The 10% Nobody Budgeted For

Germany's Energy Efficiency Act mandates that new datacentres commissioned after 1 July 2026 must reuse at least 10% of their waste heat. That date arrived six weeks ago. The regulation has been public since September 2023. The penalty for non-compliance is up to €100,000. And yet the mechanical engineers I speak to across the colocation and hyperscaler segment are redesigning cooling systems this month for builds whose excavation began in 2024, because the heat recovery specification was never translated from a legal paragraph into a line item with a supplier and a price.

This is not a grid queue problem. It is not a community opposition problem. It is the rarer and more surgical failure mode: a hard regulatory deadline that forces you to find an offtaker for a product your facility was not designed to produce, and to prove that offtaker exists before you are allowed to turn the servers on.

what the law actually requires

The EnEfG applies to facilities with a connected load greater than 300 kW and measures compliance using Energy Reuse Factor (ERF), the proportion of energy redirected outside the datacentre's IT and infrastructure operations. A 30 MW IT datacentre with a PUE of 1.2 must supply more than 3.6 MW of recoverable heat annually to hit 10% ERF, rising to 7.2 MW for the 20% target in 2028.

The law is clear on what counts. Heat reused on-site qualifies. Heat fed into existing or newly established district heating networks qualifies. Venting warm air into the atmosphere does not. The reporting obligation is annual, to the Federal Office for Economic Affairs and Export Control. The burden of proof sits with the operator, not the regulator.

The mandate escalates to 15% in 2027 and 20% in 2028, making heat reuse a non-negotiable design parameter for all new builds in Europe's largest market. There is an exception clause for technical or economic infeasibility, but if a nearby heat network operator refuses an offer to purchase heat at cost within six months, and the datacentre has built the necessary transfer infrastructure, the datacentre is exempt. That carve-out exists because the drafters understood that the bottleneck is not the engineering, it is finding someone willing to take delivery of 50-70°C water at the volume and timing your cooling loop produces it.

the retrofit that wasn't scoped

Air cooling dominates the installed base. Server exhaust in air-cooled facilities typically reaches 30-45°C. That temperature is too low for third-generation district heating networks, which require supply temperatures of 70-100°C. Bridging that gap demands a heat pump, and heat pumps for air-to-water applications achieve a coefficient of performance of 2.5-3.5, meaning you spend one watt of electricity to move 2.5-3.5 watts of thermal energy. The power demand for the heat pump was not in the original electrical infrastructure sizing. The capital cost was not in the original construction budget. And the commercial agreement with a heat offtaker was not a gating item for the construction permit, because when the site was selected in 2023 the law had been announced but not yet enacted.

Liquid cooling changes the arithmetic. Direct-to-chip systems deliver outlet temperatures of 50-60°C, enabling connection to low-temperature networks with a COP of 3.5-5.0. Fourth-generation networks, which operate at 50-70°C, can often accept liquid-cooled datacentre heat directly without a heat pump. But retrofitting liquid cooling into a build designed for air is not a change order, it is a redesign. The server racks, the coolant distribution manifolds, the redundancy architecture, and the floor loading all change. If your go-live date is Q4 2026 and you are reading the regulation properly for the first time in August, you are negotiating with your general contractor about what stays in scope and what gets valued as a variation, and the client is asking why this was not flagged when the site was selected.

The answer, in every case I have seen, is the same. The site was selected for power availability, fibre routes, and land cost. Heat offtake proximity was not a variable in the model, because the team running the site selection did not yet believe the law would bind.

the offtaker is the constraint

Germany has become a testbed for compliance technologies and business models, and success or failure here will likely influence regulatory approaches across the rest of the EU. That is the macro story. The micro story is that around 1,000 datacentres in Germany fall under the scope of the EnEfG, and every one commissioned after 1 July needs a heat sales contract or a documented refusal from a network operator to demonstrate compliance.

Site evaluations in the EU now routinely include analyses of nearby district heating networks, industrial parks, or agricultural facilities that can consume large volumes of low-grade heat, making partner availability a key gating item for development. That sentence describes the future. The present is dozens of projects whose construction began before that analysis was standard, and whose operators are now approaching municipal heating companies, asking if they will take heat that was not in the city's infrastructure planning.

The economics can work. One operator told me their team had modelled €62,000 per MW annually in potential revenue from heat sales, assuming a municipal offtaker and a fourth-generation network within 2 kilometres. But revenue requires a counterparty. The network operator needs to demonstrate to their own regulator that the added supply will not destabilise their existing heat balance, that they have the pump capacity to move the additional flow, and that their customers will accept a new heat source that was not in the rate case. That is a six-month process if it goes well, and these projects do not have six months.

The fallback is on-site reuse. ERF measures the proportion of energy redirected outside the datacentre's IT and infrastructure operations, so heating the office block next door counts if the heat would otherwise have been produced by a boiler. I have seen proposals to heat adjacent warehouses, to supply process heat to a logistics facility sharing the site, and in one case to route waste heat to a covered agriculture operation purpose-built to absorb it. Each of those requires a capital partner willing to build the heat sink, and a commercial agreement that survives the operator's next refinancing. It is solvable, but it was not solved in advance, and the clock is running.

The mandate applies to new datacentres commissioned after 1 July 2026, escalating to 15% in 2027 and 20% in 2028. Existing facilities are exempt from the heat reuse target, though they face separate PUE requirements. That creates a perverse incentive: delay commissioning until the heat contract is in place, which delays revenue, or commission without a compliant heat reuse path and argue technical infeasibility in the first annual report. Neither option appears in the investment committee materials, but both are live conversations in project meetings.

the compliance bet

The way you know a regulation has teeth is when operators start designing around the exception clauses instead of designing for compliance. If a heat network operator refuses an offer at cost within six months and the datacentre has built the transfer station, the exemption applies. I have now seen three proposals where the transfer station is specified and budgeted, the offer letter to the municipal heating company is drafted, and the operational assumption is that the offer will be refused because the network lacks the capacity to absorb the flow. The operator gets the exemption, the capital cost of the transfer station is a small fraction of a full heat pump retrofit, and the project can commission on schedule.

That is not the intent of the law. It is also not non-compliance. It is what happens when a mandate with a hard deadline meets a supply chain and a partner ecosystem that were not built to support it. The thermal infrastructure needed to make 10% ERF economically rational at scale—fourth-generation district heating networks, industrial heat consumers co-located with datacentres, regulatory frameworks that let network operators add non-traditional sources without a rate case—does not yet exist at the density the law assumes. The operators who are commissioning in Q3 and Q4 2026 are not designing for the future state, they are designing for the state that exists today, and documenting why that state made full compliance infeasible.

Germany's Energy Efficiency Act sets a target of 10% heat reuse by 2026 and 20-30% by 2028. The 2028 target will force a different conversation, because 20% cannot be met with on-site reuse and an exemption letter. It requires an offtaker, it requires liquid cooling or a heat pump sized for the full load, and it requires that those decisions be made during the site selection, not during the fit-out. The projects breaking ground now will either have that in the spec, or they will not break ground in Germany. That forcing function is working as intended. The question is whether the 10% target, applied to projects that were designed before the ecosystem was ready, becomes a compliance milestone or a documentation exercise. The answer will be visible in the BAFA reporting database in Q1 2027, when the first annual filings are due.

Tarry Singh is the founder and CEO of Real AI (realai.eu), an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan (earthscan.io) for Energy AI, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.