In August 2026, a Chinese power grid company built a data center inside a shipping container. The 40-foot unit housed four immersion-cooled racks, 80 kilowatts of IT load, and a closed-loop cooling tower. It was assembled in a factory, delivered to a substation in Gui'an New Area, and connected to the grid. The entire project took three months.
A comparable traditional data center in the United States takes considerably longer to build and costs significantly more per unit of capacity.
The container is an extreme example. But it illustrates a structural reality that is becoming impossible to ignore: China builds data center cooling infrastructure at a fraction of the cost of the United States, and the gap is widening as AI workloads push power densities higher.
The Capex Gap
The most direct comparison is capital expenditure per kilowatt of cooling capacity.
According to research compiled by Hansen Fluid, a Chinese thermal management company, the capital expenditure for a cold plate liquid cooling system in China runs 4,000 to 5,000 yuan per kilowatt (approximately $550 to $690). In North America, the same category of equipment costs more than 10,000 yuan per kilowatt (approximately $1,380).
For air cooling, the gap is similar. Chinese air-cooled systems cost 2,000 to 3,000 yuan per kilowatt. North American systems cost 4,000 to 5,000 yuan per kilowatt.
The disparity is not a single line item. It reflects a structural difference in market competition. As the Hansen Fluid analysis noted, "competition in China's liquid cooling market is far more intense than overseas," and "the price of comparable liquid cooling products in China is less than 50 percent of the North American market."
That competition is visible in the supply chain. Chinese liquid cooling manufacturers — Envicool, Shuguang Shuchuang, Shenling Environment, and others — compete on price, delivery time, and local service capability. North American projects rely heavily on established vendors like Vertiv, Motivair, and CoolIT, whose pricing reflects both higher labor costs and a less competitive supplier landscape.
The Operating Cost Advantage
The capital expenditure gap is only the beginning. The operating cost gap is where the economics become decisive.
Cooling systems account for 30 to 40 percent of a data center's total energy consumption, according to China's Ministry of Industry and Information Technology. In a traditional air-cooled facility, completing one unit of computing work requires an additional 0.3 to 0.5 units of electricity for cooling. With full immersion liquid cooling, that overhead drops to approximately 0.15 units.
The Power Usage Effectiveness (PUE) metric captures the difference. Air-cooled data centers typically operate at PUE 1.4 or higher. Cold plate liquid cooling achieves PUE 1.15 to 1.2. Immersion cooling reaches 1.03 to 1.05.
For a 10-megawatt IT load facility, the annual electricity savings from immersion cooling — compared with air cooling — range from 20 to 30 million yuan (approximately $2.8 to $4.2 million), according to calculations cited by Shuguang Shuchuang.
A separate industry analysis estimated that a 40-kilowatt cabinet scenario can save 237 million yuan annually in electricity costs compared with air cooling, with the incremental capital investment recovered in 10 months.
The math is straightforward. Liquid cooling costs more upfront. But the operating savings are large enough that the payback period is measured in months, not years.
The Prefabrication Advantage
The most dramatic cost reduction comes from prefabrication — building data center infrastructure in factories rather than on construction sites.
In September 2026, Xinke Intelligent, a Chinese data center infrastructure company, presented a "wind-liquid co-source" containerized data center at the China AI and Computing Industry Exhibition. The company's CEO, Wang Chong, stated that a traditional one-megawatt data center costs 17 to 18 million yuan to build overseas. The containerized version can be delivered for under 8 million yuan — a reduction of more than 55 percent.
The containerized system has a PUE of 1.2 and a Water Usage Effectiveness (WUE) of zero. It can be deployed in one day. A traditional data center takes two to three years.
The Gui'an containerized facility, developed by Guizhou Power Grid, achieved similar results. Its optimal PUE is below 1.15, meaning more than 87 percent of electricity is used for computing rather than cooling. The initial construction investment was 40 to 50 percent lower than a traditional facility, and the construction period was 80 percent shorter.
Alibaba Cloud has applied the same logic at scale. In August 2026, the company announced that its fully modular data center design had reduced construction time to 100 days and cut unit construction costs by more than 10 percent compared with the previous generation. The company's first-pass test success rate approached 100 percent.
The prefabrication model shifts work from the construction site to the factory, where labor is cheaper, quality control is tighter, and economies of scale are more achievable. The result is a cost structure that is difficult for traditional construction methods to match.
The Structural Reasons
The cost gap is not explained by any single factor. It reflects a combination of structural differences between the two markets.
Competitive intensity. China's liquid cooling market has more suppliers competing for the same projects. Envicool, Shenling Environment, Shuguang Shuchuang, Tongfei, and dozens of smaller companies compete on price and delivery. North America's market is more concentrated, with Vertiv, Motivair, and CoolIT holding significant share. More competition means lower prices.
Supply chain localization. Chinese liquid cooling manufacturers source components domestically. Cold plates, quick disconnects, CDUs, and cooling fluids are increasingly produced in China. The domestic supply chain reduces both cost and lead time. North American manufacturers rely more heavily on imported components, which adds both cost and logistics complexity.
Labor cost. Construction labor in China costs a fraction of North American labor. For a data center project that requires significant on-site installation work, the labor differential alone can account for 20 to 30 percent of the total cost gap.
Policy incentives. China's "East Data, West Computing" strategy directs data center construction to western provinces with cheap land, cool climates, and abundant renewable energy. The policy framework explicitly encourages liquid cooling and prefabrication. The national standard for new data centers in hub nodes requires PUE below 1.2 and a green power share of at least 80 percent. These requirements create a large, predictable market for liquid cooling suppliers, which drives down costs through volume.
Climate advantage. Western China's cool, dry climate reduces the need for mechanical cooling for much of the year. Ulanqab has an average annual temperature of 4.3°C and more than 200 days of natural cooling. Zhongwei has an average annual temperature of 8.8°C and more than 280 days of good air quality. These conditions allow free cooling or partial free cooling for a substantial portion of the year, reducing both capital and operating costs.
The Growing Market
The cost advantage is translating into market growth. China's liquid cooling data center market reached 15.98 billion yuan in 2025, up 45.2 percent year over year. It is projected to reach 23.25 billion yuan in 2026 and 47 billion yuan by 2028.
Globally, the liquid cooling market for intelligent computing centers is projected to reach 114.7 billion yuan in 2026, a year-over-year increase of 273 percent.
The growth is driven by the same force that is reshaping data center design everywhere: AI chips are getting hotter. The transition from air cooling to liquid cooling is no longer a choice. It is a requirement.
Nvidia's Vera Rubin platform, expected in 2027, will push rack power density to 220 to 230 kilowatts — a level that air cooling cannot handle. The company's GB200 NVL72 platform, already shipping, requires direct liquid cooling by design.
The question for data center operators is not whether to adopt liquid cooling. It is where to source it. And on that question, the cost data points in one direction.
What the Cost Gap Means
China's data center cooling cost advantage is not a temporary arbitrage. It is the result of a more competitive supplier market, a more localized supply chain, lower construction costs, and a policy framework that has deliberately created demand at scale.
The gap is large. A one-megawatt liquid cooling deployment in China costs roughly 4 million yuan in capital expenditure. The same deployment in North America costs 10 million yuan or more. The operating cost advantage is similarly large, with PUE differences translating into millions of dollars in annual electricity savings for large facilities.
For American data center operators, the implication is not that they should source cooling equipment from China. It is that the cost structure of their Chinese competitors is fundamentally different. A Chinese AI company building a data center in Inner Mongolia pays less for land, less for power, less for cooling infrastructure, and less for construction labor. The cumulative effect is a cost per unit of compute that American operators cannot match with the same technology.
The cooling system is one piece of that puzzle. But it is a piece that is getting larger as AI chips get hotter. And it is a piece where the cost gap is measured in multiples, not percentages.
Sources: Hansen Fluid "Recent Research Minutes of the Liquid Cooling Sector" (February 2026); CCID Consulting via Economic Daily (September 21, 2026); CCTV News via Yunnan Science and Technology Department (September 7, 2026); Guizhou Daily (August 11, 2026); Cailianshe (September 12, 2026); Shanghai Securities News (August 10, 2026); STL Partners via Data Centre Review (June 2026); G&D Chillers (August 2026); Dongwu Securities AI Liquid Cooling Report (August 2026); Shuguang Shuchuang via Shanghai Securities News (April 2026); 10jqka project bidding data (September 2026).
Disclaimer
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations
This analysis is based on reporting and public data available as of the article date; figures may be revised as sources update.
Forecasts from third-party analysts can change with market conditions.
Cost and pricing examples are point-in-time estimates; actual rates vary.
Country and company comparisons rely on public reporting, not operational data.
This sector moves fast; timelines and deal terms may be updated later.
Company deals and regulatory rulings may evolve; verify current status.
AI infrastructure is changing quickly; claims can become outdated soon.
Sources
- Hansen Fluid "Recent Research Minutes of the Liquid Cooling Sector" (February 2026)
- CCID Consulting via Economic Daily (September 21, 2026)
- CCTV News via Yunnan Science and Technology Department (September 7, 2026)
- Guizhou Daily (August 11, 2026)
- Cailianshe (September 12, 2026)
- Shanghai Securities News (August 10, 2026)
- STL Partners via Data Centre Review (June 2026)
- G&D Chillers (August 2026)
- Dongwu Securities AI Liquid Cooling Report (August 2026)
- Shuguang Shuchuang via Shanghai Securities News (April 2026)
- 10jqka project bidding data (September 2026).
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations: This analysis is based on reporting and public data available as of the article date; figures may be revised as sources update.; Forecasts from third-party analysts can change with market conditions.; Cost and pricing examples are point-in-time estimates; actual rates vary.; Country and company comparisons rely on public reporting, not operational data.; This sector moves fast; timelines and deal terms may be updated later.; Company deals and regulatory rulings may evolve; verify current status.; AI infrastructure is changing quickly; claims can become outdated soon.