Data Centre Cooling Cost Models
Executive Summary
Key Takeaways
- ✓ Cooling cost is one of the largest non-IT power draws in a data centre and the primary driver of PUE, and should be modelled explicitly as its own cost category rather than absorbed into a single blended power cost.
- ✓ Cooling technology choice, air-based, containment-assisted, or liquid cooling, carries materially different capital and operating cost profiles, and higher upfront capital cost technologies typically deliver a lower PUE and lower ongoing cooling opex.
- ✓ Climate materially affects achievable free cooling hours, periods when outside air or water temperature allows cooling with reduced or no mechanical refrigeration, and a facility's cooling cost model should use a climate-specific free cooling assumption rather than a generic industry-average figure.
- ✓ Rising rack density increases the cooling load per rack and can require a shift to a higher-capability (and typically higher-cost) cooling technology once density exceeds what air-based cooling can efficiently support.
Objective¶
This guide sets out how to model data centre cooling cost within Data Centre Financial Modelling, as a distinct cost category and the primary driver of PUE.
Cooling as a Distinct Cost Category¶
Cooling cost is one of the largest single components of non-IT power draw in a data centre and the primary driver of PUE. It should be modelled as its own cost line, with its own technology, climate, and density-driven cost behaviour, rather than absorbed into a single blended power cost assumption inconsistent with Data Centre Power Consumption Models.
Cooling Technology Choice¶
Air-based cooling typically carries lower upfront capital cost but higher PUE and ongoing cooling operating cost. Containment-assisted and liquid cooling technologies carry higher upfront capital cost but deliver a materially lower PUE and lower ongoing cooling operating cost. The model should reflect this capex-versus-opex trade-off explicitly, rather than assuming a single cooling technology and cost structure regardless of the facility's actual design.
Climate and Free Cooling Opportunity¶
Climate determines how many hours per year outside air or water temperature is cool enough to support free cooling, cooling achieved with reduced or no mechanical refrigeration. A facility in a colder or drier climate can achieve materially lower cooling cost than an otherwise identical facility in a hot, humid climate. The model should use a climate-specific free cooling hours assumption for the facility's actual location, not a generic industry-average figure.
Density-Driven Technology Thresholds¶
Rising rack density increases the cooling load per rack, and beyond a certain density threshold, air-based cooling can no longer efficiently remove the heat generated, requiring a shift to a higher-capability, typically higher-capital-cost, cooling technology such as liquid cooling. The model should trigger this technology shift, and its associated capex and opex change, explicitly once density assumptions exceed the air-cooling threshold, rather than assuming air cooling remains adequate indefinitely as density assumptions rise.
Common Construction Pitfalls¶
Cooling cost blended into a single power cost line. Obscures the technology, climate, and density-driven cost behaviour specific to cooling.
Generic industry-average free cooling hours assumption. Misstates cost for a facility in a climate materially different from the industry average.
No density-driven cooling technology threshold. Understates the capex requirement once rack density assumptions exceed what air-based cooling can efficiently support.
Recommended Practices¶
- Model cooling cost as a distinct line, linked explicitly to PUE.
- Reflect the capex-versus-opex trade-off between cooling technology choices.
- Use a climate-specific free cooling hours assumption for the facility's actual location.
- Trigger a cooling technology shift, with its associated cost change, once density assumptions exceed the air-cooling threshold.
Continue Reading¶
Related Pillars¶
Related Technical Guides¶
Related Glossary¶
How OXXON tests thisRun a free structural check with FMAE
Frequently Asked Questions
Why does cooling cost deserve its own dedicated model line rather than being part of a blended power cost assumption?
Because cooling cost is one of the largest single components of non-IT power draw and the primary driver of PUE, and its cost behaviour, technology choice, climate sensitivity, and density dependency, differs materially from a simple blended electricity rate assumption.
How does cooling technology choice affect the cost model?
Air-based cooling typically carries lower upfront capital cost but higher PUE and ongoing cooling opex; containment-assisted and liquid cooling technologies carry higher upfront capital cost but deliver a lower PUE and lower ongoing cooling opex, a capex-versus-opex trade-off the model should reflect explicitly.
Why does climate matter to the cooling cost model?
Because climate determines how many hours per year outside air or water temperature is cool enough to support free cooling, cooling with reduced or no mechanical refrigeration, and a facility in a colder or drier climate can achieve materially lower cooling cost than an otherwise identical facility in a hot, humid climate, so the model should use a climate-specific free cooling hours assumption.
How does rising rack density affect the cooling cost model?
Higher rack density increases the cooling load per rack, and beyond a certain density threshold, air-based cooling can no longer efficiently remove the heat generated, requiring a shift to a higher- capability, typically higher-cost, cooling technology such as liquid cooling, which the model should trigger explicitly once density assumptions exceed the air-cooling threshold.
Related Articles
Data Centre Financial Modelling
Data centre financial modelling is the discipline of modelling a data centre operator's revenue, cost, and capital structure from its capacity-denominated drivers, power, space, and cooling capacity, rack density, and tenant contract structure, rather than the generic market-price and headcount-growth drivers used in most corporate models, or the pure occupancy-and-lease-term drivers of conventional commercial real estate. This page is the hub for the Knowledge Centre's data centre financial modelling content: how colocation, hyperscale, and enterprise business models each require a distinct model architecture, how rack revenue and occupancy are decomposed into their separable underlying drivers, and how capacity planning and financial KPIs tie the model together, as this domain expands to cover operations, revenue, investment, and governance practice across the sector.
Power Usage Effectiveness (PUE)
Power usage effectiveness (PUE) is calculated as total facility power divided by critical IT load power, with a value approaching 1.0 indicating that nearly all power consumed is delivered to IT equipment rather than lost to cooling, power distribution, and other non-IT overhead. PUE is the standard industry measure of data centre power efficiency, and because power is typically one of the largest operating cost categories, a facility's PUE directly drives its power cost per unit of billable capacity and, in turn, its profitability.
Rack Density
Rack density measures how much power a tenant's rack draws, expressed in kW per rack, and by extension how much heat must be removed by the facility's cooling system to support it. Rising rack density, driven by higher-performance computing equipment, is the primary reason power and cooling capacity, rather than floor space, increasingly bind data centre capacity before floor space is exhausted, and density tier is a primary basis for colocation pricing.
Data Centre Power Consumption Models
Power consumption modelling forecasts a data centre's actual electricity draw and cost, distinct from the capacity planning discipline that governs how much power can be sold. This guide sets out how to model load factor, utility tariff structure (demand charges versus consumption charges), and the distinction between contracted and actual power draw, since power is typically the largest single operating cost category in this sector.