China has published GB/T 48023-2026, titled Technical Specification for Data Center Cold Plate Liquid Cooling Systems, with implementation scheduled for February 1, 2027. Chinese industry coverage describes it as the country’s first national-level standard specifically for cold plate liquid cooling in data centers. An August 24 report from China Industry Development Research says the standard was jointly released by the State Administration for Market Regulation and the Standardization Administration of China. The report identifies Aite Network Energy as one of the drafting organizations involved in technical discussion, data validation and preparation of provisions. The significance goes beyond a new document number. Cold plate cooling is moving from specialized deployments into mainstream AI infrastructure, where differences in connectors, coolant loops, monitoring and operating procedures can create expensive integration risk. Sensaka’s guide to liquid cooling in data centers explains why operators have to manage flow, temperature, pressure, leak detection, pumps and coolant condition as part of the production environment.
Standardization matters when liquid cooling becomes operational infrastructure
Air cooling evolved over decades with familiar equipment classes, maintenance procedures and facility conventions. Cold plate liquid cooling is moving much faster because accelerator density is forcing operators to adopt it on compressed timelines. That speed creates a fragmented market. Different server designs can require different flow rates, temperatures, materials, connectors and coolant conditions. Facilities also have to decide where cooling distribution units sit, how loops are monitored, how leaks are detected and what happens during maintenance. A national technical specification can give designers, equipment manufacturers and operators a more consistent reference point. It does not make every implementation identical, and the available public report does not provide the full technical clauses of GB/T 48023-2026. The important signal is that cold plate cooling now has enough deployment importance to justify a formal national standard. For buyers, standards can reduce ambiguity in procurement. Requirements can be written against an agreed technical framework rather than relying entirely on proprietary vendor descriptions. That can also make acceptance testing and long-term maintenance easier to structure.
AI rack density is pushing cooling closer to the compute
The need for liquid cooling comes from basic physics. Higher server power creates more heat, and moving that heat through air becomes increasingly difficult as rack density rises. Cold plates remove heat closer to high-power components by circulating liquid through plates attached to processors or accelerators. This can reduce the amount of heat that must be handled by traditional room air systems, although most deployments still require a broader facility cooling architecture. Sensaka’s guide to data center cooling systems places direct-to-chip cooling alongside containment, free cooling and other facility approaches. The key operational point is that cooling becomes a chain. A cold plate can transfer heat efficiently, but pumps, heat exchangers, CDUs, facility water loops and heat rejection equipment still have to work correctly. As AI facilities deploy more liquid-cooled racks, these components become part of the availability path for compute. A pump failure or loss of flow can become as operationally important as a server hardware fault.
Monitoring requirements become more specialized
Cold plate environments add telemetry that many traditional monitoring stacks were not designed to treat as first-class infrastructure data. Operators may need to track supply and return temperature, differential pressure, flow rate, pump status, valve position, leak sensors and coolant quality. Those signals need to be connected with rack location, server identity, workload criticality and facility alarms. Sensaka’s guide to data center monitoring software describes monitoring across hardware, power, cooling, networks and services rather than isolating each subsystem. That becomes especially relevant with liquid cooling because thermal events can originate in several layers of the loop. A standard can help define technical expectations, but operations teams still need a data model that shows what equipment is connected to which loop and what business services depend on it. Without that context, a liquid cooling alarm becomes another isolated notification instead of an actionable infrastructure event.
Standards can reduce integration risk without eliminating design work
It would be a mistake to interpret GB/T 48023-2026 as proof that cold plate deployment is now simple. Standards provide common requirements, but facility conditions still vary. Operators have different water temperatures, redundancy targets, rack densities, floor layouts and maintenance practices. Server manufacturers may also implement cooling interfaces differently within the boundaries allowed by the standard. The useful outcome is a stronger baseline. Engineering teams can compare vendor designs against a recognized specification, identify deviations explicitly and build test procedures around known requirements. This is similar to the role standards play elsewhere in data center engineering. They reduce unnecessary variation and create common terminology while leaving room for facility-specific design.
AI data center operations now include the cooling loop
The larger trend is that cooling is moving deeper into IT operations. When racks rely on direct liquid cooling, facilities and IT teams can no longer treat thermal systems as background building services. Sensaka’s AI data center operations guide connects GPU health with power, thermal conditions, cooling loops, networking and capacity. That operating model is likely to become more common as standards make cold plate systems easier to procure at scale. The publication of GB/T 48023-2026 is therefore an industry maturity signal. Cold plate liquid cooling is becoming standardized infrastructure rather than a collection of project-specific experiments. The next challenge is operational consistency. Standards can define how systems should be designed and tested, but reliable AI capacity will still depend on whether operators can monitor the cooling path, detect degradation early and maintain every component without disrupting increasingly valuable compute.
Originally published on the Sensaka blog.
