Every Watt Twice: The Economics of Captured Heat

Nearly every watt that goes into a data centre comes back out as heat. That's not an inefficiency to be engineered away, it's physics. Servers convert electricity into computation and waste heat in roughly equal measure, and short of a breakthrough in reversible computing, that ratio isn't moving. For most of the industry's history, the story ended there: heat was the exhaust, and the only question was how cheaply you could get rid of it.

That story is now out of date. In Odense, Denmark, the utility Fjernvarme Fyn draws heat from Meta's data centre, lifts it from around 27°C to 70–75°C with ammonia heat pumps, and pipes it into homes across the city — more than 100,000 MWh a year, enough to heat over 12,000 households. In Finland, a hyperscale operator working with the utility Fortum has built what both companies describe as the world's largest data centre heat recovery project, with combined thermal capacity up to 350 MW, supplying close to 40% of district heating demand for a metropolitan area of a quarter of a million people. Neither project treats heat as a byproduct. Both treat it as a second product line.

The regulation is catching up with the economics:

Germany's Energy Efficiency Act now requires new data centres above 300 kW to hit a minimum Energy Reuse Factor of 10% from July 2026, rising to 20% by 2028. Under the EU's recast Energy Efficiency Directive, any facility above 1 MW must formally assess the feasibility of waste heat recovery before commissioning, and increasingly, regulators are rejecting "not economically feasible" as an answer if a district heating connection exists within a few kilometres. France, Denmark, Sweden and the Netherlands all have their own versions of the same requirement arriving through the late 2020s.

For an operator, this changes the calculation. Reject heat used to be a cost to minimise. It's becoming a compliance obligation to manage — and, increasingly, a revenue line to capture. Johnson Controls' Katie McGinty, Vice President and Chief Sustainability and External Relations Officer, frames the shift plainly: data centres are moving "from consumers of resources to becoming a source of clean, affordable energy for local communities." That's not a sustainability talking point. It's a description of a new commercial relationship between a facility and its neighbours — one where heat has a counterparty, a contract, and a price.

Why grade matters more than quantity:

The obstacle has never really been whether data centres produce enough heat. A single megawatt of IT load throws off roughly 8,760 MWh of thermal energy a year at full utilisation, more than enough to matter to a district network. The obstacle is temperature. Air-cooled exhaust typically sits at 25–35°C: too cool to inject directly into most heating networks, which is why early heat-reuse projects needed a heat pump stage just to make the resource usable at all, eating into the economics before a single home was warmed.

Liquid cooling changes that maths. Direct-to-chip systems on current-generation AI racks are producing coolant outlet temperatures of 50–65°C. High enough, in many cases, to connect straight into fourth-generation district heating networks without a heat pump lift at all. That single change is why 2026 is the year heat reuse has moved from a Nordic curiosity to a standard line item in European data centre design.

The part that gets skipped:

There's also a seasonal problem that a two-stream design can't handle at all. District heating demand isn't constant, it falls away in summer, and an off-taker with no use for the heat can't simply be forced to take it. The data centre still has to reject that heat somewhere, whatever the season and whatever the network wants that week.

This is the actual job a three-stream heat exchanger does in a heat-reuse system. One stream carries the IT-side coolant. A second carries the district or industrial off-take loop, kept fully isolated from the first by geometry rather than by a second unit in series. A third provides conventional heat rejection, so that when the off-taker doesn't want the heat, the plant can dump it the traditional way without reconfiguring anything or leaving the IT load exposed. All three streams in one compact shell.

Every watt, twice:

The Odense and Finland projects are proof of concept at a scale most operators can't replicate, they sit next to city-scale district networks that already existed. But the underlying economics don't require a capital city's worth of heating demand to make sense; they require a temperature high enough to be useful, and a distance short enough to be affordable. As liquid cooling pushes exit temperatures up and regulation pushes reuse from optional to mandatory, that combination is going to apply to a lot more sites than the handful currently making headlines.

The facilities that get ahead of this won't be the ones that treat heat reuse as a retrofit bolted onto an already-built cooling system. They'll be the ones that designed the off-take into the plant from day one — where every watt of electricity is quietly doing two jobs, and the hardware connecting them was built to keep those jobs from interfering with each other.

Want to know how a compact, ASME-approved heat exchanger fits into your waste heat recovery design? Get in touch with our engineering team.