Data Center Waste Heat Recovery: Turn Cooling Costs Into Revenue

Data center waste heat recovery is the process of capturing thermal energy rejected by IT equipment and repurposing it for district heating, industrial processes, or agriculture. Nearly 100% of the electricity a data center consumes ultimately becomes heat. A 1 MW IT load produces roughly 8,760 MWh of thermal energy per year at full utilization. Germany's Energy Efficiency Act (EnEfG) now mandates a minimum Energy Reuse Factor (ERF) of 10% for new data centers from July 2026, escalating to 20% by July 2028, making waste heat recovery a regulatory requirement rather than an optional sustainability measure.

This post covers how much heat data centers actually produce, the capture methods that make reuse viable, the EU regulatory mandates driving adoption, revenue models with a worked calculation for a 1 MW facility, real-world case studies operating at scale, and why factory-built modular data centers are structurally better suited for heat recovery than traditional builds.

How much waste heat does a data center produce?

The thermodynamics are unambiguous. The Uptime Institute's 2024 heat reuse primer confirms that virtually all electricity consumed by IT equipment converts to heat, with data transmission accounting for less than 0.1% of energy output. Every watt of compute becomes a watt of thermal energy.

For a 1 MW IT load running at full capacity, that translates to 8,760 MWh of heat per year (1,000 kW × 8,760 hours). But data centers consume more than just IT power. Cooling systems, power distribution units, UPS losses, and lighting all add overhead. At a facility-level PUE of 1.3, total energy input rises to 1.3 MW, which means total heat output climbs to roughly 11,400 MWh annually. At a PUE of 1.5, common among older enterprise facilities, the figure reaches 13,140 MWh.

This is not a small number. For context, 11,400 MWh of thermal energy is enough to heat approximately 500 to 700 European homes for a year, depending on building efficiency and climate zone.

What determines whether waste heat is usable?

Temperature is the deciding factor. Not all heat is created equal, and the distinction between low-grade and high-grade waste heat determines whether you can pipe it directly into a heating network or need to invest in heat pump infrastructure first.

Traditional air-cooled data centers exhaust air at 25–45°C. That falls below the threshold for most district heating systems. Third-generation district heating networks, still common across Germany and Central Europe, require supply temperatures of 70–100°C. To bridge the gap, operators install large-scale heat pumps that boost the temperature from cooling water returns. These systems typically achieve a Coefficient of Performance (COP) of 2.5 to 3.5, meaning every kilowatt of electricity powering the heat pump delivers 2.5 to 3.5 kilowatts of usable heat.

Direct liquid cooling (DLC) changes the equation. Cold plate systems mounted directly on CPUs and GPUs capture 50–75% of rack-level heat into liquid at 50–60°C. Single-phase immersion cooling captures nearly 100% of component heat, with coolant return temperatures exceeding 60°C. Fourth-generation low-temperature district heating networks, increasingly common in Scandinavia and new German developments, operate at 50–70°C supply with 25–35°C returns. That is a near-perfect match for liquid-cooled data center output, often requiring no heat pump at all.

The table below summarizes the relationship between cooling technology and reusable heat temperature:

Germany's EnEfG: the most aggressive heat reuse mandate in Europe

Germany's Energieeffizienzgesetz (EnEfG), which entered into force on 18 November 2023, sets binding Energy Reuse Factor (ERF) targets for all new data centers with a non-redundant nominal power connection capacity of 300 kW or more. ERF, defined per DIN EN 50600-4-6, measures the fraction of total energy consumption that is meaningfully reused outside the facility. Approximately 1,000 data centers in Germany fall under this law.

The ERF requirements for new data centers escalate on a fixed timeline:

These targets must be achieved as a sustained annual average within two years of commissioning. Non-compliance carries fines of up to €50,000–€100,000 per violation, according to legal analyses by White & Case (2023) and Mayer Brown (2024). Exemptions exist: if a municipality commits to building heat infrastructure within 10 years, or if a nearby network operator refuses an offer of heat at prime cost within six months, the obligation is considered fulfilled.

Existing data centers are not subject to ERF mandates, though they face separate PUE requirements: ≤1.5 by July 2027 and ≤1.3 by July 2030. New facilities must achieve PUE ≤1.2 within two years of commissioning.

The distinction between Germany's EnEfG and the EU's Energy Efficiency Directive (EED, 2023/1791) matters. The EED is primarily a reporting and transparency framework. Article 12 mandates annual reporting for facilities with ≥500 kW of installed IT power, covering 24 data points including waste heat utilization. Article 26(6) requires member states to ensure data centers above 1 MW total rated power utilize waste heat, but with a feasibility escape: operators may present a cost-benefit analysis demonstrating economic or technical infeasibility. Germany's EnEfG goes further by setting hard numerical quotas with no feasibility opt-out for new builds.

France enacted binding waste heat recovery obligations through Law No. 2025-391 (the DDADUE law) in April 2025, requiring data centers above 1 MW to valorize their waste heat, though implementing decrees remain pending. Denmark abolished the tax on surplus heat for certified businesses in January 2022 and is removing price caps that limited payments to heat suppliers, creating strong market incentives. The Netherlands banned new hyperscale data centers outside two designated locations in January 2024, with Amsterdam requiring waste heat integration as a condition for building permits.

Revenue calculation: what can 1 MW of waste heat actually earn?

Let's work through the numbers for a realistic 1 MW data center connected to a fourth-generation district heating network.

Assumptions:

• 1 MW IT load at 75% average utilization (realistic for enterprise and edge deployments)
• Facility PUE of 1.3
• Liquid cooling with 70% heat capture rate
• Heat sold at €7/MWh (mid-range for EU district heating contracts)

Step 1: Total facility heat output1,000 kW × 1.3 (PUE) × 8,760 hours × 0.75 (utilization) = 8,541 MWh/year of total thermal output.

Step 2: Recoverable heat8,541 MWh × 70% capture rate = 5,979 MWh/year of exportable heat.

Step 3: Direct heat revenue5,979 MWh × €7/MWh = €41,853/year in heat sales.

Step 4: Avoided cooling costsExporting heat reduces the load on chillers and cooling towers. At typical cooling energy costs of €15,000–€25,000 per MW annually, assume €20,000 in savings.

Step 5: Total annual economic benefit€41,853 + €20,000 = approximately €62,000 per MW per year.

At the premium end, the Stockholm Data Parks model, where data centers are paid based on the district heating network's avoided production cost, yields approximately €190,000 per MW annually. This figure has been reported consistently by the EU Covenant of Mayors, Stockholm Exergi, and independent analysts. The range reflects contract structure, local energy prices, and whether heat pumps are provided by the data center or the utility.

Capital expenditure for heat recovery infrastructure typically runs €400,000–€700,000 per MW on the data center side (heat exchangers, piping, controls, metering) and €600,000–€1,000,000 per MW on the district heating side (heat pumps if needed, network connection, pipeline). Published payback periods range from 2 to 8 years. Fourth-generation low-temperature connections without heat pumps sit at the favorable end. Traditional high-temperature networks requiring large heat pump installations push toward 8–14 years. Economics deteriorate rapidly beyond 3–4 km of pipeline distance to the nearest heating network.

Real-world projects operating at scale

The theory is proven. Multiple EU projects are running commercially and delivering measurable results.

Meta's Odense facility in Denmark is among the most mature implementations. Fjernvarme Fyn, the local utility, recovers heat using ammonia heat pumps that boost cooling water from roughly 27°C to 70–75°C. The system has 45 MW of heat production capacity and delivers an estimated 100,000 MWh or more annually, heating over 12,000 homes.

In Finland, a major hyperscaler partnered with Fortum to build what both parties describe as the world's largest data center waste heat recycling project, with combined thermal power of up to 350 MW. When fully operational during the 2025–2026 heating season, it will supply approximately 40% of district heating for a metropolitan area serving 250,000 users. Fortum invested roughly €225 million for heat pump plants and pipeline infrastructure. The project is expected to eliminate approximately 400,000 tonnes of CO₂ emissions annually.

Google's Hamina data center in Finland takes a different approach. A 7.5 MW heat pump plant and 1.3 km pipeline provide 40 GWh per year of district heating, covering 80% of the city's total demand, with heat supplied free of charge to the municipal utility. This is Google's first offsite heat recovery project globally.

The Stockholm Data Parks initiative, operational since 2014, has connected over 20 data center operators to Stockholm Exergi's district heating network. By 2022, recovered heat was warming 30,000 apartments, with a target of 80,000 at 40 MW of total recovered capacity.

Beyond district heating, creative applications are expanding. Green Mountain in Rjukan, Norway operates a 1.75 MW heat connection to a land-based trout farm producing 8,000–9,000 tonnes of fish annually. In the Netherlands, Blockheating deploys 200 kW containerized data centers using liquid cooling at 65°C outlet temperatures to heat 2 hectares of greenhouse per container, requiring no heat pumps. In the UK, an edge computing company places immersion-cooled micro data centers beneath public swimming pools, reducing gas consumption by 62%. Octopus Energy invested £200 million to expand the model to approximately 150 pools across the UK.

Why modular data centers are structurally better for heat recovery

Factory-built modular data centers offer measurable advantages for waste heat capture that traditional raised-floor facilities cannot replicate easily. The reasons are physical, not theoretical.

In a sealed, contained module, heat is concentrated rather than dispersed across large open halls. There is no mixing with ambient building air, no thermal leakage through suspended ceilings, and no variable airflow paths that make heat capture unpredictable. Every watt of IT heat exits through a defined set of cooling connections. That makes measurement precise and export connections simple.

Liquid cooling integrates more naturally in factory-built enclosures. Because liquids carry approximately 3,000 times the heat capacity of air per unit volume, they transport concentrated thermal energy through compact piping rather than the voluminous ductwork required for air systems. A modular data center with integrated direct liquid cooling can ship with heat exchanger connections pre-plumbed and factory-tested, reducing on-site commissioning to pipe connection and controls integration.

The HPE/Danfoss partnership, announced in June 2024, illustrates this trajectory. Their standardized 40-foot container-sized Heat Recovery Module ships factory-tested with heat transfer station, controls, and management software. Deployment timelines compress to roughly 6 months versus 18 months for conventional builds. This is not a ModulEdge-specific claim, but a category advantage of factory-built modular infrastructure: the cooling system, heat capture loop, and export connections are engineered as a single integrated system rather than assembled from separate vendor packages on a construction site.

For high-density AI inference workloads at ≥40 kW per rack, where liquid cooling becomes a physical necessity rather than an option, the heat recovery opportunity compounds. NVIDIA GPU thermal design power has climbed from 300W (V100, 2017) to 1,200W (Blackwell, 2024). Cooling these chips with liquid inherently captures higher-grade waste heat at temperatures directly compatible with fourth-generation district heating. The waste heat is not just a byproduct to manage. It is a predictable, measurable thermal output that can be contracted and sold.

What to do with this information

If you are planning a data center deployment in Germany, the July 2026 ERF deadline is less than 15 months away. For new builds, waste heat recovery must be part of the design from day one, not retrofitted later. The distance to the nearest district heating network, the network's operating temperature, and the availability of a willing heat off-taker are now site selection criteria on par with power availability and fiber connectivity.

For deployments outside Germany, the regulatory direction is clear. France's binding obligations are enacted, Denmark's market incentives are proven, and the EU's EED establishes reporting requirements that will increase transparency and public pressure on operators with zero heat reuse.

The most pragmatic first step is a feasibility assessment: map the nearest heat networks, quantify the recoverable thermal energy from your planned IT load, and model the revenue against infrastructure capital expenditure. For modular deployments, the factory integration of cooling and heat export connections can compress this timeline and reduce design risk.

Frequently asked questions

How much heat does a 1 MW data center produce per year?

A 1 MW IT load at full utilization produces approximately 8,760 MWh of thermal energy annually from IT equipment alone. Including infrastructure overhead at a typical PUE of 1.3, total facility heat output rises to roughly 11,400 MWh per year. At 75% average utilization, the figure is approximately 8,500 MWh.

What is the Energy Reuse Factor (ERF) and how is it calculated?

ERF is defined by DIN EN 50600-4-6 and measures the fraction of a data center's total energy consumption that is reused outside the facility for productive purposes. It is calculated as reused energy divided by total facility energy consumption, expressed as a percentage. Germany's EnEfG mandates a minimum ERF of 10% from July 2026 for new data centers with ≥300 kW non-redundant power capacity.

What temperature does data center waste heat need to reach for district heating?

Fourth-generation low-temperature district heating networks require supply temperatures of 50–70°C, which liquid-cooled data centers can often meet directly. Third-generation networks common in Central Europe require 70–100°C, necessitating heat pump upgrades. Heat pumps used for this purpose typically achieve a COP of 2.5–6.0 depending on the temperature lift required.

How much revenue can a data center earn from selling waste heat?

Revenue depends on contract structure, local energy prices, and cooling technology. Conservative estimates for a 1 MW data center range from €30,000 to €62,000 per year in combined heat sales and avoided cooling costs. Premium market arrangements, such as Stockholm's avoided-cost model, yield approximately €190,000 per MW annually.

Does Germany's EnEfG apply to existing data centers?

No. The ERF requirements apply only to new data centers commissioned after the respective deadline dates. Existing facilities face separate PUE requirements: ≤1.5 by July 2027 and ≤1.3 by July 2030. However, existing operators above 300 kW must report waste heat data annually.

Can data center waste heat be used for purposes other than district heating?

Yes. Operational projects across Europe use data center waste heat for aquaculture (trout farming in Norway), greenhouse agriculture (the Netherlands), swimming pool heating (UK), and industrial drying processes. The viability depends on proximity, temperature requirements of the off-taker, and contract economics. Greenhouses and aquaculture facilities require lower temperatures (20–35°C) and can often use air-cooled data center reject heat directly.

What is the payback period for data center heat recovery infrastructure?

Published payback periods range from 2 to 8 years for fourth-generation low-temperature district heating connections without heat pumps, extending to 8–14 years for traditional high-temperature networks. Key variables include pipeline distance (economics deteriorate beyond 3–4 km), local energy prices, heat purchase rates, and whether the data center or utility funds the heat pump infrastructure.

How does liquid cooling improve waste heat recovery economics?

Liquid cooling captures heat at higher temperatures (50–65°C versus 25–45°C for air), reducing or eliminating the need for heat pump investment. It also captures a higher percentage of IT heat (70–100% versus 40–60% for air cooling), increasing the volume of exportable energy. For high-density deployments above 40 kW per rack, liquid cooling is often a physical requirement regardless of heat recovery plans, making the incremental cost of heat export infrastructure minimal.

‍