Heat reuse from data centres works well at the edge and at scale. It’s everyone in the middle who’s stuck with the expectations but not the infrastructure.

Economists talk about the ‘squeezed middle’ – households that earn too much to qualify for income support but not enough to feel totally secure. They sit in the uncomfortable space between extremes, expected to behave like the very affluent while lacking the advantages that make those expectations possible.

Over the last few weeks – in Cannes, in Manchester, in London – I’ve spoken to many interesting people about data centre heat reuse. It’s an intriguing, fast evolving subject, and as someone who is passionate about the positive footprint data centres can provide communities – it’s an area with massive potential upside as outlined in a great recent LinkedIn post by David Davies.

For decades, our industry has treated “heat rejection” as a polite euphemism for ejecting potentially useful thermal energy into the atmosphere. The mission was simple: build fast, scale the compute, keep the servers cool. But with data centre electricity consumption rising, and sustainability shifting from marketing language to regulatory mandate, pure heat rejection is becoming harder to justify.

However, from those discussions it’s clear to me that heat reuse from data centres has arguably developed its own ‘squeezed middle’.

At one end of the spectrum, smaller edge sites and facilities (ranging from approx. 25kW to 400kW) can hand off their waste heat to a nearby swimming pool, leisure centre, greenhouse or similar. We’ve all heard about (blimey even my Mother-in-Law referenced this on Sunday…) Deep Green’s model of warming UK pools and there are other interesting projects like Thermify’s concept of distributed cloud infrastructure in homes that also deliver low-cost, low-carbon home heating.

At the other end, huge multi-megawatt (MW) Nordic hyperscalers or supercomputing sites sit inside cities built around district heating, where industrial-scale thermal output is a feature, not a problem. Data centre certification and advisory firm Uptime Institute calls this the “structural advantage” of cold climates and mature heat networks.

For example, Finland’s LUMI supercomputer in Kajaani already supplies around a fifth of the city’s heating demand, and Stockholm Data Parks links its facilities directly into the city’s district heating network. Plans for 100MW-1GW facilities are also at the scale where heat reuse can be designed-in from the start – which sounds like what Anna Carolina Dowson and the clever people at Greenscale are intending to do in Northern Ireland and Norway.

Stuck in the middle

In the middle – where many UK colocation providers operate — the picture looks very different. With many colo facilities using somewhere in the region of 5-40MW, these operators generate far too much heat to treat reuse as a niche, swimming pool-esq side-project, yet sit in dense urban areas where the off-take infrastructure is fragmented, immature, or more often than not, simply absent.

The reality is, in urban areas like London and Manchester – retrospectively fitting a 20-40MW heat network into and around pre-War “Coronation Street” style terraced housing is a considerable undertaking, and a challenge that would come at great cost, great inconvenience and would probably inspire a lot of chin-stroking in the ‘Rovers Return’.

So, in the UK while there’s a growing expectation that heat should be reused, the infrastructure and ecosystem isn’t built to receive the load from the vast majority of data centres – which is probably one of the main reasons the UK is lagging behind on this critical issue, as highlighted in Fleur Doidge’s excellent recent article in Computer Weekly.

The regulatory picture

Regulation is never my favourite subject and in the case of heat reuse it is starting to tighten, though admittedly the picture varies by jurisdiction. As John Booth MBCS, CDCAP, CDCSP highlights there are 5 EU Code of Conduct for Data Centres (Energy Efficiency) best practices dedicated to the “Reuse of Data Centre Waste Heat” and a standard, the ISO/IEC 30134-6 Energy Reuse Factor. In addition, the EU’s Energy Efficiency Directive (EED) requires data centres with more than 1 MW of total energy input to recover their waste heat or demonstrate that doing so is technically or economically unfeasible. EU member states are now writing this into national law — Germany’s transposition, for instance, sets a minimum waste heat reuse rate of 10% from July 2026, rising to 20% by 2028, with non-compliance carrying fines of up to €100,000.

The UK obviously sits outside the EED, but Whitehall has its own levers. For example, the £288 million Green Heat Network Fund is part of a broader push to lift heat networks from around 3% of UK heat demand today to a government target of 20% by 2050.

The clearest UK example of where this funding is going is the Old Oak and Park Royal Energy Network (OPEN) in West London — the UK’s first scheme to take data centre waste heat at scale. It will recover 17 MW of waste heat from two data centre sites delivering up to 95GWh of heat annually and eventually warming as many as 25,000 homes between 2028 and 2040, backed by £36 million from the Green Heat Network Fund. A second scheme is in development near Gatwick Airport, where a data centre and the airport’s own waste heat could supply 46GWh to homes in North Crawley.

While these projects are significant, sticks and carrots alone won’t make the business case. That has to come from the physics and the economics lining up.

The physics are improving

Like regulation, physics isn’t also one of my favourite subjects, but heat reuse dynamics is actually pretty simple. So, traditional air cooling usually produces low-grade, diffuse heat that’s difficult to reclaim. Liquid cooling – which is what we will be providing in KLON-03 in Harlow, changes the equation. Direct-to-chip and immersion systems can deliver outlet temperatures of 45-70°C, the range experts describe as the “Goldilocks zone” for district heating: hot enough to be useful with minimal upgrading, cool enough to remain practical to handle.

However in the colocation space, workload volatility represents the other half of the heat-reuse challenge. While district heating networks require a stable, predictable baseload, data centre heat output fluctuates alongside client compute cycles. This variation is manageable in traditional enterprise and hyperscale facilities, but AI workloads—particularly large-scale training—introduce intense volatility. Furthermore, while AI inference is more stable, the market remains dynamic and unpredictable. All of this means the heat generated could fluctuate massively. Compounding this, colocation facilities often take years to fully lease, meaning actual heat output will fall short of utility expectations until the site reaches full capacity.

Getting over these challenges will require a shift in facility design but also integration with related infrastructure, with local government taking the lead. A data centre operator can potentially assist by exhausting heat to one point on their perimeter boundary, but that needs to then interface and connect with a local government owned, managed and metered heat off-take system – and as previously highlighted, there aren’t many of those in the UK right now to accommodate the ‘squeezed middle’.

From my discussions with heat reuse experts – I had a great conversation with Edward Galvin in Cannes – I think the real opportunity lies in so-called Integrated Energy Systems: combining on-site generation, thermal storage, and liquid cooling to buffer output. Storage becomes the shock absorber that smooths the peaks and troughs of IT demand, letting data centres behave as true prosumers rather than bystanders in the energy transition. However, who foots the bill for this kind of technology and infrastructure?

Where this leaves the squeezed middle

Crucial questions remain: How aggressively will regulators continue to ratchet up targets? How will utilities respond regarding infrastructure investments and heat-reuse incentives? How will local governments tackle the challenge of deploying (or retrofitting) networks into existing housing stock.

Because the inherent volatility of data centre loads requires a physical storage buffer between operators and public infrastructure, the answers to these questions will determine whether the squeezed middle remains hot and bothered, or learns to keep calm, cool, and carry on..