AI's Thirst for Computing Power Turns Liquid Cooling Into a Booming Business
The unglamorous business of keeping computer chips from overheating has suddenly become one of the hottest corners of China’s technology supply chain. Companies that build liquid-cooling systems for AI data centers say their order books are full through 2027, with some reporting revenue growth of more than 400 percent year on year, according to a report in Shanghai Securities News. Demand is arriving faster than factories can ship, a rare constraint in a country known for manufacturing scale.
Liquid cooling is exactly what it sounds like: instead of blowing air across racks of servers, operators pump coolant through plates mounted directly on the processors, or even immerse entire server boards in specialized fluid. Air conditioning cannot keep up with the heat generated by the graphics processors that train and run today’s large AI models. As data centers pack more of those chips into each rack, liquid systems have moved from an optional upgrade to essential plumbing.
The shift is visible in how the industry now describes itself. Once treated as a supporting player, liquid cooling is being recast as core infrastructure for the AI era, on par with power supply and networking. Chinese media reports describe a supply chain firing on all cylinders: component makers adding production lines, new entrants crowding in, and established manufacturers signing long-term contracts with data center operators.
The boom traces back to the sheer scale of AI construction in China. Major technology companies, telecom carriers and regional governments have all announced computing clusters in the past two years, and each of those facilities needs a way to shed heat. The same dynamic is playing out globally, which gives Chinese suppliers an export opportunity alongside domestic demand. Industry analysts note that Chinese firms already dominate the manufacturing of many cooling components, from pumps and cold plates to the fluids themselves.
Not everyone is celebrating. Several people interviewed by Chinese media cautioned that the field is becoming crowded, with too many players chasing the same contracts. Liquid cooling involves real engineering — leak prevention, corrosion control, compatibility with different chip designs — but the basic promise of growth has attracted companies without deep technical backgrounds. If AI infrastructure spending slows, or if standards consolidate around a few suppliers, weaker entrants could be left with idle factories.
There is also a technical race within the category. Cold-plate systems, which circulate fluid through metal plates touching the chips, are the current mainstream because they retrofit relatively easily into existing data centers. Immersion cooling, which submerges whole servers in non-conductive fluid, offers better efficiency but requires rethinking everything from maintenance procedures to building design. Which approach wins at scale remains unsettled, and suppliers are hedging by offering both.
For global readers, the story matters beyond China. AI’s physical footprint — the electricity, water and hardware it consumes — has become a central concern for technology companies and governments alike. Cooling is one of the few places where efficiency gains directly reduce both energy use and operating costs. If Chinese manufacturers can drive down the price of liquid cooling the way they did with solar panels and batteries, they could accelerate the build-out of AI infrastructure worldwide.
What happens next depends on whether demand keeps outpacing supply. For now, the order backlog tells its own story: a technology that was once an afterthought is now a bottleneck, a business opportunity and, increasingly, an industrial battleground.