性赛道，在普通公司财务账上可能是成本中心，在摆摊创业者的直播手机里，或许只是一个用来生成更吸睛产品描述的便捷工具。

热潮之中，最需要警惕的不是技术的瓶颈，而是叙事的泡沫。当一家公司的产能扩建公告都能成为资讯，当存储芯片涨价的影响被放到AI热榜旁分析时，我们或许该冷静一下：AI确实在重塑一切，但这个过程不是平滑的指数曲线，而是充满结构性摩擦、成本转嫁和需求验证的复杂博弈。中际旭创们扩产的光缆，最终需要承载的是无数京东方们屏幕上流转的、能创造真实利润的数据与应用，而不仅仅是光怪陆离的演示视频和发布会上的豪言壮语。在算力与需求、泡沫与实干、科幻与账单之间，这场豪赌的终局，不取决于谁的铲子更多，而取决于用这些铲子到底能挖出什么。