Return-water grade: the first number that decides whether your heat has a buyer.
Every liquid-cooled data centre produces warm water. Whether that water is an asset or a disposal problem is settled by one figure that rarely appears on a datasheet: the temperature it comes back at.
Definition
Return-water grade (回水品位) is the temperature at which a liquid-cooling loop hands its heat over at the data-centre boundary. Before load profile, distance or price, it decides which off-takers can use the heat without a heat pump: greenhouses and low-temperature district heating typically need 45 °C or more; pools and heat-pump sources can take less.
Heat is not a quantity. It is a quantity at a temperature.
A megawatt of heat at 30 °C and a megawatt at 60 °C are not the same product. The first can warm a swimming pool or feed a heat pump. The second can go straight into a low-temperature district network, a greenhouse, an aquaculture farm or a laundry with little or no lift. Buyers do not want "as much heat as possible". They want the right grade, at the right time, at a temperature their own system was built for.
That is why we design to return-water grade before we design to PUE. A loop tuned for the lowest possible facility overhead often returns water too cool to sell. A loop designed to return warm water may add pump and heat-exchanger energy to the facility overhead, and can turn a disposal cost into a metered revenue line at the fence.
Temperature is the first screen, not the whole answer. Whether a buyer actually signs depends as much on load profile and seasonality, distance and pumping cost, availability and back-up, hydraulic design and heat-exchanger approach, and the commercial terms. Return-water grade decides which of those conversations are worth having.
What the grades look like
Warm-water direct-to-chip systems can be designed to return water in the 45–60 °C band under the OEM's stated inlet limits and load, and some immersion and two-phase systems are designed to return above 60 °C, depending on the fluid and its saturation temperature. Air-cooled halls often hand off heat at 20–35 °C, whether as warm air or as facility water, and so do liquid loops that are mixed down with facility water before they reach the fence. At that grade almost every buyer needs a heat pump, and the economics move to whoever owns the heat pump.
The boundary where the number is measured
European reporting rules (Commission Delegated Regulation (EU) 2024/1364, which applies to data centres in the EU with an installed IT power demand of at least 500 kW) require reused heat to be measured at the data-centre boundary, where it is handed off for use outside. Outside the EU it is not binding, but it is the clearest published statement of where the meter belongs, and it is more precise than most heat-reuse marketing. It also tells you where the industry is heading: heat reuse is moving from a claim you make to a quantity somebody meters. The return-water grade at that meter is the number that will be audited.
Three paradoxes that live in the same gap
- A low PUE does not mean low water use. Evaporative cooling is thermodynamically cheap and asks for very little electricity. The metric that improved and the resource consumed are not the same resource.
- A low site WUE does not mean low total water use. Swap the tower for a dry cooler and your meter falls. Fan energy rises. The water used to generate it lands on someone else's meter.
- Heat recovery can make PUE worse. If the export pump and heat exchanger are counted as facility energy, PUE rises, and the exported heat appears nowhere in the ratio. You improved the energy system and damaged your own headline metric. Report the Energy Reuse Factor (ERF, ISO/IEC 30134-6 and EN 50600-4-6) alongside PUE so the improvement is visible.
None of these metrics is wrong. The boundary is what is rarely stated. We state it.
Questions this page answers
Who can buy data-centre waste heat, and what temperature do they need?
Buyers are sorted by the supply temperature they need, not by how much heat you have. Greenhouses and fourth-generation district heating typically need 45 °C or more (4GDH is defined around 50–60 °C supply, up to 70 °C in winter); swimming pools, aquaculture and heat-pump sources can take less. A cold-plate loop returning around 55 °C at rated IT load can serve the lower end of that window directly, once the export heat-exchanger approach is subtracted and if the network runs at fourth-generation temperatures; warmer networks and the winter end of the range need a heat pump for the last few kelvin. A 30 °C loop needs a heat pump for almost all of them. Check your own return temperature →
Why can heat recovery make PUE worse?
PUE counts the export pump and heat exchanger as facility energy but gives no credit for the heat delivered, so a site that starts selling heat can see its PUE rise while its total energy system improves. Report the Energy Reuse Factor (ERF, ISO/IEC 30134-6) next to PUE and state the metering boundary for both. Which metrics and standards apply →
What does a 45 °C inlet liquid-cooled data centre mean?
It means the coolant entering the servers may be as warm as 45 °C, so the facility can reject heat with dry coolers in most climates for most of the year and the loop returns water warm enough to sell. The figure is the OEM's stated inlet limit at rated load; the return temperature is higher by the loop's temperature rise across the servers, which the OEM states for its rated flow. Read the 45 °C article →