摘要：AsianFin -- The world’s data centers are moving rapidly toward liquid cooling, as the twin pressures of artificial intelligence co

AsianFin -- The world’s data centers are moving rapidly toward liquid cooling, as the twin pressures of artificial intelligence computing and energy efficiency reshape the sector’s economics. Analysts expect the global liquid cooling market to quadruple in size by the end of the decade, with China playing a pivotal role in driving adoption.

According to Omdia Research, the global data center liquid cooling market will grow at a compound annual rate of 25% to reach $16.8 billion by 2028. In China, Kezhi Consulting projects the market will surpass 100 billion yuan ($13.7 billion) as early as 2027, as AI workloads scale and supply chains mature.

A report from CITIC Securities underscored the shift, noting that the combination of escalating computing power requirements and rising energy costs will accelerate the replacement of traditional air-cooling systems with liquid-based solutions. “Liquid cooling has moved from experimental to inevitable,” the report said.

Only a few years ago, liquid cooling was rarely more than a PowerPoint slide at industry conferences. Air-based systems, bolstered by indirect evaporative and natural cooling technologies, dominated the market. Industry executives were cautious about introducing change in data centers, where uptime and stability are paramount.

“Back then, most operators viewed liquid cooling as unnecessary risk,” one industry veteran recalled. “The sentiment was, ‘We’d rather stick with what works than introduce uncertainty.’”

That changed with the explosion of generative AI in 2022. Tools such as ChatGPT triggered a wave of new demand for graphics processing units (GPUs) and high-density servers, stretching the limits of existing cooling systems.

According to IDC, the global AI server market will reach $125.1 billion this year and is projected to surge to $222.7 billion by 2028. Generative AI servers will account for nearly 38% of the market by then.

“In China, intelligent computing power is expected to grow at a five-year CAGR of more than 46%, compared to less than 19% for general computing,” said Zhou Zhengang, Vice President of IDC China. “The data clearly shows incremental demand is coming almost entirely from intelligent computing.”

That demand is translating into unprecedented rack power density. Standard CPUs that once consumed 300 watts now draw closer to 500 watts, with some approaching 1 kilowatt. GPU-driven AI workloads push power consumption even higher.

Where a 2.5kW rack was once standard, 8kW, 20kW, or even 50kW racks are increasingly common. Traditional air-cooling systems simply cannot dissipate heat at that density without massive energy penalties.

“Liquid cooling has become the only viable option for balancing efficiency and reliability at these power levels,” said a Beijing-based data center consultant.

TrendForce estimates that liquid cooling penetration in AI-focused data centers will climb from 14% in 2024 to 33% in 2025, continuing to expand thereafter.

As liquid cooling adoption accelerates, the ecosystem is becoming more sophisticated. Upstream suppliers of coolants, pipes, and quick connectors are innovating rapidly, while downstream integrators are building out standardized cabinet and immersion systems.

One of the most active participants has been Intel, which has invested heavily in both cold-plate and immersion technologies. In 2024, Intel partnered with Green Cloud Chart to launch G-Flow immersion cabinets using ExxonMobil’s PFAS-free coolant, a hydrocarbon synthetic oil designed to be more environmentally sustainable than traditional fluorinated fluids.

Intel has also co-developed the first certified cold-plate solution for its Xeon 6 processors with Chinese partner Inovance. “When chip manufacturers endorse liquid cooling, it signals to customers that the technology is not only viable but also supported at the hardware level,” said Green Cloud Chart CEO Hu Shixuan.

Beyond product launches, Intel is attempting to reshape the ecosystem in China through its Data Center Liquid Cooling Innovation Acceleration Program. The initiative brings together suppliers across the chain—from quick connector makers to coolant formulators—to standardize interfaces and improve component reliability.

“By working with leading partners, we empower them to develop new products and materials, then connect them with OEMs and ODMs in our ecosystem,” said Du Liguang, Director of Platform Technology R&D at Intel China’s Data Center and AI Division. “It’s about building a scalable, reliable supply chain.”

Earlier this year, Intel spearheaded the creation of the Universal Quick Disconnect (UQD) Interconnect and Interchange Alliance. The group certified five local hardware suppliers, including Invt, Danfoss, Liminda Technology, BlueSky Electric, and Zhengbei Connection.

Quick connectors, essential for ensuring leak-free performance in liquid-cooled servers, have historically varied widely in design and quality. Standardization promises to reduce maintenance costs, ease future upgrades, and improve overall reliability.

“Cabinet lifecycles run eight to 10 years, while servers turn over every three to five years—or faster in the AI era,” Du said. “If internal components are standardized, operators don’t need to replace entire cabinets during upgrades, saving millions.”

To validate the durability of UQD connectors, Intel’s alliance subjected components to more than 3,000 manual connection and disconnection cycles. Automated tests proved inadequate because real-world conditions vary too much.

“We had to replicate the imperfect motions of human operators,” Du explained. “That level of rigor ensures connectors perform reliably in practice, not just in theory.”

Lanke Electric, one of the testing participants, said the comprehensive approach provided valuable design standards. “This is not just about passing a test,” a company executive said. “It delivers a blueprint for the industry to follow.”

As more components achieve standardization, analysts expect deployment costs for liquid cooling to fall. In the long term, total costs could drop below those of air-cooling systems, especially as energy efficiency gains accumulate.

“The economics are tilting decisively toward liquid cooling,” said a CITIC Securities analyst. “We’re seeing a rare case where performance, sustainability, and cost all align in the same direction.”

That could encourage broader adoption beyond AI data centers, extending to cloud platforms, enterprise workloads, and even edge computing environments.

The numbers underscore the scale of the shift. By 2027, China alone will host a liquid cooling market exceeding 100 billion yuan ($13.7 billion). Globally, the market will approach $17 billion by 2028, according to Omdia.

With AI computing demand growing at more than 40% annually, the stakes are enormous for both technology providers and data center operators. Standardization efforts like Intel’s could accelerate adoption by reducing risk and cost, creating a virtuous cycle of innovation and deployment.

“Liquid cooling is no longer an alternative—it’s a must-have,” said the Beijing consultant. “The digital economy of the future will be built on it.”

The rise of generative AI has propelled liquid cooling from a niche concept into the center of global data center strategy. With power densities soaring and efficiency imperatives mounting, industry leaders are betting heavily on liquid cooling technologies and supply chain standardization. If current projections hold, liquid cooling won’t just replace air—it will redefine the economics of the world’s data centers.

来源：钛媒体