NVIDIA (NVDA.US) Rubin CPX Adopts Innovative Decoupled Reasoning Architecture, Expected to Drive PCB Market Upgrade

Stock News
5 hours ago

A recent research report indicates that Rubin CPX is a GPU released by NVIDIA (NVDA.US) specifically designed for processing ultra-long context AI reasoning tasks, featuring an innovative decoupled reasoning architecture. Rubin CPX brings significant changes to the hardware side, particularly in connector/PCB aspects, adopting a wireless cable architecture while adding Paladin B2B connectors connected to a PCB mid-plane located in the center of the chassis. Rubin CPX introduces significant innovations in PCB, connector, and cooling architectures, which is expected to substantially increase its hardware-side market scale. The NVIDIA AI PCB market is projected to reach $6.96 billion in 2027, representing a 142% increase from 2026.

Main viewpoints are as follows:

**Rubin CPX Achieves Rebalancing of Efficiency and Cost Through Hardware-Level Optimization**

Rubin CPX is a GPU released by NVIDIA designed specifically for processing ultra-long context AI reasoning tasks, featuring an innovative decoupled reasoning architecture. In terms of computational performance, it provides 30 Peta FLOPS of computing performance under NV FP4 precision. Regarding storage performance, it is equipped with 128GB GDDR7 memory with a memory bandwidth of 2TB/s. In terms of process and packaging, the Rubin CPX chip adopts a traditional FC-BGA flip-chip single-chip design.

**Rubin CPX Brings Significant Hardware Changes**

In tray architecture, the front end of the tray adopts a modular design with four sub-cards distributed on both sides of the tray, consisting of 2 Rubin CPX GPX chips and two CX-9 network cards. For cooling, front-end cooling will be upgraded from air cooling to liquid cooling, with an additional liquid cooling plate added within the tray. Heat pipes and heat spreaders will conduct heat from the back of each GDDR7 memory module to this cooling plate. In connector/PCB aspects, it adopts a wireless cable architecture while adding Paladin B2B connectors connected to a PCB mid-plane located in the center of the chassis.

**Rubin CPX Drives Further Enhancement of Single PCB Value**

The estimated VR200NVL144 single cabinet PCB value is approximately 456,000 yuan, with single GPU corresponding PCB value of 6,333 yuan ($880), representing a 113% increase compared to GB300. It is projected that GB300 NVL72/VR200 NVL144/VR300 NVL576 shipments in 2027 will be 1/7/2 thousand racks respectively, totaling 100,000 racks, corresponding to a PCB market scale of $6.96 billion, representing a 142% increase from 2026.

**Related Industry Chain Targets**

Industry chain related targets include: Shengyi Technology (600183.SH), Shenznan Circuits (002916.SZ), Xingsen Technology (002436.SZ), Pengding Holdings (002938.SZ), Dongshan Precision (002384.SZ), Shenghong Technology (300476.SZ), Shenzhen Fastprint (002463.SZ), Shengyi Electronics (688183.SH), Kingboard Laminates (603228.SH), Founder Technology (600601.SH), Guanghe Technology (300964.SZ), and others.

**Risk Factors**

Rubin CPX implementation progress may fall short of expectations; AI and computing infrastructure demand may be lower than anticipated.

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Most Discussed

  1. 1
     
     
     
     
  2. 2
     
     
     
     
  3. 3
     
     
     
     
  4. 4
     
     
     
     
  5. 5
     
     
     
     
  6. 6
     
     
     
     
  7. 7
     
     
     
     
  8. 8
     
     
     
     
  9. 9
     
     
     
     
  10. 10