Energy Source: Thermal Integration
Cooling is the second largest energy load in AI data centers after IT power. Integrating thermal energy systems with the electrical and DER portfolio improves efficiency, resilience, and sustainability. Advanced campuses now use district cooling, thermal storage, and waste-heat reuse to reduce peak loads and align with energy autonomy strategies.
Overview
- Purpose: Remove heat from IT equipment while minimizing electrical overhead and enabling reuse of waste heat.
- Scale: Cooling can represent 25–40% of total facility energy consumption.
- Integration: Links to CHP, absorption chillers, thermal storage, and district energy networks.
- Drivers: AI workload density, sustainability goals, and urban regulations on waste-heat recovery.
Architecture & Design Patterns
- Chilled Water Plants: Electric or absorption chillers supplying CRAHs/CRACs, CDUs, or rear-door HX units.
- Liquid Cooling: Direct-to-chip (D2C) and immersion systems reduce reliance on traditional air cooling.
- Absorption Chillers: Driven by CHP waste heat, reducing electrical cooling demand.
- Thermal Storage: Chilled water or ice tanks shift cooling load off-peak; duration typically 4–8 hours.
- District Cooling: Shared utility-scale chilled water networks serving multiple facilities.
- Waste Heat Reuse: Exported to district heating, greenhouses, or industrial processes.
- Digital Twins: CFD models simulate airflow, liquid loops, and thermal storage dispatch strategies.
Waste Heat Reuse
AI data centers produce massive amounts of low-grade heat. While most facilities reject this heat through cooling towers or dry coolers, advanced campuses capture and reuse it to offset external energy needs. Waste heat reuse is most viable in urban regions with district heating networks or colocated industries.
- District Heating: Exporting 40–70°C water into municipal heating loops, common in Northern Europe (Denmark, Finland, Sweden).
- Industrial Integration: Supplying heat to nearby factories, aquaculture, or food processing plants.
- Greenhouses & Agriculture: Low-grade heat supports year-round cultivation and reduces fossil fuel heating.
- Aquatic Applications: Warm water used for fish farming or algae growth systems.
- Combined Systems: Pairing CHP with waste heat reuse creates electricity + cooling + exportable heat.
- Digital Twins: Simulate thermal flows and external demand patterns to size exchangers and interconnects.
Challenges
- Temperature Quality: Server waste heat is often low-grade and may require heat pumps for district use.
- Infrastructure: Requires physical pipelines and agreements with municipalities or industrial partners.
- Geography: More feasible in colder climates where heating demand is year-round.
- Business Models: Revenue sharing or tariff structures must be negotiated with utilities or city governments.
Examples
- Stockholm Data Parks (Sweden): Multiple operators supply waste heat into city heating grids.
- Facebook (Meta) Odense, Denmark: Reuses data center heat for district heating serving thousands of homes.
- Amazon (Ireland, UK): Pilots with heat reuse for district networks near urban campuses.
Bill of Materials (BOM)
| Domain | Examples | Role |
|---|---|---|
| Chillers | Trane CenTraVac, Carrier AquaEdge, York YMC2 | Primary cooling generation |
| Absorption Chillers | Broad, Thermax, York absorption | Cooling from CHP waste heat |
| Thermal Storage | Calmac ice tanks, chilled water reservoirs | Load shifting, peak shaving |
| Liquid Cooling Systems | Asetek D2C, Submer immersion, CoolIT CDUs | Efficient server- and rack-level heat removal |
| Distribution | Pipes, pumps, valves, manifolds | Circulate chilled water/liquid to racks |
| Heat Reuse Systems | Plate heat exchangers, district energy tie-ins | Export captured waste heat to external users |
| Controls | BMS integration, PID loops, predictive optimization | Maintain setpoints, reduce energy use |
Key Challenges
- Density: AI racks exceed 50–100 kW; liquid cooling adoption is accelerating.
- Integration: Coordinating chillers, storage, and DER dispatch adds complexity.
- Reliability: Leaks or contamination in liquid loops can impact IT equipment.
- Urban Deployment: District cooling and heat reuse require municipal agreements and infrastructure.
- Capex vs Opex: Absorption chillers and thermal storage reduce operating costs but require higher upfront investment.
- Carbon Accounting: Reusing waste heat earns credits but requires rigorous metering and reporting.
Vendors
| Vendor | Solution | Domain | Key Features |
|---|---|---|---|
| Trane | CenTraVac, ice storage solutions | Chillers / Storage | High-efficiency chillers, modular storage |
| Johnson Controls (York) | YMC2 chillers, absorption chillers | Cooling | Oil-free centrifugal, waste heat integration |
| Carrier | AquaEdge, AquaForce chillers | Chillers | Variable-speed, low-GWP refrigerants |
| Asetek / CoolIT | Direct-to-chip liquid cooling | Server / Rack | Efficient, rack-level heat removal |
| Submer | Immersion cooling tanks | Rack | Immersion-cooled servers for >100 kW racks |
| Thermax / Broad | Absorption chillers | CHP Integration | Heat-driven cooling from turbines/engines |
| Siemens / ABB | District cooling tie-ins, BMS integration | Controls | Smart building + energy management platforms |
Future Outlook
- Liquid Cooling Mainstream: Direct-to-chip and immersion systems becoming baseline for AI racks.
- Thermal-Storage Scaling: Multi-hour ice and chilled water storage standard for peak shaving.
- Waste Heat Valorization: Expanded partnerships with municipalities and industrial clusters.
- Zero-Liquid-Discharge Plants: Integration of water reuse and advanced treatment to minimize withdrawals.
- AI-Driven Optimization: Digital twins + AI control loops predict and adjust cooling setpoints dynamically.
FAQ
- How much energy does cooling consume? Typically 25–40% of total facility energy use; lower with liquid cooling and thermal storage.
- What’s the difference between CRAH, CDU, and immersion? CRAH handles room air; CDU circulates liquid to racks; immersion submerges IT directly.
- Why use absorption chillers? They convert CHP waste heat into cooling, reducing electric chiller loads.
- Is waste heat reuse practical? Yes in urban sites near district heating, greenhouses, or industrial users; harder in remote campuses.
- Do digital twins help cooling? Yes, they simulate airflow/liquid loops, optimize storage dispatch, and predict failure risks.