How GPU server waste heat can warm buildings in 2025
- Cedric KTORZA
- 6 days ago
- 8 min read
How waste heat from GPU servers can be recovered to heat buildings is no longer theory—it’s a practical path to decarbonise heat in 2025. With AI clusters now running at high duty cycles and embracing liquid cooling, the “waste” heat they produce can be captured, lifted in temperature if needed, and reused in space heating, domestic hot water, or district heating. In this guide, we explain the end-to-end chain—capture, upgrade, distribute—plus engineering, KPIs, and how Score Group integrates energy, digital, and new tech to deliver it.
At a glance
GPU clusters are ideal heat sources: high heat density, long operating hours, and liquid cooling make heat recovery viable.
The recovery chain: capture at rack level, upgrade via heat pump when required, distribute to building or district networks.
Key metrics: track PUE alongside ERF (Energy Reuse Factor) and carbon displacement for real benefits.
Integration patterns: building heating, DHW preheat, and export to district heating with hybrid/back-up systems.
Score Group unites data center, energy, and smart control expertise to design, integrate, and operate robust heat reuse solutions.
Why GPU clusters are ideal heat sources in 2025
Heat density and utilisation fit heat reuse
Modern GPU servers concentrate significant thermal loads in compact footprints. Unlike sporadic workloads, AI training and inference drive sustained utilisation, making heat output predictable and continuous—exactly what heating systems need. The International Energy Agency notes that data center electricity demand is rising rapidly with AI growth (IEA, 2024), increasing the opportunity—and responsibility—to reuse heat where feasible. Continuous, high-grade heat makes GPU clusters particularly attractive for building heating and hot water preheating.
Source: IEA on data centres and networks
From air to liquid cooling: enabling higher-grade heat
Heat reuse improves markedly when moving from air-cooled rooms to liquid-cooled racks. Liquid removes heat closer to the source, enabling higher coolant return temperatures and easier coupling to hydronic systems. Industry adoption of rear-door heat exchangers, direct-to-chip cold plates, and immersion cooling is accelerating (Uptime Institute, 2023; OCP ACS). These technologies unlock return water temperatures often in the 30–45°C range—and with immersion or tailored loops, higher—reducing or eliminating the temperature lift required from heat pumps.
Sources: Uptime Institute Annual Survey 2023, Open Compute Project – Advanced Cooling Solutions, ASHRAE TC 9.9 resources
The heat recovery chain: from rack to radiator
1) Capture: rack-level cooling that feeds hydronics
Rear-door heat exchangers (RDHx): Retrofit-friendly; remove heat at the rack with chilled or tempered water.
Direct-to-chip liquid cooling: Transfers heat from CPUs/GPUs to a coolant loop, enabling higher return temps.
Immersion cooling: Submerges servers in dielectric fluid; often achieves the highest and most stable outlet temperatures.
Each approach uses plate heat exchangers to isolate IT coolant from building water and includes leak detection, flow/temperature sensors, and redundancy.
2) Upgrade: heat pumps for temperature lift
When building circuits or domestic hot water require 50–60°C or higher, a heat pump raises coolant temperature with high efficiency. Typical seasonal COPs of 3–5 are reported depending on lift and source temperature (IEA, 2023). Minimising temperature lift through good liquid-cooling design yields better COP and lower electrical overhead. For mixed-use buildings, cascading heat pumps can supply multiple setpoints (e.g., 60°C DHW and 35–45°C space heating).
Source: IEA – Heat pumps
3) Distribute: integrate with building or district networks
Recovered heat connects to:
Building hydronic loops for space heating via fan coils/air handlers or radiant systems.
Domestic hot water preheat with stratified buffer tanks.
District heating export via heat interface units and metering.
Critical elements are hydraulic separation, buffer volume for demand smoothing, smart controls, and safe operating envelopes for both IT and building systems.
Integration patterns for buildings and campuses
Direct building heating
For offices, labs, and mixed-use buildings, GPU heat can offset gas or electric boilers during heating seasons. Low-temperature heating (e.g., 35–45°C) suits modern fan coils and radiant floors; a moderate lift via heat pump covers legacy emitters. Domestic hot water preheat is often the most efficient first step, as it requires year-round load and can accept variable inlet temperatures.
District heating export
Where district heating exists, data centers can export surplus heat via heat exchangers and heat pumps. A well-known example is Meta’s Odense facility in Denmark, designed to supply significant heat annually to the local network, demonstrating technical and commercial feasibility for hyperscale-class reuse.
Reference case: Meta – Odense heat recovery, Euroheat & Power resources
Seasonal balance and hybrid operation
Heat output from IT may exceed winter demand or be underutilised in summer. Hybrid control pairs:
Free/reject cooling when there’s no heat sink,
Thermal storage for short-term shaving,
Auxiliary boilers or chillers as back-up,
Export to district networks where viable.
Well-designed controls maintain IT thermal budgets first, then optimise for heat reuse.
Sizing and engineering considerations
Heat balance and simultaneity
Calculate hourly profiles: IT heat availability vs. building heat demand (space + DHW).
Determine base vs. peak: Right-size heat pumps and buffers to capture high-utilisation hours without overbuilding.
Consider future IT growth, GPU refresh cycles, and redundancy (N+1) in both IT cooling and heat pumps.
Water quality, safety, and reliability
Use closed loops with appropriate water treatment; employ double-wall plate heat exchangers where required.
Add leak detection, dripless quick-connects, and containment for serviceability.
Ensure fail-safe modes: if heat sink fails, cooling automatically diverts to heat rejection to protect IT.
Guidance: ASHRAE thermal guidelines, OCP ACS best practices
Regulation, metering, and reporting
Include calibrated metering for thermal energy (kWhth), electricity to pumps/heat pumps, and temperatures/flows for verification.
In the EU, evolving efficiency and data center transparency requirements encourage heat reuse; keep designs audit-ready.
Policy context: EU Energy Efficiency Directive
KPIs that matter: quantify performance and benefits
Beyond PUE: add ERF to the scorecard
Power Usage Effectiveness (PUE) remains valuable, but it doesn’t credit reuse. The Energy Reuse Factor (ERF) measures the fraction of total energy exported for beneficial use. Track both PUE and ERF to understand overall efficiency. For transparency, report where the heat goes and any temperature lifts applied.
Background: LBNL – Energy Reuse Factor overview
Carbon impact: what heat are you displacing?
The real climate benefit depends on the carbon intensity of the displaced heat (e.g., gas boiler vs. district heat mix) and the electricity used by pumps/heat pumps. Use location-based or market-based emissions factors consistently, and consider marginal grid intensity for heat pumps during peak hours. Publish a clear methodology to avoid double counting.
How Score Group delivers heat reuse end-to-end
Where efficiency meets innovation.
As a global integrator, Score Group unites energy, digital infrastructure, and new technologies to turn GPU heat into a dependable, monitored heat resource. Our approach aligns three complementary divisions:
Noor ITS – The digital backbone for reliable cooling
Designs and optimises data center infrastructure, from networks to resilient power and cooling.
Engineers liquid-cooling topologies (RDHx, direct-to-chip, immersion) and integrates BMS/DCIM.
Implements cybersecurity and monitoring to keep thermal operations secure and observable.
Noor Energy – Intelligent building energy integration
Connects recovered heat to building systems: GTB/GTC, hydronic loops, DHW preheat, and storage.
Specifies and commissions high-efficiency heat pumps and control logic for optimal COP and comfort.
Orchestrates renewable inputs (e.g., solar) and hybrid operation with existing boilers/chillers.
Noor Technology – Smart control and predictive optimisation
Uses IoT sensors and real-time analytics to balance IT cooling, heat pump operation, and building demand.
Applies AI-driven forecasting to improve ERF, reduce peaks, and maintain SLAs.
Develops custom applications and dashboards for operational visibility and reporting.
Learn more about our integrated approach at Score Group.
Case examples and references to learn from
Hyperscale to district heating: Meta’s Odense project demonstrates large-scale export with heat pumps, validating technical feasibility and multi-stakeholder coordination. Read more.
Industry guidance: ASHRAE TC 9.9 and OCP ACS provide design envelopes and best practices for liquid cooling and heat reuse integration. ASHRAE TC 9.9, OCP ACS.
Policy and markets: EU and national programmes increasingly recognise waste heat reuse in efficiency planning, reporting, and district energy development. EU Energy Efficiency Directive, IDEA.
Heat recovery options for GPU clusters: indicative integration fit
Notes: Temperatures are indicative and depend on design, flow rates, and vendor specs. Always validate envelopes with equipment manufacturers and standards.
Practical steps to get started
Baseline: Measure IT heat profile, cooling topology, and building heat demand by hour/season.
Concept design: Select capture method, hydraulic separation, and heat pump sizing based on simultaneity.
Controls: Define priorities (IT protection first), setpoints, and failure modes; integrate BMS/DCIM.
Pilot: Start with DHW preheat or a single building loop; validate metering and KPIs (PUE, ERF, COP).
Scale: Extend to multiple buildings or district export, adding storage and advanced optimisation.
FAQ
Can GPU server heat be reused without a heat pump?
Yes—if your liquid-cooling loop returns water warm enough for the target use. For example, low-temperature space heating or DHW preheat can accept 30–40°C feeds in preheat stages. Immersion and well-designed direct-to-chip systems may reach higher return temperatures, enabling partial direct use. However, many building systems and DHW setpoints need 50–60°C, particularly for hygiene standards, so a water-to-water heat pump is often added to lift temperature efficiently while maintaining high COP.
How does liquid cooling change the business case for heat reuse?
Liquid cooling captures heat closer to the source, increasing return water temperatures and raising the share of recoverable energy. That reduces the temperature lift required from heat pumps, improving seasonal COP and lowering electrical overheads. It also simplifies hydraulic coupling to building or district networks via plate heat exchangers. In parallel, liquid cooling can improve rack density and thermal stability. Together, these factors typically raise ERF and the proportion of building heat covered by IT waste heat.
What happens in summer or when the building doesn’t need heat?
Control logic prioritises IT safety, so if there’s no available heat sink, the system diverts to conventional heat rejection (dry coolers, towers). Options to improve year-round utilisation include DHW preheat, exporting to district heating, and adding short-term thermal storage. Some sites also shift compute (where SLA allows) to align with local heat demand. The design should include clear switchover criteria, buffer capacity, and metering to verify performance across seasons.
Is heat reuse compatible with reliability and SLAs?
Yes, provided the design maintains fail-safe cooling paths. Good practice includes hydraulic separation via plate heat exchangers, N+1 pumps and heat pumps, leak detection, and automatic bypass to heat rejection if the heat sink is unavailable. Controls should keep IT cooling independent from building-side faults. Rigorous commissioning and continuous monitoring (BMS/DCIM) help ensure setpoints are maintained and service levels are preserved, even as heat reuse operates opportunistically behind the scenes.
What KPIs should we track to prove benefits?
Track PUE for total efficiency, ERF for the fraction of energy beneficially reused, and heat pump COP to understand lift efficiency. Add thermal energy delivered (kWhth), temperature/flow logs, and the carbon intensity of displaced heat (e.g., gas boiler vs. district mix). Use consistent emissions factors and document methodology. With these metrics, you can demonstrate both energy and carbon benefits, compare scenarios, and guide optimisation over time.
Key takeaways
GPU clusters are excellent, steady heat sources; liquid cooling makes their heat easier to capture and reuse.
The capture–upgrade–distribute chain enables practical building heating, DHW preheat, and district export.
Design for reliability first, then optimise for ERF, COP, and carbon displacement with metering and controls.
Start small (e.g., DHW preheat), validate performance, then scale across buildings or to district heat.
Score Group integrates energy, digital infrastructure, and new tech to deliver robust, measurable heat reuse.
Ready to explore a feasibility study or pilot? Visit Score Group to connect with our teams.
