Why the AI Revolution Is Really an Energy Revolution
Imagine a metal box about the size of a carry-on suitcase.
Inside, processors perform trillions of calculations every second—powering artificial intelligence, discovering new drugs, designing aircraft, optimizing supply chains, and answering millions of questions.
That same box also produces roughly as much continuous heat as a household electric water heater.
Now imagine forty of those servers stacked into a single rack.
Then hundreds of racks inside a data center.
Then thousands of new data centers being built around the world to power the AI revolution.
Finally, imagine spending additional electricity to cool all of that heat—only to dump it into the atmosphere.
That is how we are building one of the largest infrastructure systems in modern history.
And I believe we’re optimizing for the wrong outcome.
Every Computation Becomes Heat
There is one law of physics that every engineer designing AI infrastructure must ultimately obey.
Every watt of electricity consumed by a processor eventually becomes heat.
Not most of it.
Not almost all of it.
All of it.
Every search query.
Every AI inference.
Every scientific simulation.
Every financial transaction.
Every image generated.
Every token produced by a large language model.
Eventually becomes heat.
Today’s data centers are designed to remove that heat as efficiently as possible.
Tomorrow’s data centers should be designed to use it.
That simple shift changes how we should think about AI infrastructure.
The Silent Externality of the AI Boom
The public conversation around AI infrastructure has focused on electricity demand, water consumption, semiconductor supply chains, and carbon emissions.
Far less attention has been paid to thermal pollution.
Yet heat is the inevitable byproduct of every computation.
Today, nearly every data center treats that heat as waste.
It is exhausted into the surrounding environment through cooling towers, dry coolers, evaporative systems, or massive banks of fans.
As artificial intelligence accelerates demand for computing, the world is preparing to invest hundreds of billions of dollars in new data centers.
According to current projections, global data-center electricity demand is expected to increase dramatically over the next decade, reaching approximately 980 TWh by 2030, according to Gartner.
That also means approximately 980 TWh of heat] will be generated every year.
Most discussions stop there.
But they shouldn’t.
Because that heat represents one of the world’s largest untapped energy resources.
Seeing the Problem at Every Scale
The scale of this opportunity is difficult to appreciate until you zoom in—and then back out.
One Server
A modern AI server occupies only a few cubic feet.
Yet it can continuously consume—and therefore reject—several kilowatts of heat.
Comparable to a household electric water heater.
Except the water heater cycles on and off.
The server runs all day.
Every day.
One Rack
Now stack dozens of those servers together.
Modern AI racks routinely operate between 50 and well over 100 kilowatts, with even higher densities on the horizon.
A single rack can produce enough continuous heat to rival dozens of household water heaters operating simultaneously.
One Data Center
Scale that to a 20-megawatt facility.
That single building continuously produces roughly 20 megawatts of thermal energy—every hour of every day.
Over a year, that’s approximately 175,000 megawatt-hours of heat.
Enough thermal energy to serve thousands of homes under the right conditions—or support hospitals, universities, food processors, district energy systems, or industrial facilities.
A Nation
Now multiply that by thousands of data centers.
The AI revolution isn’t simply creating a new computing infrastructure.
It’s unintentionally creating one of the largest new sources of continuous thermal energy humanity has ever built.
We’ve simply chosen not to think of it that way.
We’ve Been Thinking About Data Centers Incorrectly
For decades we’ve described data centers as digital infrastructure.
That description is incomplete.
Data centers are energy infrastructure.
They simply happen to perform computation.
The real design question is no longer:
“How do we get rid of the heat?”
It is:
“Who can use this heat?”
That single question fundamentally changes site selection, engineering, economics, and public policy.
Introducing Distributed Data Center Cogeneration
I call this design philosophy Distributed Data Center Cogeneration (DDCC).
Its central idea is remarkably simple.
Instead of placing data centers wherever inexpensive land and electrical capacity exist, place them where continuous thermal demand already exists.
Food manufacturers.
Cold-storage facilities.
Hospitals.
Universities.
Apartment developments.
Industrial campuses.
Greenhouses.
District energy systems.
Rather than treating neighboring facilities as unrelated buildings, DDCC treats them as components of a single integrated energy system.
The data center provides computing.
The host facility provides thermal demand.
Together they create value neither could achieve independently.
Waste Heat Isn’t Waste
Many people assume data-center heat is too low-quality to matter.
That assumption becomes less true every generation.
Liquid cooling captures heat far more effectively than traditional air cooling.
Increasing rack densities produce hotter coolant temperatures.
That heat can power absorption chillers, reducing cooling loads elsewhere.
Residual heat can produce domestic hot water.
Lower-temperature heat can support industrial processes or thermal storage.
Instead of using energy once before rejecting it into the atmosphere, DDCC allows the same unit of energy to perform multiple jobs.
Nature already produced the heat.
The only remaining question is whether we choose to use it.
How DDCC Actually Reduces Total Energy Use
The value of DDCC is not that the data center suddenly consumes less electricity to perform the same computation.
The value comes from using energy that the data center has already consumed to eliminate energy that another facility would otherwise need to purchase.
Consider a simplified 20-megawatt data center operating continuously.
Over one year, it consumes approximately:
20 MW × 8,760 hours = 175,200 MWh of electricity
Nearly all of that electricity ultimately becomes heat. But not all of the heat will necessarily be recoverable or useful.
Assume:
- 90 percent of the data center’s heat can be captured through liquid cooling.
- 85 percent of the captured heat remains useful after heat exchangers, piping and distribution losses.
- A neighboring facility can use the recovered heat throughout the year.
The amount of useful thermal energy delivered would be approximately:
175,200 MWh × 90% × 85% = 134,000 MWh of useful heat per year
Without DDCC, the data center would still consume its 175,200 MWh of electricity. The neighboring facility would then produce its heat separately.
If that facility used natural-gas boilers operating at 85 percent efficiency, providing 134,000 MWh of useful heat would require approximately:
134,000 MWh ÷ 85% = 158,000 MWh of natural-gas energy
Under the conventional model, the two facilities therefore consume:
- 175,200 MWh of electricity for computing
- Approximately 158,000 MWh of fuel for heating
- Additional electricity for the data center’s cooling equipment
Under the DDCC model, the computing electricity still performs its original job. But the resulting heat performs a second job before it is rejected.
The recovered heat can therefore eliminate most or all of the 158,000 MWh of fuel that the host facility would otherwise consume, while potentially reducing some of the electricity required to reject the data center’s heat.
The same input energy has now produced two useful outputs:
- Computation
- Heating, hot water, industrial heat or—in appropriate systems—cooling
That is the fundamental energy advantage of DDCC.
It does not violate the laws of thermodynamics or create free energy. It prevents one organization from buying new energy while another organization simultaneously pays to discard energy it has already purchased.
The Difference Between Available Heat and Useful Heat
The theoretical amount of heat produced by a data center is not the same as the amount of energy a neighboring facility can actually avoid purchasing.
A realistic DDCC assessment must account for four quantities:
Heat produced: The total heat created by the servers and supporting equipment.
Heat captured: The portion collected by liquid cooling or another heat-recovery system.
Heat delivered: The amount remaining after pumps, heat exchangers, piping and other system losses.
Heat used: The amount that coincides with actual demand at a compatible temperature.
Only the final quantity—heat actually used—should be counted as displaced energy.
If a data center produces 20 MW of heat but its neighbor needs only 5 MW, the remaining 15 MW must still be stored, transferred elsewhere or rejected. DDCC creates the greatest value when production and demand remain closely matched over many hours of the year.
Where DDCC Works Best
The best DDCC locations are not simply facilities that need heat occasionally. They are facilities with large, predictable and preferably continuous thermal demand.
A strong host generally has:
- Substantial electrical and thermal demand
- Long operating hours
- Heating, hot-water, cooling or process loads throughout the year
- A central HVAC or thermal-distribution system
- Demand at temperatures compatible with recovered data-center heat
- Physical proximity to the data center
- Enough load to use a high percentage of the available heat
- A long-term need that justifies the connecting infrastructure
Hospitals and Medical Campuses
Hospitals may be among the strongest DDCC targets.
They operate around the clock and require continuous electricity, domestic hot water, sterilization, laundry service, ventilation, reheating, humidity control and cooling. Many hospitals already operate centralized hot-water, steam and chilled-water systems.
A data center located on or near a hospital campus could provide base-load hot water, preheat boiler makeup water, support space heating and potentially contribute to chilled-water production where recovered temperatures are sufficient.
The hospital’s continuous demand reduces one of the largest risks in heat-recovery projects: having large quantities of heat available when no one needs it.
Central HVAC Plants
Buildings served by central heating and cooling plants are better targets than collections of buildings dependent on independent rooftop units.
A central plant creates a practical point where recovered heat can enter a hot-water loop, preheat return water, charge thermal storage or support compatible heat-driven cooling equipment.
The best opportunities include:
- University and college campuses
- Hospital systems
- Military installations
- Government campuses
- Large corporate campuses
- Airports
- Convention centers
- Mixed-use developments
- District energy networks
These facilities can distribute recovered energy across multiple buildings and combine different load profiles, increasing the percentage of heat that can be productively used.
Food Processing and Industrial Operations
Food manufacturers often require simultaneous refrigeration and heat.
They may use energy for washing, sanitation, pasteurization, drying, cooking, boiler makeup water and other processes while also operating substantial refrigeration systems.
This creates two potential DDCC uses:
- Recovered heat can replace boiler or water-heating energy.
- Where temperature conditions permit, it can support cooling technologies or reduce the lift required of conventional refrigeration equipment.
Other promising industrial users include laundries, paper and pulp operations, chemical facilities, textile plants, breweries, dairies and manufacturing processes requiring large quantities of low- or medium-temperature heat.
Hotels, Apartments and Residential Districts
Hotels and large residential developments create recurring demand for domestic hot water and space heating.
A single building may not absorb the output of a large data center, but a district energy network connecting apartments, hotels, offices, schools and public buildings can aggregate enough demand to create a strong thermal market.
Cold-weather locations offer particularly strong space-heating demand, although summer demand must also be considered. Domestic hot water, swimming pools, laundry facilities, thermal storage and appropriately designed cooling systems can help increase year-round utilization.
Greenhouses and Controlled Agriculture
Greenhouses can use large quantities of low-temperature heat to extend growing seasons, maintain nighttime temperatures and support year-round production.
They can be attractive DDCC partners because they may be built near the data center rather than requiring an existing urban host. Their demand, however, can vary by season, crop and climate, so the data center may still require backup heat-rejection capacity.
Cold Storage and Refrigerated Warehouses
Cold-storage facilities are attractive because they operate continuously and have significant energy demand.
However, recovered server heat does not automatically produce refrigeration. The practical opportunity depends on whether the heat can drive an absorption or adsorption chiller, regenerate a desiccant system, provide defrosting, heat water, serve office or warehouse heating, or reduce demand elsewhere in the operation.
The temperature of the recovered coolant is therefore critical. Some applications can use relatively low-temperature heat directly. Others require heat pumps or higher server outlet temperatures.
A Practical DDCC Screening Test
Before considering a location, developers should answer five questions:
- How many megawatts of heat will the data center produce during each hour of the year?
- How much of that heat can the cooling system capture, and at what temperature?
- How much thermal demand exists nearby during those same hours?
- What fuel or electricity would the recovered heat actually displace?
- How much pumping, heat-pump, storage and backup energy is required to make the system work?
A project should not be evaluated only by its annual heat production.
Two facilities might each consume 100,000 MWh of heat annually, but they may still be a poor match if one needs the heat mostly during winter while the data center produces it evenly throughout the year.
The proper calculation compares production and demand hour by hour.
The most valuable DDCC projects will be those that maximize what might be called the Useful Heat Utilization Rate:
Useful Heat Utilization Rate = Recovered heat actually used ÷ Recoverable heat produced
A project that productively uses 80 percent of its recoverable heat provides far greater energy and economic value than one that uses 20 percent, even when both data centers are the same size.
This changes the site-selection process.
Instead of first finding land and electrical capacity and then asking how to reject the heat, developers should search for locations where electricity, computing demand and thermal demand can be designed as one integrated system.
AI Makes the Business Case Better Every Year
Most infrastructure depreciates as technology advances.
DDCC behaves differently.
Each new generation of AI hardware increases compute density.
Higher compute density produces more concentrated heat.
Better liquid cooling captures that heat more efficiently.
Higher coolant temperatures make that heat more valuable.
The host facility receives increasing thermal value without expanding its footprint.
Few infrastructure investments become more valuable simply because technology advances.
DDCC is one of them.
A Rare Alignment of Incentives
The best infrastructure creates value for everyone involved.
DDCC does exactly that.
Data-center operators gain lower cooling costs, greater flexibility in site selection, improved sustainability performance, and potentially faster permitting.
Host facilities gain lower heating costs, lower cooling costs through heat-driven chilling, new lease revenue, and greater energy resilience.
Utilities benefit from reduced peak electrical demand and improved system efficiency.
Communities experience lower thermal pollution, reduced emissions, and more efficient land use.
The combined value exceeds what either organization could achieve independently.
That is the hallmark of transformative infrastructure.
We Already Know the Engineering Works
Around the world, pieces of this model already exist.
Scandinavian cities heat neighborhoods with recovered data-center heat.
Universities recover server heat for campus buildings.
Greenhouses extend growing seasons using industrial waste heat.
Aquaculture operations use recovered thermal energy to improve production.
The technology is no longer the limiting factor.
The missing ingredients are standardized commercial models, engineering frameworks, financing structures, and public policies that recognize heat as an asset instead of a nuisance.
The Next Five Years Matter
Artificial intelligence is driving one of the fastest infrastructure build-outs in history.
The site-selection decisions made over the next five years will shape cities for decades.
Every conventional data center approved today locks in another generation of wasted thermal energy.
Every DDCC deployment creates something fundamentally different:
A distributed urban energy asset.
A building that computes.
Heats.
Cools.
Stores energy.
Supports industry.
Strengthens the surrounding community.
Infrastructure lasts a long time.
The choices we make today will still be shaping our cities in 2050.
A New Way to Think About AI
Imagine building a natural gas power plant and intentionally throwing away every kilowatt of usable heat it produced.
Few engineers would consider that acceptable today.
Yet that is effectively how we design nearly every data center.
We celebrate the computation.
We discard the energy.
Artificial intelligence is often described as a digital revolution.
It is.
But it is also an energy revolution.
We are about to build one of the largest new energy systems in human history.
We simply haven’t realized it’s an energy system yet.
The next generation of data centers should not merely process information.
They should heat buildings.
Cool factories.
Support industrial processes.
Strengthen electrical grids.
Reduce urban heat.
And become an integral part of the communities they serve.
The question is no longer whether AI will reshape our infrastructure.
It already is.
The real question is whether we will build infrastructure intelligent enough to use the energy it already produces.
Written with assist from AI. My ideas,with assistance from AI in the research and writing the article.
Kevin Mireles 