NVIDIA GTC 2026 Redefines Liquid Cooling for Next-Gen AI Computing
Most industry practitioners still define liquid cooling as a simple chip heat dissipation method. However, NVIDIA delivered a landmark move at GTC 2026 that overshadowed its new chip launches. For the first time, NVIDIA has taken direct control of the liquid cooling supply chain from traditional OEMs and established an officially certified supplier roster covering four core components: cold plates, quick connectors, manifolds, and CDUs. Liquid cooling is no longer a generic thermal solution, but a strictly certified computing infrastructure locked by NVIDIA’s authoritative qualification system. The upgraded liquid cooling data center ecosystem has become the indispensable foundation for supporting ultra-high-power next-generation AI supercomputing platforms, completely overhauling the traditional heat dissipation logic of artificial intelligence infrastructure.
NVIDIA’s newly launched Vera Rubin NVL72 supercomputing platform has fully eliminated air cooling configurations and established 100% full liquid cooling as the factory standard. Equipped with 72 Rubin GPUs, 36 Vera CPUs, ConnectX-9 SuperNICs and BlueField-4 DPUs, the platform delivers five times the inference performance of the Blackwell architecture. With a single GPU power consumption of 2300W and a rack power range of 190–230kW, it adopts 45°C warm water circulation and a 75% water plus 25% propylene glycol cooling medium, removing the need for conventional chillers entirely. The upgraded MCCP micro-channel cold plate drastically improves heat exchange efficiency, cutting infrastructure parasitic losses from 30% to 5% and shortening on-site installation from two days to two hours. Professional and verified rack liquid cooling supplier partners strictly follow NVIDIA’s unified thermal design specifications to ensure consistent and reliable cooling performance for mass-deployed Rubin AI racks.
Cold plates represent NVIDIA’s first critical move to regain control of the liquid cooling supply chain. Previously, cold plate specifications were independently defined by OEMs including Quanta, Foxconn and Supermicro. NVIDIA has now centralized standard formulation, unified certification and centralized procurement, and enforced full component interchangeability across all ODM and OEM servers for the Rubin platform. Four tier-one suppliers including Cooler Master, AVC, Delta and Jentech have obtained official certification and commenced mass production, with single-system cold plate costs rising to USD 57,000 and above. As a professional data center air liquid CDU manufacturer, qualified vendors deliver matched high-precision CDU equipment that perfectly coordinates with standardized cold plate systems, realizing efficient and stable heat circulation for full liquid-cooled AI data centers.
Quick connectors are the most vital yet easily overlooked core components of liquid cooling systems, with each Rubin rack deploying 200 to 300 units. A single leakage during plugging and unplugging can trigger full rack downtime, highlighting extreme reliability requirements. The industry has undergone a generational upgrade from the traditional UQD04 solution dominated by European and American brands to the compact NVQD03 solution, with Taiwanese and mainland Chinese manufacturers achieving rapid substitution. Adopting hose-free design and blind plugging supports hot-swap maintenance and greatly improves operational convenience. The high-stability air liquid CDU cooling unit cooperates with upgraded high-reliability quick connectors to eliminate system leakage risks, ensuring continuous and safe operation of ultra-high-density AI computing clusters.
As the core component responsible for evenly distributing coolant to each cold plate, manifolds serve as the key bridge transforming single-channel rack cooling into multi-loop efficient heat dissipation. The Rubin platform adopts the unified MGX 44RU standard and replaces traditional rubber hoses with 316L mirror stainless steel corrugated pipes, compatible with glycol warm water, fluorinated liquid and silicon-based cooling media. Liminda stands out as the sole certified mainland Chinese manifold supplier, owning dual NVIDIA AVL and RVL certifications with leading processing precision and explosive business growth. A complete and optimized rack manifold liquid cooling system achieves balanced coolant delivery across all cooling branches, effectively avoiding local overheating and maximizing the overall thermal management efficiency of high-power AI racks.
CDU, known as the control heart of liquid cooling racks, undertakes core functions including heat exchange, pressure stabilization, flow distribution and real-time leakage monitoring. The previous GB200 generation relied almost exclusively on Vertiv for CDU supply, while NVIDIA has expanded the qualified supplier pool to six to seven mainstream manufacturers for the new Rubin platform, covering international industry giants and leading domestic enterprises. Frequent cross-border mergers and acquisitions reflect global leaders’ pursuit of systematic full-link thermal management capabilities. The high-value rack air liquid cooling CDU realizes precise intelligent regulation of liquid cooling loops, adapting flexibly to diverse power load changes and meeting the stringent heat dissipation demands of next-generation full liquid-cooled racks.
NVIDIA has further deepened strategic cooperation with the OCP Open Compute Project to promote comprehensive industry standardization for next-generation AI liquid cooling infrastructure. The collaboration focuses on unified liquid cooling interface compatibility, standardized UQD/MQD quick connector specifications and universal MGX 44RU rack design norms. Multiple industry giants jointly participate in formulating new industry rules, building a closed-loop standard system from chips and servers to terminal infrastructure. The integrated liquid cooling CDU manifold complies fully with upgraded OCP UQD V2.0 specifications, realizing cross-brand component interchangeability and laying the foundation for large-scale standardized industrial deployment.
NVIDIA has established a complete hierarchical supplier certification system covering NPN registration, RVL reference list, AVL qualified list and top-tier NPN Tier 1 system certification, forming extremely high industrial entry barriers. The full certification cycle lasts 1.5 to 3 years with strict requirements on cleanliness, zero leakage and megawatt-level heat dissipation capacity. In the current industry landscape, product competition will shift to certification and production capacity competition from 2027 onward. Certified suppliers have secured orders extending to 2029, while late entrants face rising technical and qualification thresholds. The authoritative data center liquid cooling manifold has become a core qualification barrier for enterprises to enter NVIDIA’s exclusive supply chain, determining the long-term competitive pattern of the global AI liquid cooling industry.
As AI evolves, so does its thermal output. EXTRCOOL provides a complete liquid cooling solution set, including CDUs, cold plates, manifolds, and piping. You can choose standard units or have them customised to your facility's layout. Our systems are designed for high heat dissipation, low power draw, and reduced carbon emissions. They have proven reliable at gigawatt-scale data centres across multiple continents. When uptime is non-negotiable, EXTRCOOL is the reliable choice.
Browse our full catalogue or ask for a personalised quote at: https://www.extrcool.com
