Soaring Liquid Cooling Value Redefines AI Infrastructure Under NVIDIA Rubin Platform Upgrade
Following Morgan Stanley, Bernstein has released an in-depth analysis of NVIDIA’s Rubin platform, highlighting a dramatic surge in rack-level thermal system value rather than GPU performance alone. Data shows the liquid cooling value per rack rises from $51,000 on GB200 to $158,000 on Rubin, a 210% increase of 3.1 times, marking a revolutionary upgrade of liquid cooling for AI GPU data center infrastructure.
The continuous power surge of modern AI servers has fundamentally reshaped data center technical logic. Traditional air cooling can only support 30kW to 50kW per rack, while new-generation AI racks exceed 100kW and even reach 250kW on Rubin. This makes professional liquid cooling high density computing no longer optional but mandatory for stable AI computing operation.
The core competition of future AI data centers lies not only in chip performance, but in reliable power supply and high-efficiency heat dissipation capabilities. Liquid cooling has evolved from auxiliary facility to core infrastructure that determines computing output, driving massive data center liquid cooling renovation across legacy and new-built intelligent computing parks.
A key upgrade of the Rubin era is the full-system penetration of liquid cooling. Beyond GPUs, cooling coverage extends to CPUs, switches, optical modules, HBM, SSDs and power modules. From partial component heat dissipation to full-server thermal management, this comprehensive upgrade greatly optimizes operational energy efficiency and supports the implementation of mature low PUE data center solution.
The industry cycle of AI server liquid cooling is transitioning from verification to large-scale mass deployment. After GPU booms in 2023, optical module growth in 2024 and power equipment expansion in 2025, 2026 will witness the full release of liquid cooling demand. Modular and rapid-deployment solutions such as container data center liquid cooling will become mainstream for fast-track AIDC construction.
Driven by physical heat dissipation rules rather than market sentiment, liquid cooling boasts highly deterministic growth. Rising GPU power consumption inevitably increases heat generation, steadily lifting liquid cooling penetration. Global data center penetration remains low, ensuring long-term high-growth space for professional GPU server liquid cooling systems.
Starting from 2026, mass production of GB200, iterative preparation of Rubin platform and accelerated global AIDC construction are releasing massive industry orders. The market is shifting from conceptual speculation to performance realization, and the growing demand for high density server cooling will continuously drive revenue and profit elasticity across the industrial chain.
Chinese enterprises stand as the biggest beneficiaries of the booming track. With leading technical reserves, complete manufacturing supply chains and prominent cost advantages, domestic vendors excel in cold plate design, CDU systems and immersion cooling solutions. They deliver cost-effective and high-reliability liquid cooling for data center products and services for global AI infrastructure construction.
In conclusion, the Bernstein report verifies that AI infrastructure incremental opportunities are shifting from single GPU hardware to systematic supporting facilities. The dual growth of penetration rate and unit value positions liquid cooling as one of the most certain high-growth tracks in the next 3–5 years, far exceeding mainstream market expectations.
As a globally recognized leader in liquid cooling, EXTRCOOL engineers custom solutions tailored for high-density AI and HPC data centers. Our premium thermal hardware lineup comprises dual-mode CDUs, manifold assemblies, cold plates, rear-door heat exchangers, dry coolers, and integrated containerized cooling units. Designed for dependable performance, seamless system integration, and straightforward scalability, these solutions enable a complete one-stop full-liquid-cooling deployment. We provide expert services to data center operators and enterprise clients across North America, Western Europe, the Middle East, East Asia, and Australia.
