英偉達引領新一代人工智能硬件升級

🧠 NVIDIA’s Rubin Series Ushers in Next-Gen AI Hardware Upgrades — Implications for PCB Value Chain $英偉達(NVDA.US)

NVIDIA is poised to maintain its “one major upgrade per year” cadence in product development. Its upcoming Rubin GPU series will see significant advancements across multiple technical dimensions:

• HDI (High-Density Interconnect) levels will be upgraded from the current 5-layer design to a more advanced 7-layer structure.

• Core CCL (Copper Clad Laminate) materials will move from existing grades to M9, offering superior heat resistance and signal integrity.

• The high-end Rubin Ultra variants are expected to incorporate Kyber backplanes made from M9 or PTFE materials, featuring 60–70 PCB layers per GPU.

These advancements translate to a higher per-unit PCB value for each GPU, benefiting upstream suppliers in the PCB and advanced materials ecosystem.

Meanwhile, ASIC-based AI servers, which currently mostly adopt M8 materials, are also evolving. Future ASIC boards are expected to exceed 30 layers, and their per-chip PCB value will significantly surpass that of the NVIDIA ecosystem, presenting another long-term growth vector.

Simultaneously, the rapid evolution of data center networking equipment—particularly 800G and 1.6T switches—is expected to drive strong growth in both penetration and shipment volumes over the next two years. These switches demand higher-spec PCBs and carry elevated ASPs, further reinforcing the industry’s bullish outlook.

All of these technological transitions point to increasing product complexity and rising entry barriers. As a result, the “technology + capacity” flywheel effect will increasingly favor leading PCB and material manufacturers, amplifying industry consolidation and value capture by top-tier players.