IT Home reported on November 11, 2026, that IDC said China’s liquid-cooled server market has grown rapidly over the past two years.
IDC’s China liquid-cooled server tracking puts vendor revenue in the first half of 2026 up 153.1% year over year. Over the next five years, shipments are expected to grow at a compound annual rate approaching 49%.
IDC said the procurement shift may matter more than the growth rate. More high-density computing projects are no longer choosing equipment by the individual server. Compute nodes, links between nodes, power supply, and in-rack liquid cooling are being configured and delivered as one system. The cabinet is moving from a physical container in the machine room to the unit in which computing is purchased and delivered.
When a full rack is the delivery unit, the server vendor supplies an integrated system. The data center then takes on a load with defined requirements for power, cooling, and interfaces, rather than a set of separate machines.
The report ties the market’s growth to computing demand from artificial intelligence. Internet companies are the largest buyers, accounting for 65.6% of China’s liquid-cooled server market. Telecom operators and IT services are also major application areas. Some traditional-industry customers have not adopted liquid cooling, mainly because their current work is general-purpose computing that air cooling can already handle. Those industries are exploring liquid-cooling approaches suited to their own operations.
IDC’s quarterly tracking of China’s server market projects shipment growth of about 6.8% a year over the next five years, while revenue is expected to grow at a compound annual rate of 24.6%. Moderate growth in units alongside much faster growth in revenue indicates that added value is concentrating in high-configuration, high-density systems, which is also where liquid cooling is expanding.
Cold-plate liquid cooling remains the main technical route, at about 90.2%, supported by a mature supply chain, cooling capacity, and cost advantages at scale.
IDC also described a change in the mix of AI workloads:
- Early deployments centered on large-model training, with large and concentrated clusters.
- As models move into use, the share of inference continues to rise. Online inference is closer to end users and requires lower latency and continuous availability.
- In agent-style applications, one user request often triggers multiple rounds of model and tool calls, multiplying inference demand.
IDC statistics put whole-cabinet delivery at 47.4% in North America in the first half of 2026, compared with 5.0% in China. In China, tighter availability of high-end accelerator cards is an additional reason buyers use supernode and full-rack configurations to secure computing capacity. Whole-cabinet delivery is therefore partly a path under supply constraints, not only an efficiency choice.
Original report: IT Home
