AI Computing Power Contest, Next Phase: Power Competition Over, Liquid Cooling Industry Goes to War

The competitive landscape of the AI industry is quietly shifting.

In the past two years, the entire computing power sector raced frantically to secure computing resources and power supplies. Power availability once became a core bottleneck holding back the deployment of intelligent computing centers. With the large-scale installation of high-end AI chips and the popularization of high-density server cabinets, industry pain points have undergone a fundamental shift. Computing power is no longer a threshold; heat dissipation capability now defines the upper limit of AI data centers.

Currently, the global AI industry has fully entered a fierce competition centered on liquid cooling. The rivalry over computing infrastructure has extended from the chip layer to the heat dissipation infrastructure layer. As power consumption of computing resources surges, traditional air cooling has hit a physical ceiling. While NVIDIA’s Blackwell architecture chips have pushed AI computing performance to new heights, they have also brought exponentially growing heat dissipation pressure. According to the GB200 NVL72 reference architecture jointly released by Vertiv and NVIDIA, a single cabinet can reach a maximum power output of 132 kilowatts, more than ten times that of conventional server cabinets.

Traditional data centers have long relied on air cooling systems for heat dissipation. Air conditioning units and hot-cold aisle isolation could reliably meet cooling demands, featuring low operational costs and proven practicality, which worked perfectly for low-power servers in earlier years. However, this long-used cooling solution is no longer viable for high-density AI cabinets. When high-power GPU clusters run continuously, traditional fans and room air conditioners fail to remove accumulated heat in a timely manner, causing rapid temperature rises inside server rooms. Servers will automatically activate frequency reduction protection via built-in temperature control mechanisms, which directly slows down AI model training, raises costs for online inference, lowers the computing utilization rate of entire cabinets, and ultimately undermines the overall operational efficiency of intelligent computing centers.

Research firm TrendForce released data reflecting this industry shift: the penetration rate of liquid cooling in global AI data centers stood at merely 14% in 2024, and jumped sharply to 33% in 2025. Within just one year, liquid cooling has evolved from an optional auxiliary solution into a standard requirement for high-end AI server rooms.

Three Core Reasons Why AI Server Rooms Are Adopting Liquid Cooling

1. Innovation in hardware architecture brings a dramatic rise in power density per cabinet

AI computing deployment has moved beyond the operation of individual graphics cards. The industry now widely adopts integrated deployment for entire cabinets and clusters. Take NVIDIA’s GB200 NVL72 cabinet as an example: it houses 72 Blackwell architecture GPUs in a single unit. The highly integrated hardware delivers powerful computing performance while generating concentrated heat. High-density computing clusters demand heat dissipation systems with ultra-fast thermal conduction and full-range temperature control. Air cooling, inherently limited by low heat dissipation efficiency, cannot keep pace with the new generation of hardware architectures.

2. Transformation of the PUE metric: from an environmental indicator to a core factor of computing costs

Previously, the industry focused on Power Usage Effectiveness (PUE) mainly to meet green and low-carbon construction standards. In the AI era, PUE is directly linked to enterprises’ operational costs for computing resources.

Cooling systems rank as the second largest power consumer in data centers. Air cooling leads to excessive energy consumption for cooling, adding substantial electricity costs to every AI token generated. For major cloud vendors and AI technology enterprises, besides developing higher-performance AI chips, optimizing cooling systems and cutting unnecessary power consumption in server rooms have become critical measures to reduce costs and improve efficiency. A lower PUE translates to stronger cost competitiveness.

3. Heat dissipation has evolved from standalone equipment procurement to a comprehensive system engineering project

Many practitioners hold a misconception that liquid cooling is simply a replacement of cooling equipment. In fact, a complete liquid cooling system consists of cold plates, Cooling Distribution Units (CDUs), specialized coolants, pump and valve assemblies, pipeline systems, heat exchangers and full-range leakage monitoring systems. All components must work in tandem to ensure stable operation.

Currently, cold-plate liquid cooling has become the mainstream solution for large-scale implementation thanks to its low renovation difficulty and strong compatibility. Immersion liquid cooling offers superior heat dissipation efficiency, yet its widespread adoption takes time due to constraints including operational standards, material compatibility and established industry practices.

Fundamentally, air conditioners, refrigerators, cold storage facilities and data center temperature control systems all operate on the same core principle: heat transfer. Household temperature control devices can tolerate occasional downtime, but AI data centers operate 24/7 as industrial facilities, leaving no margin for error in cooling systems. A single system failure will cause server outages and computing interruptions, resulting in enormous economic losses for enterprises.

Global manufacturers are now competing on this track, where core competitiveness no longer lies merely in hardware. Well-established international temperature control enterprises have made deep inroads into liquid cooling. Overseas players including Vertiv, Schneider Electric, Eaton, ABB, Siemens and Johnson Controls leverage decades of expertise in precision temperature control to hold a major share of the early high-end market.

The domestic industrial chain also boasts solid technical foundations, rather than being late entrants. Companies such as Envicool, Shenling Environment, Jialit and Imicon have long specialized in precision air conditioners, in-row cooling systems and water chillers for server rooms, with a comprehensive portfolio of mature temperature control products. Meanwhile, KEHUA Data and KSTAR expand their business into integrated liquid cooling solutions, drawing on their strengths in UPS power supplies, cabinets, power distribution and modular data center infrastructure.

Nevertheless, the booming market does not guarantee all temperature control enterprises can benefit from the AI liquid cooling trend. Next-generation liquid cooling for high-density AI scenarios tests enterprises’ overall capabilities across the entire industrial chain, rather than the cooling capacity of a single device. Strict standards apply to every link: the compatibility of hardware components such as cold plates, pipelines and connectors, the commissioning of full-cabinet servers, real-time leakage monitoring, as well as project delivery and full-lifecycle operation and maintenance services.

The true moat of this industry is not the manufacturing capacity of individual cold plates or water pumps, but official certifications from leading cloud vendors, proven track records in large-scale intelligent computing center projects, round-the-clock system stability and end-to-end operation and maintenance capabilities. Given the huge losses caused by downtime in AI data centers, mainstream IDC service providers and cloud vendors will never entrust their core cooling systems to suppliers without experience in large-scale projects.

Challenges persist in the industry, yet the shift toward liquid cooling is irreversible.

The liquid cooling sector still faces multiple uncertainties. Unified industrial standards are constantly being updated, and diversified demands arise from different chip platforms and cabinet designs. A growing number of low-end players have triggered price wars, squeezing profit margins for hardware products. In addition, water resource regulations and environmental policies continue to influence the selection of data center cooling solutions. Despite short-term industry fluctuations, the development trajectory is clear.

Modern AI data centers are no longer simple server rooms, but high-density energy processing facilities. Chips deliver computing power, power supplies provide energy, and cooling systems sustain stable computing operations. The higher the cost of computing resources, the greater the value of heat dissipation. The first phase of AI competition centered on chips and power supplies. In the second phase, liquid cooling holds the key to winning the race for computing power.

Jiayida (Suzhou) Fluid Power Technology Co., Ltd. focuses on providing one-stop liquid cooling fluid system solutions and services for data center scenarios. The product portfolio covers OCP-standard UQD, UQDB and BMQC quick disconnect couplings, water manifold assemblies, Continental EPDM hoses and hose assemblies, FF9 full-flow double interlock quick disconnect ball valves as an alternative to FD83 series products, API heat exchangers, Gates data center-dedicated EPDM hoses, as well as WIKA pressure gauges and sensors from Germany.