On 23 October 2026, a guest article by Leonard Burtscher on netzpolitik.org reported that Germany's CDU/SPD coalition plans to weaken the data centre rules in the Energieeffizienzgesetz (EnEfG). The reported changes raise the permitted PUE values and add exemptions for waste-heat reuse, which the coalition agreement describes as returning the law to a "European minimum". The Bundestag is expected to vote in November. Burtscher works for the Umweltinstitut München, which opposes the draft, so his account is an advocate's, though the draft itself is public.
The case for the rollback
The strongest argument for the government is practical. The 2023 EnEfG goes further than EU law. It requires new data centres to reach a PUE of 1.2 or lower, and it phases in waste-heat reuse of 10% from July 2026, 15% from July 2027 and 20% from July 2028 (EnEfG § 11). An operator can only sell heat if a network exists to take it. Many sites have no district heating nearby, and a mandate cannot create one. German operators also compete for capital with sites in countries that impose no such duties, while grid connections and power prices are already the main constraints on AI buildout.
The government's draft, the Gesetz zur Beschleunigung der Umsetzung der Energieeffizienzrichtlinie (Drucksache 21/8027), is framed as implementing the recast EU Energy Efficiency Directive. As reported by the Bundestag's hib service, it would:
- extend the time new data centres have to meet efficiency requirements from two years to four;
- move the 100% renewable-electricity obligation from 2027 to 1 January 2030;
- make the waste-heat rule more flexible, so that heat must be used only where a suitable district heating network is nearby.
The waste-heat change has some logic. The existing statute already exempts operators when a heat network operator declines a cost-price offer, or when a municipal agreement covers a planned network (§ 11(3)). Tightening that exemption into a clearer "where a network exists" test is defensible.
Where the argument breaks down
The case for flexibility does not support a wholesale retreat, for three reasons.
First, the baseline is poor, not ambitious. Netzpolitik reports that existing data centres reuse only just under 3% of their waste heat, against a 20% target for new builds by 2028. Of roughly 1,000 reportable data centres, 319 published named data and 166 reported anonymously. A rule that is mostly unmet because nobody checks is a weak starting point for loosening it. The better response is to fix measurement and enforcement.
Second, the stakes are rising. Netzpolitik cites the government's data centre strategy, which aims to more than double IT capacity in Germany by 2030 and to quadruple AI capacity. (Other coverage cites different capacity figures, so treat the gigawatt numbers with care.) It also cites a Borderstep study for the Economy Ministry projecting about 80 TWh of annual data centre electricity use by 2045. Opponents, including the Umweltinstitut, Fraunhofer ISI and AlgorithmWatch, estimate the weaker rules would mean roughly 9 TWh more electricity use and 10 TWh less useful heat in 2045. These are advocacy-commissioned projections, not official figures, but they point to a real trade-off. Wasted heat and extra load raise system costs that the whole economy pays, including other innovators.
Third, the "European minimum" framing may be legally shaky. At the 5 October hearing, Burtscher argued that lowering standards to implement the directive could be unconstitutional, because an efficiency gap already exists. Netzpolitik adds that Germany has rejected obligation schemes such as white certificates, so the savings the directive requires have to come from somewhere. This is a contested legal argument, not a settled one. The European Commission says the recast directive set an 11.7% final-energy reduction target for 2030 and required transposition by 11 October 2025, and that a data centre efficiency package is planned for 2026. Germany would be loosening its own rules just as Brussels moves to set common ones.
The lobbying question
The Umweltinstitut, LobbyControl and Campact allege "conspicuous parallels" between the amendments and Big Tech lobby papers, with some passages nearly identical, according to Netzpolitik. The article also notes that the Economy Ministry did not consult environmental groups on data centre standards. Industry input into legislation is legitimate and often valuable, and Germany's lobby register exists to make it visible. Similarity of text is not proof of improper influence. The criticism lands, however, because the process looks one-sided. The same week, Chancellor Merz and Nvidia's Jensen Huang announced closer AI cooperation at an investment summit. A government courting investment should be more transparent about who shaped its rules, not less.
The hearing record shows the debate is not simply industry against environmentalists. Günter Eggers of the German Datacenter Association said he welcomes ambitious efficiency requirements if they are technically sound, verifiable and within operators' control, and proposed a new PUE calculation method. DIHK's Erik Pfeifer wants one-to-one directive implementation and market incentives. Even Deutscher Städtetag, which welcomed simplification, called the draft a "deutliche Schwächung" of key efficiency instruments.
A proportionate alternative
Good regulation here would keep the duty and change the instrument:
- Fix the metric. PUE ignores what is done with the heat and what the IT load achieves. Eggers' call for a better calculation method deserves a serious hearing before values are raised.
- Tie heat duties to infrastructure. Require reuse where a network exists or is funded, and require operators to publish what they offered and what was refused.
- Keep deadlines short and conditional. Doubling the grace period from two to four years is blunt. A staged extension tied to documented grid or heat-network delays would give operators certainty without removing the incentive.
- Publish the lobbying record. Release the consultation list and the meetings behind the amendments before the vote.
Germany wants to be a place where AI infrastructure is built. That is better served by predictable, measurable rules than by a quiet retreat that invites a legal challenge and a second round of rules from Brussels. The Bundestag has a few weeks to separate the flexibility the sector needs from the delay it merely wants.