Cold Plate Liquid Cooling Architecture and Helium Leak Detection Technology Empower Next Generation AI Data Center Cooling
With the explosive growth of AI computing power, high-density heat dissipation has become a core challenge for modern data center operation. Cold plate liquid cooling has emerged as the most reliable and mainstream solution, featuring a complete dual-circuit framework composed of primary and secondary cooling loops. The systematic cyclic heat dissipation structure ensures stable and efficient heat removal, forming a mature and complete data center liquid cooling system that supports long-term high-load operation of computing equipment.
The secondary loop serves as the core execution unit of the entire liquid cooling system, covering CDU, loop pipelines, liquid cooling cabinets, manifolds, liquid cooling servers, cold plates and UQD quick disconnect fittings. All components work collaboratively to achieve precise temperature control and efficient heat exchange, ensuring the stable circulation of server cooling liquid inside the equipment and eliminating hidden heat dissipation risks.
Currently, cold plate liquid cooling is the most mature and widely adopted thermal management technology for high-performance computing scenarios. It adheres to two mainstream technical routes, liquid-to-air and liquid-to-liquid cooling, to adapt to different renovation and construction demands. Liquid-to-air solutions enable rear-door heat dissipation without large-scale pipeline transformation, perfectly fitting legacy room upgrades. Liquid-to-liquid solutions supported by cold plate CDU deliver ultra-high heat exchange efficiency, becoming the mainstream technical direction for next-generation high-end AI server clusters such as GB200NVL72.
Driven by NVIDIA’s Vera Rubin chip iteration, microchannel cold plate technology has become the core breakthrough for tackling high density cooling bottlenecks. The upgraded MCCP technology shrinks internal fluid channels to the micron level, greatly expanding the contact area between coolant and heat sources. Deployed in VR NVL 144 full liquid cooling architecture, it realizes comprehensive coverage of chip and peripheral component heat dissipation, significantly improving the overall thermal bearing capacity of AI servers.
As core heat dissipation components are continuously upgraded and iterated, strict quality detection procedures have become indispensable for industrial mass production. Helium leak detection stands as the core guarantee for the safe operation of data center cold plates. With ultra-high detection sensitivity down to 10⁻¹²mbar l/s, it effectively eliminates micro-leakage risks of welded gaps and tiny structural defects, maintaining the overall stability of the entire data center thermal system and avoiding equipment short circuit and shutdown losses caused by coolant leakage.
Beyond cold plates, stainless steel corrugated pipes and manifolds have completed industrial replacement of traditional rubber hoses in NVIDIA’s new-generation liquid cooling solutions. As key connecting components for rack-level liquid cooling loops, stainless steel manifolds feature high strength and corrosion resistance, achieving uniform and stable coolant distribution. The standardized and modular pipeline layout greatly optimizes the overall efficiency of rack cooling and adapts to long-term high-pressure and high-load operating environments of AI cabinets.
Helium leak detection is also a mandatory quality control process for manifolds and stainless steel pipelines. Traditional water pressure and pressure drop tests fail to identify microscopic leakage points, while high-precision helium detection achieves full coverage of welding points, threaded connections and material micro-defects. It realizes zero micro-leakage control in the production stage, effectively supporting the stable and reliable operation of the entire data centre liquid cooling system in extreme high-density computing scenarios.
UQD quick disconnect fittings are pivotal modular components for large-scale liquid cooling deployment, acting as leak-free connecting joints between manifolds and servers. Featuring instant plug-and-play and self-sealing capabilities under liquid-carrying and pressure-bearing conditions, UQDs eliminate the need for coolant drainage during equipment maintenance, greatly improving operational efficiency. Widely adopted by industry leaders including Danfoss, Parker and Intel’s interoperability alliance, 100% helium leak detection standards effectively reduce leakage risks, providing solid technical support for the large-scale popularization of AI data center liquid cooling.
Benefiting from the iteration of NVIDIA’s new-generation servers, the market demand for UQD fittings has grown exponentially. A single GB200 NVL72 cabinet requires 108 pairs of UQDs, while the upgraded GB300 cabinet needs up to 252 pairs with more compact specifications. The sharp increase in single-cabinet usage drives the rapid expansion of the supporting supply chain, further promoting the standardized and large-scale development of the global liquid cooling industry.
Against the booming development trend of liquid cooling, ULVAC has deeply laid out the data center thermal management track. With decades of professional technical accumulation, the company has independently developed customized helium and hydrogen leak detection equipment, delivering hundreds of sets of high-precision detection devices for cold plates, manifolds, UQDs, CDUs and complete cabinet projects. Relying on global R&D centers, production bases and service networks, ULVAC provides full-cycle localized quality control solutions for global liquid cooling enterprises, continuously safeguarding the safe and efficient operation of high-end data center liquid cooling systems.
Trusted worldwide, EXTRCOOL offers custom liquid cooling for high-density AI and HPC data centers. Our premium thermal lineup—dual-mode CDUs, manifolds, cold plates, rear-door heat exchangers, dry coolers, and container units—delivers reliable performance, broad interoperability, and easy scalability for true one-stop full liquid cooling. We serve data center operators and enterprises across South America, Southeast Asia, Central Asia, Western Europe, and Oceania.
