DataCentreNews UK - Specialist news for cloud & data centre decision-makers
United Kingdom
Schneider Electric backs liquid cooling software for AI

Schneider Electric backs liquid cooling software for AI

Wed, 18th Nov 2026 (Today)
Joseph Gabriel Lagonsin
JOSEPH GABRIEL LAGONSIN News Editor

Schneider Electric has outlined a software-led approach to managing liquid cooling in AI data centres, treating cooling as a coordinated system rather than a set of separate assets.

Its position centres on what it calls liquid cooling orchestration: a supervisory software layer designed to manage cooling infrastructure across AI facilities as rack power densities rise and liquid cooling becomes more common.

The company argued that operators now face the practical challenge of matching cooling systems to fast-changing AI workloads without wasting energy or water. It said orchestration software can help facilities maintain thermal headroom, improve the use of cooling capacity and support GPU performance.

In Schneider Electric's description, AI factories often rely on a hybrid thermal design that combines liquid-to-chip cooling for high-density AI systems with more conventional air cooling for building infrastructure and other IT equipment. This can include coolant distribution units, pumps, liquid distribution networks, computer room air handlers, fan walls, chillers, dry coolers, cooling towers and adiabatic heat rejection systems.

These systems do not all respond at the same speed. AI workloads can shift rapidly, but the thermal and mechanical equipment that supports liquid cooling tends to operate over slower timescales. That creates a need for higher-level coordination rather than simple reactions to every change in chip activity.

Supervisory layer

Under this model, local controls would still manage equipment such as coolant distribution units within expected operating ranges. Orchestration software would sit above them and adjust setpoints across the broader cooling chain. Schneider Electric said this approach would let operators coordinate chillers, heat rejection equipment, pumps and airflow systems using both historical and real-time operating data.

The aim is to ensure enough cooling is available before sustained heat loads affect the coolant loop. Predictive analytics could be used to anticipate thermal demand, flag abnormal conditions and suggest supervisory changes that improve overall facility efficiency.

Schneider Electric linked the approach to efforts to improve power usage effectiveness and water usage effectiveness, two measures data centre operators increasingly track as AI infrastructure expands. It also said better visibility into cooling requirements could help data centre owners make more accurate decisions on facility design, construction and infrastructure investment.

That includes estimates of likely power and water demand for cooling in different locations. According to the company, a more detailed view of operating requirements could help reduce unnecessary capital spending and lower operating costs.

Mixed environments

A key issue for operators is the growing number of mixed air and liquid environments inside AI facilities. Schneider Electric said orchestration software can ingest telemetry from both types of cooling infrastructure and present a unified operational view, allowing coordinated control across systems that may support several server platforms with different thermal profiles.

The company also made the case for multi-site management. As AI data centre fleets expand across regions, cooling optimisation is becoming a fleet issue rather than a single-site engineering task, it said. A centralised view of thermal performance, infrastructure utilisation, energy efficiency and water consumption could allow operators to compare sites, identify weaker assets and standardise operating practices.

That argument reflects a wider shift in the data centre sector as operators move from conventional enterprise facilities to larger AI-focused campuses with denser racks and more complex power and thermal requirements. Cooling, once managed largely as a local facilities function, is becoming more tightly linked to IT workload planning and hardware performance.

Link to compute

Schneider Electric said the software can also be tied more directly to compute behaviour. In its example for NVIDIA GPUs, orchestration takes account of MAX-LPS, MAX-P and MAX-Q profile configurations when assessing how much cooling capacity is needed to support the required level of performance.

The company said the software can monitor GPU and rack power telemetry as an early indicator of heat load and correlate that information with manifold return temperature and flow. Predictive models can then recommend, or automatically make, adjustments to coolant distribution units and chilled water plant settings before thermal demand rises.

Schneider Electric also argued that managing the full heat transfer chain as a single system could let operators run warmer supply temperatures, increase the number of economiser hours, reduce pumping energy and recover unused cooling capacity without undermining thermal resilience or GPU output.

The company said: "Managing the full heat transfer chain as one system, from the cold plate to heat rejection, lets operators run warmer supply temperatures, gain more economizer hours, cut pumping energy, and recover stranded cooling capacity without compromising thermal resilience or GPU performance."