CITIC Securities: Distributed Expansion of Optical Interconnect Boundaries, DCI Industry Chain Faces Systemic Restructuring 中信证券：分布式拓展光互联边界，DCI产业链迎系统性重构

This is the true moment of "disenchantment." While public debates still rage over which model is smarter or which application is more disruptive, Wall Street analysts and engineers have long begun measuring AI’s future in kilowatts and optical modules. The term "computing wall" in the report is precise. No matter how many advanced GPUs are stacked in a data center, it will eventually collide with the iceberg of local grid capacity and substation approval. This is not a technical issue; it is a dual iron curtain of physics and administration. Thus, the logic becomes crystal clear: since data and tasks can no longer be "involutionally" contained within a single park, they must "spill over" to distributed computing and collaboration across broader geographical spaces. This is the essence of "Scale Across"—a forced yet imaginative relocation across physical space.

At the heart of this relocation is DCI (Data Center Interconnect). In the past, DCI might have been used for backup, disaster recovery, or handling routine traffic—essentially the "connection lines" between data centers. But today, when it must support training trillion-parameter models, the requirements become formidable: high density, high bandwidth, low latency, and high reliability—all indispensable. This is not an upgrade but a complete overhaul. It directly signals that the old DCI business model, dominated by a few downstream system integrators, is collapsing. Value is undergoing a quiet but profound "upward shift."

Previously, the most profitable players in optical transmission equipment were the "general contractors" who assembled devices into complete solutions. Now, the situation has changed. AI demands the utmost performance squeezed into every bit of fiber. Whoever can advance modules from 100G to 400G, 800G, or even 1.6T, and whoever can produce more stable and compact optical amplifiers, becomes the shovel-sellers and cement-makers of this new world—those who collect tolls on the essential pathways. Components like chips, modules, and amplifiers, once overshadowed by vast systems, have suddenly ascended to the forefront of value. The report’s mention of "profound industrial chain restructuring" centers on this: discourse power and profit pools are undergoing an epic shift from downstream integration and assembly to mid- and upstream precision optoelectronic manufacturing.

For Chinese manufacturers, this is undoubtedly a tempting window of opportunity. The report highlights "mature manufacturing advantages and supply chain collaboration capabilities"—a polite way of putting it. To be blunt, our cost control, mass production capabilities, and rapid response speeds in the optical module sector are now an undeniable force in the global industrial chain. In the past, we relied on this to secure orders and handle contract manufacturing; now, we have the opportunity to directly enter the core supply chain of global AI computing infrastructure. This is no longer just about earning processing fees but potentially participating in defining the technological standards for next-generation infrastructure.

But hold off on cheering. Behind the opportunities lie deeper challenges. The upstream shift in value means competition will revolve around harder technological barriers: the design and manufacturing of high-speed optoelectronic chips, the application of new materials, and ultra-precision packaging processes. If we settle for scale advantages in module assembly without breakthroughs in upstream "brain" components like optical chips and DSP (Digital Signal Processor) chips, then the majority of profits from this feast will ultimately flow to the international giants that control core technologies. We may build most of the "highways," but the keys to the most expensive toll stations remain in others’ hands.

Therefore, CITIC Securities’ report is less an industry forecast and more a warning of an "infrastructure crisis." It reveals a stark reality beneath the AI frenzy: the path to artificial general intelligence is paved with cold concrete, humming transformers, and fiber-optic cables threading through desert sands. While everyone looks up at the stars of algorithms, the truly decisive battle is unfolding underfoot on this land. The future of computing power lies not only in code but also in fiber optics, in substation blueprints, and in the tedious R&D that might improve optoelectronic conversion efficiency by 0.1%. Whoever controls this physical "pipeline" truly holds the lifeline of the AI era’s energy.

当全球AI巨头们还在为“规模法则”（Scaling Law）的无限前景而兴奋时，真正的瓶颈正在以一种物理的方式悄然降临——不是算法，不是数据，而是电和土地。中信证券的这份报告，看似在分析一个产业链的技术迁移，实则捅破了一层窗户纸：AI军备竞赛的终极战场，正在从炫目的模型参数比拼，转移到最枯燥、最硬核的基础设施领域。所谓“从Scale Up/Out到Scale Across”，翻译成人话就是：单个“算力城堡”已经建到极限了，现在得在这些城堡之间，修一条前所未有的高速公路。

这才是真正的“祛魅”时刻。当舆论场上还在争论哪个模型更聪明、哪个应用更颠覆时，华尔街的分析师和工程师们早已开始用千瓦和光模块来丈量AI的未来。报告里“算力墙”这个词用得精准。一座数据中心，再怎么堆叠最先进的GPU，最终都会撞上当地电网容量和变电站审批的冰山。这不是技术问题，是物理和行政的双重铁幕。于是，逻辑就变得无比清晰：既然不能让数据、让任务继续在单个园区内“内卷”，那就只能把它们“外溢”出去，在更广阔的地理空间上进行分布式计算和协同。这就是“Scale Across”的本质——一场被迫的、却又充满想象力的物理空间大挪移。

而这场挪移的核心“管道”，就是DCI。过去的DCI，可能只是备份、容灾或者分担一些常规流量，算是数据中心的“连接线”。但今天，当它要承载的是训练一个万亿参数模型时，需求就变得狰狞起来：高密度、高带宽、低延迟、高可靠性，缺一不可。这根本不是升级，而是彻底的颠覆。它直接宣告了，旧有的、由少数几家下游系统集成商掌控的DCI商业模式，正在崩塌。价值，正在发生一场静悄悄但意义深远的“上移”。

以前，做个光传输设备，最赚钱的是最后把这些设备组装成完整解决方案的“总包商”。现在，情况变了。AI需要的是能挤进光纤里的每一比特极致性能。谁能把100G的模块做到400G、800G甚至1.6T，谁能做出更稳定、更密集的光放大器，谁就成了这个新世界里卖铲子的、卖水泥的，成了那个坐在必经之路上收过路费的人。芯片、模块、放大器这些曾经在庞大系统阴影下的“零部件”，一跃成为了价值的制高点。报告里说的“产业链深刻重构”，核心就在这里：话语权和利润池，正在从下游的集成组装，向中上游的精密光电制造环节进行一次史诗级转移。

对于中国厂商而言，这无疑是一个极具诱惑力的窗口。报告提到“成熟的制造优势和供应链协同能力”，这话说得客气。直白点讲，我们在光模块领域的成本控制、量产能力和快速响应速度，已经是全球产业链里无法忽视的一股力量。过去，我们靠这个接订单、做代工；现在，我们有机会靠这个，直接切入全球AI算力基建的核心供应链。这不再只是赚点加工费，而是可能参与到定义下一代基础设施技术标准的游戏里。

但且慢欢呼。机遇的背后，是更深层次的挑战。价值向上游转移，意味着竞争的核心将是更硬核的技术壁垒：高速光电芯片的设计与制造、新材料的运用、超高精度的封装工艺。如果我们满足于在模块封装环节的规模优势，而无法在最上游的光芯片、DSP（数字信号处理器）芯片等“大脑”环节取得突破，那么这场盛宴的大部分利润，最终仍将流向那些掌握核心技术的国际巨头手中。我们或许能修好大部分的“高速公路”，但收费最贵的那个收费站，钥匙还在别人手里。

所以，中信证券的这份报告，与其说是一份行业预测，不如说是一份“基础设施危机”预警。它揭示了AI狂热下的一个冷峻现实：通往通用人工智能的道路，是由冰冷的混凝土、嗡嗡作响的变压器和穿梭于沙漠地下的光缆铺就的。当所有人都在抬头仰望算法的星空时，真正的决定性战役，正在脚下这片土地上展开。算力的未来，不只在代码里，更在光纤里，在变电站的图纸里，在那些能把光电转换效率提升0.1%的枯燥研发中。谁掌控了这条物理维度的“管线”，谁才真正握有AI时代的能源命脉。