Guosen: NVIDIA Corporation (NVDA.US) Rubin architecture release, CCL upstream material system upgrade.

date
11:38 03/06/2026
avatar
GMT Eight
In AI servers, the GPU board group is an increment compared to regular servers. Taking NVIDIA servers as an example, the GPU board group mainly includes GPU components, module board NVSwitch, and all three parts will be applied to high-level CCL boards.
Guosen released a research report stating that it recommends paying attention to Jinan Shengquan Group Share Holding (605589.SH), which has independently developed polyphenylene ether resin that has been certified by key domestic enterprises. The company also has production capabilities for hydrocarbon resins (ODV) with rapidly expanding capacity. Sichuan Em Technology (601208.SH) has independently developed electronic-grade resin materials such as hydrocarbon resin, polyphenylene ether resin, pyromellitic dianhydride resin, and special epoxy resin, which are used by various leading copper clad laminates (CCL) manufacturers downstream. In the future, due to its extremely low dielectric loss (Df) value, polytetrafluoroethylene is expected to enter the M10 material system, and it is recommended to pay attention to the leading domestic polytetrafluoroethylene company DONGYUE GROUP (00189). Guosen's main points are as follows: Demand for AI servers, general servers, etc., driving high-end CCL market space Due to the rapid deployment of AI applications and the increased capital expenditure by technology giants in AI, the shipment volume of AI servers is rapidly increasing. Within AI servers, the GPU board group is an incremental component relative to general servers. For example, in NVIDIA servers, the GPU board group mainly includes GPU components, module board NVSwitch, all of which will utilize high-grade CCL boards. General servers are in a key transitional period for upgrading to CCL, and there is a noticeable increase in the number of CCL layers required after server iteration. The demand for high-speed CCL in high-performance servers is continuously expanding, driving the growth of CCL material demand. NVIDIA's Vera Rubin architecture fundamentally shifts the role of PCB in AI racks From primarily passive carriers for intraboard connections in the past, PCBs have been upgraded to become active media for high-speed interconnects within racks. Some of the value that originally belonged to copper cables, connectors, backplane systems has transferred to PCBs, leading to the "semiconductorization" of PCBs. This has significantly increased the value of upstream materials such as CCL, with a significant upgrade to the M8-M9 material system. The future Rubin Ultra is expected to upgrade to the M10 material system, with a significant increase in PCB usage. Pay attention to the rapid growth in demand for upstream materials and product iterations and upgrades. Special electronic resins are an important direction for material upgrades Electronic resins are the only organic materials in the production of copper clad laminates that can be designed. Factors affecting dielectric loss include the polarity of molecules and the density and mobility of polar groups. Reducing dielectric constant (Dk) and dielectric loss (Df) is an important direction. Currently, common resins used in high-end copper clad laminates include polyphenylene ether resin (PPO), hydrocarbon resin (CH), and polytetrafluoroethylene resin (PTFE). The electrochemical properties of hydrocarbon resin and polytetrafluoroethylene resin are excellent. As the NVIDIA material system upgrades, there will be rapid growth in demand for hydrocarbon resin and polytetrafluoroethylene resin in the future. Supply and demand for glass fiber electronic cloth is tight, a bottleneck in the production of copper clad laminates The M8 & M9 materials for NVIDIA's AI servers correspond to glass fiber electronic cloth with low dielectric constant, low dielectric loss, and low thermal expansion characteristic. However, due to supply capacity constraints and loom equipment limitations, prices of the product have continued to rise, making it the most scarce segment in upstream materials. Silicon microfillers have risen from "ordinary fillers" to "core functional materials" In the material upgrade for NVIDIA's Rubin platform-driven AI server, silicon microfillers have transitioned from auxiliary fillers to core main materials that determine the performance of copper clad laminates. Their quantity, price, and technological requirements have undergone generational leaps, driving silicon microfillers towards higher filling, lower dielectric loss, and smaller particle sizes, significantly increasing their product value. Risk factors: Risks of intensifying competition; risks of technological iterations falling short of expectations; risks of downstream demand falling short of expectations; risks of capacity expansion falling short of expectations.