① Orders for current liquid cooling equipment are generally scheduled until the end of this year, with the CDU segment being the most crowded; ② Nvidia Vera Rubin has established a fanless all-liquid cooling architecture, and the penetration rate of liquid cooling in AI chips is expected to rise to 53%; ③ Domestic quick connectors have joined Nvidia Rubin’s ecosystem, and orders for liquid cooling pumps and equipment are beginning to materialize. There is still a significant gap in fluorinated fluids, micro-channel coolers, and overseas products.
Nvidia’s Vera Rubin establishes a fanless all-liquid cooling architecture, and the penetration of liquid cooling in AI chips is expected to rise to 53%. Currently, orders for liquid cooling equipment are generally scheduled for completion by the end of this year, with the CDU segment being the most competitive. Domestic quick connectors have joined Nvidia Rubin’s ecosystem, and orders for liquid cooling pumps and equipment are beginning to materialize, but there is still a significant gap in fluorinated fluids, micro-channel cold plates, and overseas markets. Global AI hardware power consumption is increasing, and traditional air cooling is approaching its limits; liquid cooling has become the mainstream choice. Vera Rubin fully adopts a fanless all-liquid cooling solution, increasing the power consumption per cabinet to approximately 225kW, and PUE can be reduced to 1.08 to 1.12. The penetration of liquid cooling in AI chips is expected to rise from 33% in 2025 to 53% in 2026, with overseas cloud manufacturers driving 60% of the global growth. With the adoption of the all-liquid cooling architecture, the value of liquid cooling per cabinet has increased from approximately $41,500 for the GB200 to $57,500 for the Rubin NVL144 platform. CDU, Manifold, U…