Summary
- Georg Fischer reported a year-on-year doubling of data centre liquid-cooling orders during the first half of 2026.
- Its systems combine polymer pipework, valves, sensors, controls, and prefabricated modules for direct-to-chip cooling.
- Larger deployments are placing greater weight on materials, joints, leakage control, interoperability, commissioning, and long-term serviceability.
Georg Fischer says orders for its data centre liquid-cooling systems doubled year on year during the first half of 2026, as direct-to-chip designs move beyond isolated trials and into larger procurement programmes.
The Swiss industrial supplier is assembling a portfolio around polymer pipework, control valves, sensors, prefabricated modules, and connection components. Several proof-of-concept projects are under way, while discussions with hyperscale and colocation customers have progressed into a wider order pipeline.
Its half-year report shows broader growth across the Industry and Infrastructure division, where order intake reached CHF1.095bn, compared with CHF889m a year earlier. Data centre cooling sits within that division rather than appearing as a separately disclosed revenue line.
Pipework joins the critical compute chain
Direct-to-chip cooling transfers heat through cold plates mounted close to processors and memory, reducing dependence on server fans and room-level airflow. Once fluid enters the rack, pipes, hoses, couplings, valves, sensors, and water-quality controls become part of the critical path supporting the compute load.
Georg Fischer’s LiquidCore system uses polymer pipework and prefabricated assemblies in the secondary cooling loop. Lower weight, corrosion resistance, and faster installation are among the supplier’s stated advantages over conventional metallic systems, although each project still has to assess pressure, temperature, fire performance, mechanical protection, and maintenance access.
Material selection reaches far beyond installation cost. A leak inside an operating hall can affect power distribution, servers, flooring, controls, and customer confidence, so designers must account for joint integrity, thermal expansion, pressure transients, chemical compatibility, permeation, accidental damage, and repeated operating cycles.
Factory-built assemblies can reduce cutting, welding, and site work, particularly where hundreds of racks follow a repeatable design. That advantage depends on stable dimensions and connection standards; late changes to server hardware or rack layouts can turn a prefabricated system into a sequence of field modifications.
Orders do not equate directly to commissioned capacity. Liquid-cooling programmes commonly move through evaluation, proof of concept, framework agreements, site orders, installation, flushing, pressure testing, water treatment, and operational acceptance before customer workloads begin producing heat.
Interoperability will govern deployment speed
Cooling interfaces remain less consistent than the air systems used across conventional data halls. Server manufacturers specify different cold plates, couplings, manifolds, flow rates, pressure limits, fluids, and temperature ranges, while facility systems must connect those requirements to cooling distribution units and external heat rejection.
Operators increasingly want mechanical infrastructure that can accept hardware from several suppliers without rebuilding the secondary loop. Component manufacturers therefore face closer examination of interoperability, pressure containment, leakage detection, isolation, monitoring, and the availability of replacement parts.
Controls carry equal weight. Flow and temperature must follow rapidly changing server loads without creating instability across the loop, while valves and sensors need to feed usable information into facility monitoring, alarm management, and maintenance systems.
Higher coolant temperatures can reduce mechanical refrigeration in suitable climates and can improve the quality of recoverable heat. Neither outcome is automatic: server inlet limits, outdoor conditions, redundancy design, pumping energy, and the temperature required by a local heat customer will determine the result.
Purpose-built halls can allocate risers, drainage, floor loading, leak detection, and maintenance space from the outset. Retrofitting a live facility requires new distribution equipment and pipe routes to pass through operating areas without compromising customers, fire separation, access, or existing cooling.
Water quality will also influence reliability. Corrosion, biological growth, particulate contamination, and incompatible materials can damage cold plates or restrict flow, so the secondary loop needs defined treatment, sampling, filtration, and maintenance procedures.
Georg Fischer’s order growth shows mechanical infrastructure taking a larger share of the AI data centre bill of materials. Commissioned megawatts, operating reliability, service requirements, and the emergence of common connection standards will provide a clearer measure of how far the market has progressed.
