NVIDIA has finalized its next-generation liquid cooling strategy: shielded pumps will become standard equipment in CDUs, marking a technological turning point for the liquid cooling pump market.
The surge in original AI computing power is driving a complete transformation in data center architecture. At GTC Taipei and Computex, NVIDIA announced that its next-generation 800V HVDC (High-Voltage Direct Current) CDU will fully adopt high-voltage DC electronic shielded pumps. This marks a pivotal technological shift in the liquid-cooled CDU pump sector, with zero-leakage, high-performance shielded pumps now officially integrated into the global top-tier AI computing supply chain.
The evolution from the Blackwell platform to the Vera Rubin platform has brought dual leaps in both chip power consumption and cabinet power density, compelling a deep integration of power supply and cooling technologies. As the “heart” of liquid cooling systems, the technological transformation of water pumps reflects data centers ‘stringent requirements for extreme energy efficiency, zero-leakage safety, and high space utilization. This report provides an in-depth analysis of the underlying rationale and technical mechanisms behind NVIDIA’s pioneering high-voltage DC electronic shielded pumps, as well as their profound impact on China’s water pump industry chain.
1. NVIDIA’s GTC event in Taipei officially announced the launch of its first “screening pump” product.
At a major industry event in Taipei, NVIDIA’s supply chain ecosystem unveiled for the first time its high-power electronic-shielded pump CDU solution designed for the 800 HVDC architecture. This solution has been incorporated into the DSX AI factory reference design, marking a technological shift in the global liquid-cooled CDU pump market.
The primary driver behind this upgrade is the exponential increase in AI computing power consumption. NVIDIA’s Vera Rubin platform employs a 100% liquid cooling solution, with both the entire chip and network components cooled by 45°C coolant, achieving near-zero water usage under optimal conditions and significantly improving energy efficiency. However, the maximum thermal power consumption per GPU reaches 2300 W, coupled with the cabinet’s power density escalating to 100 kW or even 1 MW, imposing stringent demands on CDU performance; this platform requires high-voltage direct power supply via a solid-state transformer (SST), making traditional industrial-frequency transformers unsuitable due to space constraints in high-density cabinets.
The limited space in AI server racks further complicates the design and deployment requirements for CDUs. When LG Electronics developed a CDU compliant with NVIDIA’s certification standards, the company faced significant challenges as NVIDIA substantially raised the cooling capacity benchmark from 600–700 kW to over 1.4 MW, delaying certification approval until the end of 2026—a clear indication of the mounting pressure on cooling infrastructure due to soaring computing density.
As the core heat exchange module linking the primary building complex circulation with the secondary cabinet internal circulation, the stability and efficiency of the CDU’s pump system directly determine the operational safety of the computing power cluster. Currently, CDUs are evolving from traditional centralized deployment to distributed in-rack configurations to enhance local temperature control accuracy and system redundancy.
Amid the dual trends of distributed deployment and high density, traditional AC centrifugal pumps—due to their large size, multiple conversion stages, and high noise levels—can no longer meet the requirements of next-generation rack-mounted CDUs. NVIDIA’s flagship solution integrates 800 HVDC direct-current power supply with high-power electronic-shielded pumps, ushering in a new era of end-to-end liquid cooling.
II. Why did NVIDIA adopt an electronic screening pump?
NVIDIA’s shift from traditional pump designs to high-pressure DC electronic shielded pumps represents not merely a component replacement, but an inevitable technological choice resulting from comprehensive trade-offs in electrical engineering, thermodynamics, and mechanical dynamics under the stringent physical constraints of high-density computing cabinets.
(1) Technical Collaboration in the 800 V High-Voltage DC Power Distribution Ecosystem
As AI server cabinets evolve toward megawatt (MW)-
級
power capacities, traditional 48V or 54V low-voltage DC power architectures face a “physical wall” due to excessive current, overly thick copper busbars, and high system losses. According to fundamental electrical principles, transmission losses are proportional to the square of the current.
At the same power level, when the system voltage reaches 800 V, the current is only one-sixteenth of that in a 48 V architecture; theoretical resistance losses in cables and transmission links can be reduced to one-two-hundred-fifty-sixth of the original value. This not only enhances end-to-end power supply efficiency by 3% to 5%, but also reduces copper consumption by over 45% and frees up approximately 30% of cabinet space.
In the 800V DC power architecture, traditional AC pumps and low-voltage DC pumps suffer from technical compatibility issues, requiring the installation of AC-DC/DC-DC voltage conversion modules on the CDU or cabinet side, which introduces additional energy losses and potential failure points. High-voltage DC electronic shielded pumps can directly interface with the 800V DC busbar, with motors receiving direct high-voltage DC power—eliminating intermediate voltage transformation stages and minimizing the power consumption of the liquid-cooled circulation system.
NVIDIA has outlined a phased technical roadmap:
During the transition period from 2025 to 2026, the UPS Upgrade Edition will be equipped with a dedicated Power Sidecar power rack to convert 480 V AC to 800 V DC for power supply.
Starting in mid-2026, standardized HVDC solutions will be deployed at scale, phasing out traditional UPS systems, with backup power supplied by BBU units on the rack side and BESS units on the grid side.
In its ultimate form, the Solid-State Transformer (SST) efficiently converts medium-voltage electricity directly into 800 V DC power, achieving a 5% higher efficiency under light loads than conventional transformers. Large data centers can save millions of kWh annually through this system. The electronically shielded pump connected directly to the high-voltage busbar serves as the core component of this energy-efficient solution.
(II) The unique physical mechanism of the shielded pump and its “zero-leakage” safety feature
Within data centers, even a minor leakage of coolant can prove catastrophic for GPU server clusters valued at hundreds of thousands to millions of dollars. Traditional liquid cooling pumps primarily face the physical bottleneck of shaft seal failure.
Conventional centrifugal pumps rely on mechanical shaft seals for leak prevention; under prolonged high-frequency operation, wear of the stationary and rotating rings is inevitable, with a service life of approximately 15,000 hours and an annual failure rate exceeding 6%. Although magnetic pumps achieve contactless power transmission through magnetic coupling, they are prone to magnetic degradation during transient pressure fluctuations or in high-temperature media environments, and their bulky size makes them unsuitable for the compact space requirements of Central Distribution Units (CDUs).
The shielded pump is a shaftless centrifugal pump in which the motor rotor is coaxially integrated with the pump impeller and submerged in a cooling medium, while the stator is physically isolated from the rotor by an ultra-thin, non-magnetic, corrosion-resistant metal shielding sleeve made of stainless steel or titanium alloy. The stator windings are completely enclosed within the shielding sleeve, allowing the fluid to flow exclusively within the sealed chamber and achieving structural zero leakage. Additionally, the cooling liquid simultaneously lubricates both the sliding bearings and the motor windings, eliminating the need for external lubrication systems and enabling a system lifespan exceeding 30,000 hours.
3) The stringent requirements of the distributed CDU architecture on pump performance
The Data Center CDU is rapidly evolving toward distributed data center or cabinet-level (in-row/in-rack) architectures, which impose stringent requirements on pump size, noise control, and control accuracy.
The high-voltage DC electric shielded pump features an integrated motor-pump design, making it suitable for compact distributed rack spaces and reducing its size by over 30% compared to conventional centrifugal pumps or magnetic pumps of equivalent power. It eliminates the friction associated with traditional mechanical seals, ensuring smooth operation with noise levels below 60 dB. The pump supports intelligent flow regulation across a wide range of 30–130 L/min, with flow variation below±2% and temperature control accuracy under 0.5°C; it dynamically adjusts cooling output in response to load changes, achieving up to a 50% improvement in overall heat transfer efficiency.
III. The Critical Significance of NVIDIA’s Use of Electronic Shielded Pumps for Domestic Water Pump Manufacturers
NVIDIA’s establishment of a dominant position in high-voltage DC electronic pumps within the liquid-cooled CDU supply chain marks a historic turning point for China’s domestic pump manufacturers, which have long competed in the mid-to-low-end market and urgently need to transition into high-value-added sectors.
(1) The liquid cooling market, valued at hundreds of billions, is experiencing explosive growth, with domestic players expected to capture a significant share of the emerging market segment.
Driven by both policy incentives and technological advancements, liquid cooling has become an essential requirement for high-density data centers. The National Data Administration mandates that the Power Usage Effectiveness (PUE) of data centers must be strictly controlled below 1.3 by 2025, which is expected to boost the penetration rate of liquid cooling in China from the current 10% to over 40%. The domestic liquid cooling market size is projected to exceed 108.2 billion yuan by 2027. Globally, the market size is estimated at 94.2 billion yuan in 2026 and 147.8 billion yuan in 2027, representing a year-on-year growth of 56.8%. As a key component, the market for data center cooling pumps (CDU pumps) has experienced explosive growth alongside this trend.
The global data center liquid cooling pump market has long been dominated by foreign giants such as Grundfos, Veolia, Xylem, Moog, Sulzer, and Colfax, collectively holding over 75% of the market share. As liquid cooling solutions shift from integrated delivery to decoupled delivery models, major cloud providers and computing power integrators increasingly bypass equipment manufacturers to independently procure core components—including cooling plates, quick connectors, manifold pipes, and CDUs—thereby mitigating supplier lock-in risks and reducing capital expenditure (CAPEX).
Currently, the cold plate module market is dominated by Taiwanese and overseas manufacturers (KooChill 54%, AVC 23%, JianCe Precision 15%, Delta 6%). However, in the critical component of power pumps, domestic manufacturers are accelerating domestic substitution through price competitiveness, a four-week delivery cycle (compared to 12 weeks for global giants), and rapid technical responsiveness, actively entering the North American supply chain.
(II) Strategic Planning and Technological Breakthroughs in Domestic Water Pump and Faucet Products
In response to the new industry trends set by NVIDIA, several leading domestic pump manufacturers and automotive thermal management innovators have already built up comprehensive product and technological reserves, precisely capitalizing on this industry opportunity.
1. Dayuan Pump Industry: A global leader in civilian and high-end shielded pumps, achieving significant breakthroughs overseas
As China’s leading manufacturer of shielded pumps, Dayuan Pump Industry’s “Xinhu” brand has specialized in pump products for 35 years. Its high-end shielded liquid-cooled pumps feature a unique “two-stage pressure-bearing structure design,” enabling zero-leakage transportation of hazardous fluids, with the technology already validated in advanced equipment sectors such as aerospace and nuclear power.
The company has launched mature products such as the CHM25 and PBEC permanent magnet high-efficiency shielded pumps, tailored for liquid cooling applications in data centers. Its self-developed 37 kW high-power electronic shielded pump has not only passed rigorous certification from North American internet giants including Google but has also secured substantial orders from clients in the region, breaking the monopoly of overseas competitors. Currently, its liquid cooling pumps have become integral parts of the supply chains of leading domestic and international liquid cooling integrators—including ZTE Corporation, Sugon Digital Innovation, Invic, Tongfei Co., Ltd., and Vidi Technology—demonstrating significant first-mover advantages.
2. SDPPUMP: A pioneering solution utilizing high-power permanent magnet technology
When NVIDIA officially announced its 800V HVDC architecture, Leo Pump Industry, leveraging its forward-looking strategy, had already established technical capabilities and commercialized high-power shielded pumps. Its LHS magnetic levitation high-speed high-pressure pump and full range of high-power permanent magnet shielded pumps integrate core magnetic levitation and permanent magnet technologies, addressing industry challenges such as high energy consumption, leakage risks, and complex maintenance in conventional pumps. These products cover flow rates from 800 L/min to 2000 L/min, perfectly meeting the cooling medium circulation requirements for ultra-large computing clusters demanding high flow rates and elevated head pressures.
3. Cross-industry strategic moves by automotive thermal management companies to transform their business models
As the value of the liquid cooling sector is being re-evaluated, domestic automotive component manufacturers are rapidly entering the data center liquid cooling market, leveraging their mass production capabilities and technical expertise in new energy vehicle thermal management solutions—including 800V high-voltage platform electronic water pumps, permanent magnet motor drives, and microchannel heat exchangers.
For example, Feilong Co., Ltd.’s liquid cooling circulation pumps, CDU pumps, and electronic shielded pumps have been supplied to mainstream system integrators such as Invensys, Delta, and Huawei, establishing themselves as key players in the computing power infrastructure sector. Leading automotive thermal management companies like Yinlun Co., Ltd., Top Group, Zhongding Co., Ltd., and Minshi Group have also accelerated their market penetration through manufacturing capabilities, creating cross-industry competitive advantages; manufacturers including AVIC Optoelectronics, Qiangrui Technology, and Chuanhuan Technology have simultaneously entered emerging segments such as quick connectors, rack manifold pipes, and piping systems, collectively building a comprehensive domestic liquid cooling supply chain ecosystem.
(3) Conclusion and Prospects
NVIDIA has deeply integrated its next-generation 800HVDC architecture CDU with high-voltage DC electronic shielding pumps, systematically redefining the three-dimensional framework of “electricity, heat, and mechanical forces” in AI data centers. This represents not merely a technical upgrade of individual components, but a fundamental paradigm shift in computing efficiency, space utilization, and operational reliability.
As data centers evolve toward hyper nodes, high-power chips, and megawatt-scale cabinets, the technological advancement of high-voltage DC shielded pumps has opened a pathway for domestic pump manufacturers and thermal management companies to achieve breakthroughs and integrate into the global top-tier AI computing power supply chain. Leveraging core strengths in 800V DC control, high-efficiency permanent magnet shielded motors, and agile manufacturing, China’s liquid cooling supply chain can, if it capitalizes on this technological turning point, gradually replace foreign brands in the global computing infrastructure landscape and establish itself as a key player in the global industrial chain.
