CICC Investment: High-frequency and High-speed Demand such as Computing Power is Growing Rapidly, Electronic-grade PTFE is Expected to be Widely Used 中信建投：算力等高频高速需求快速增长，电子级PTFE有望大规模应用

PTFE, the so-called "plastic king," is suddenly at the center of a quiet but seismic shift in AI hardware architecture. This isn't about some breakthrough chemical synthesis. It’s about the ruthless, practical reality of physics catching up with software’s grand ambitions. As NVIDIA’s next-generation Rubin Ultra servers approach, the industry is seriously discussing ditching traditional backplane materials for PTFE-based orthogonal designs. That single sentence is packed with implications that go far beyond a material substitution.

First, this exposes a fundamental blind spot in the AI narrative. For years, we’ve been mesmerized by the silicon—the GPUs, the TPUs, the dazzling TOPS figures. But the physical substrate that connects them, the "high-frequency high-speed transmission" plumbing, has been a boring, assumed constant. Now, with data rates pushing into the stratosphere for AI training clusters, that plumbing has become the bottleneck. The signal integrity and dielectric properties of your circuit board are suddenly as mission-critical as your algorithm. The industry is effectively admitting that its software ambitions have outpaced its materials science. It’s a humbling, necessary correction.

Second, watch the strategic chessboard. This is not just about better plastics. It’s about supply chain control and technological sovereignty. The mention of Chinese firm Shengyi Technology actively validating PTFE for this use case is the real headline. In the global tech cold war, dominance isn’t just about who designs the best chip, but who can reliably produce the critical, next-generation materials that go into the system. If PTFE becomes the substrate standard for next-gen AI servers, then the companies mastering its high-purity, high-performance manufacturing for electronics—not just chemical giants—will wield immense power. This is materials science as geopolitics.

Third, let’s be blunt about the industry’s hype cycle. We’re so fixated on the "next big thing" in software—foundation models, agents, reasoning—that we undervalue the incremental, hard-won advances in enabling hardware. This PTFE shift isn’t a sexy headline, but it’s a profound enabler. It’s the difference between an AI cluster that’s merely powerful and one that’s actually stable, efficient, and capable of scaling without the backplane turning into a noisy, heat-warped disaster. The most important innovations are often the least glamorous.

So, what does this redefine? It redefines the bottleneck. For the past decade, we talked about the memory wall and the I/O wall. Now, we’re talking about the material wall. The performance ceiling of large-scale AI is being set not just by lithography, but by polymer chemistry. It forces a reevaluation of the entire R&D pipeline, elevating material scientists to the same echelon as chip architects. It suggests that the next leap in AI performance might not come from a smaller transistor, but from a better dielectric.

Ultimately, this is a wake-up call about the physicality of intelligence. We can’t code our way out of electromagnetic interference or signal degradation. The cloud is not an ethereal realm; it’s a very hot, very physical stack of metal, plastic, and silicon. The sudden spotlight on PTFE is a stark reminder that in the race to build artificial minds, we’re still fundamentally constrained by the humble, remarkable properties of the materials we use to house them. The plastic king is having its moment, not because it’s new, but because we finally hit the wall where its old, reliable virtues become the most cutting-edge feature in the server room.

券商研报把PTFE叫做“塑料王”时，总带着一种发现新大陆的兴奋。但这次关于英伟达Rubin Ultra服务器可能采用PTFE正交背板的讨论，确实戳中了产业链一个真实的痒点——当算力狂奔到一定程度，瓶颈往往不在芯片本身，而在那些连接芯片的“管道”和“地基”上。

PTFE（聚四氟乙烯）的老本行是“耐操”：不粘锅涂层、化工管道衬里、电线绝缘。它的化学稳定性近乎顽固，介电性能在塑料里属于顶级，但加工难度也是出了名的大。以前在军工和高端通信领域用，是因为那些场景对“可靠性”的溢价愿意买单。现在突然被AI算力基建盯上，本质是高频高速传输对材料性能的极限压榨。铜缆到一定频率就扛不住了，传统PCB材料在高频信号下的损耗像漏水的水管，而PTFE那种“信号穿过去损耗极低”的特性，突然变成了香饽饽。

研报里罗列的三大需求——军工、服务器线缆、高速板——其实指向同一个逻辑：全球算力军备竞赛正在倒逼上游材料革命。英伟达的GPU迭代速度已经让散热和供电焦头烂额，但信号传输这条“神经网络”的升级同样迫在眉睫。Rubin Ultra这种下一代怪兽级服务器，内部互联带宽的需求可能是当前产品的数倍，传统材料架构很可能成为性能天花板。讨论用PTFE做正交背板，不是在追时髦，而是在解决一个工程上的物理极限问题。

国内生益科技“配合积极做验证”这句话值得玩味。PCB行业长期在追赶，从FR-4到高频材料，核心专利和工艺know-how大多捏在罗杰斯、泰康利等海外巨头手里。PTFE基板对加工精度、表面处理、层压工艺的要求极为苛刻，国内厂商能挤进验证名单，至少说明在某些细分赛道上不再只是看客。但“配合验证”和“量产供应商”之间，隔着良率爬坡、成本控制、客户信任三大高山。券商研报里“有望高速增长”的预测，往往线性得过于美好，忽略了新材料导入周期里那些咬牙切齿的攻坚细节。

更辛辣的是，这种材料层面的“军备竞赛”背后，是算力话语权的隐秘争夺。当英伟达定义下一代服务器架构时，连背板用什么塑料都可能影响整个供应链格局。PTFE的下游被“重新定义”，本质是AI基础设施的标准化战争向前端材料的蔓延。国内产业链如果只能跟在验证环节，等标准定型后再做适配，那么“高速增长”的利润蛋糕大概率还是被上游材料厂和设计厂商切走大半。

不过，把宝压在一种材料上也有风险。实验室数据到工程量产之间，永远有距离。PTFE加工温度高、与铜箔结合力弱等痛点，不是一句“材料性能优异”就能带过。更何况，替代方案也在暗流涌动：改性聚烯烃、液晶聚合物（LCP）等材料都在虎视眈眈，谁都不想让出高频高速这个未来的主战场。

所以，这篇研报与其说是看多PTFE，不如说是在警告：算力的瓶颈正在从硅片蔓延到沙子（硅的源头）。当所有人都盯着芯片制程时，那些不起眼的连接材料、封装基板，可能才是决定AI大厦能盖多高的地基。生益科技们的积极验证，至少是踏进了这场游戏的内场——虽然离真正改变牌局还很远，但总比站在门外讨论“有望”来得实在。券商喜欢用“有望”这个词，既留足了预测退路，又点燃了市场想象。而对于产业而言，真正的战场永远在“有望”和“量产”之间那条布满荆棘的路上。