Every watt of electricity that enters a data center eventually leaves as heat. For many years, this has been just an idea rather than a practical business case.

Anyone who has actually built or operated a data center understands why. It's easy to imagine a water-cooled facility heating a bathhouse, an apartment building, a greenhouse, or a factory.

But as soon as you take a look at the map, it becomes clear that many of these sites are often far away from densely populated areas because they are chasing cheap power. Unless you are close to homes, commercial activity, and district heating networks, heat is just waste that must be dealt with.

That’s why heat reuse has not scaled widely in the past. Heat reuse historically failed because of mismatches between location, scale, temperature, and demand.

Things are beginning to change

With the AI boom, we are looking at much larger, denser, and more continuous loads. While a small server room will not produce enough heat to justify a new infrastructure layer, a small city, an industrial park, a greenhouse cluster, or a district heating operator can now plan to benefit from the heat produced by AI data centers.

While training clusters will continue chasing large-scale power resources, we are beginning to see inference and edge AI are moving compute closer to users and communities, making heat reuse more practical and sustainable at scale.

Heat reuse requires three conditions: a sufficient amount of heat, appropriate temperature, and local real heat demand. Historically, at least one of these conditions was missing, but AI is beginning to change that equation.

Another major change is also occurring as more data centers embrace liquid cooling. Liquid cooling makes heat easier to capture, measure, and commercialize than traditional hot-air exhaust systems.

The industry is now gradually moving from chilled-water cooling toward higher-temperature warm-water liquid cooling. High-density AI racks need more efficient cooling, and operators want to reduce dependence on chillers and improve system efficiency.

When combined with heat pumps, 40°C (104°F) water can be useful. But it's not the same as higher-grade heat going to a greenhouse, industrial dryer, or district heating system. This is where the Bitcoin mining industry has years of practical experience.

Bitcoin mining was the first computing industry to operate for years under very challenging economic conditions. Miners had to learn early how to optimize power, flexible loads, and thermal management under extremely tight margins.

To be clear, Bitcoin mining ASICs are different from your general-purpose AI servers. They are designed to operate at higher temperatures, which is why heat recovery has advanced so much. At my company, our production systems have already reached outlet water temperatures of around 81°C (178°F). At those levels, you have to think about heat as an input resource rather than a cooling problem.

There is a great deal of know-how involved. If I had to pick one lesson we learned, it would be to plan heat reuse before construction begins. Once site design and permitting are complete, most opportunities disappear.

Before selecting a site, it's important to map heat demand. Look for greenhouses, industrial facilities, warehouses, and municipal heating networks. Small cities and industrial parks are the perfect choice. A data center cannot heat large metropolitan areas, but it can make a difference to a university district or an industrial park.

The benefits of heat reuse

Heat reuse can also improve permitting and community support.

Many local communities are currently seeing the costs of data centers before they see the benefits. They see substations, water use, traffic, and noise. Heat reuse can directly help businesses or residents lower their heating costs, making the benefit visible and giving the community a concrete return.

Heat reuse can make a difference when it comes to political durability for infrastructure. Regulation is already moving in this direction, including Germany’s Energy Efficiency Act, which introduced waste heat reuse requirements for new data centers.

I expect energy efficiency to no longer be judged only by how much electricity a data center consumes, but also by how much useful work each watt performs before it leaves the system. The next generation of metrics will look at tokens per watt, water use, load flexibility, and eventually heat reuse efficiency. Operators that monetize both compute and waste heat should have a structural advantage.

In the long term, this could even become a trend in people's houses. Routers and gaming consoles all generate heat that cannot be meaningfully integrated into a home system. Again, Bitcoin mining changed that, with household devices that compute, generate revenue, and produce steady heat. Local AI inference is likely to follow a similar path, with privacy, latency, and cost pushing AI workloads closer to the user.

AI’s larger and more stable thermal loads, liquid cooling, and the lessons from Bitcoin mining are creating a unique opportunity for operators worldwide to consider. Every data center already pays to get rid of the heat. They might as well design the site correctly from day one, and turn that cost into a product, a community benefit, and a long-term advantage.

The "AI-ready" premium won’t last forever. Markets usually cool down, and margins will compress. Heat reuse alone will not determine the success or failure of a project, but in the right conditions, it can turn a burden into durable infrastructure.