Orphaned Access: The Silent Cybersecurity Gap
Benedikt Langer
5 Min. Read Time Service accounts, API keys, and AI agents often outnumber human accounts. Many of these ...
The planned amendment to the Energy Efficiency Act (EnEfG) is loosening-not tightening-Power Usage Effectiveness (PUE) requirements for data centres. Anyone waiting for regulatory pressure to drive energy efficiency will have to wait a little longer. A far more immediate lever remains the electricity bill: in 2025, German data centres and smaller IT installations consumed a record 21.3 billion kilowatt-hours.
Key Takeaways
Related:CSRD demands Scope-3 IT emissions: CIOs under pressure / DGX-class cost trap: electricity OPEX splits cloud regions
What is the PUE value? The Power Usage Effectiveness compares a data centre’s total electricity consumption with the power used by the IT hardware itself. A value of 1.0 would be perfect and technically unattainable; a value of 1.46 means that almost half the electricity is spent on cooling, power supply and building services rather than computing. That is why the PUE value is the key control metric on which the Energy Efficiency Act is based.
What exactly is changing in the Energy Efficiency Act? The draft bill from the responsible Federal Ministry dated 9 April 2026-later confirmed as the cabinet draft-relaxes the requirements adopted in 2023. For existing data centres the maximum permissible PUE rises from 1.5 to 1.6 from 1 July 2027 and from 1.3 to 1.4 from 1 July 2030, each measured as an annual average. The obligations on waste-heat utilisation are watered down. The threshold above which companies must operate a certified energy-management system climbs from an average of 7.5 to 23.6 gigawatt-hours of annual consumption. A new, less stringent band between 2.77 and 23.6 gigawatt-hours now requires only an energy audit instead of a full management system.
Bitkom and the German Data Centre Association had criticised the original rules as bureaucratic and a competitive disadvantage versus the USA. The amendment is the political response. For CIOs this means: anyone who previously justified energy-efficiency investments with an impending legal obligation now loses that argument. Yet the rise in consumption itself does not lose momentum. According to a Bitkom study conducted by the Borderstep Institute, German data centres and smaller IT installations consumed 21.3 billion kilowatt-hours in 2025, up from about 20 billion the previous year. Cloud utilisation, AI workloads and growing edge capacity are the main drivers. The average PUE in 2024 was 1.46; modern colocation facilities already reach 1.3.
The figure that sets the pace
21.3 billion kilowatt-hours. That is how much electricity German data centres and smaller IT installations consumed in 2025, according to Bitkom and Borderstep Institute-more than double the figure for 2010. The electricity bill is growing faster than any legislative amendment can regulate it.
Anyone aligning their own investment planning with statutory deadlines is currently negotiating a moving target. The sequence shows how much the legal landscape has shifted since 2023-and continues to shift:
Between each of these milestones lie amendment motions, industry hearings and postponements. An investment decision based on today’s draft of an as-yet-unpassed amendment may stand on a different foundation in two years’ time.
If you want to cut energy consumption without waiting for the next regulatory update, there are four independent starting points you can implement immediately-no reporting obligations required.
First, workload placement based on electricity mix. Cloud providers now offer tools to select regions by CO₂ intensity of the local grid and shift load peaks to times with higher renewable energy shares. Carbon-aware scheduling can reduce a cluster’s carbon footprint by up to 41 percent in studies, with minimal latency trade-offs-an obvious switch for batch workloads without real-time demands.
Second, chip selection. ARM-based processors like AWS Graviton consume up to 60 percent less energy than comparable x86 instances at the same performance, while also lowering instance costs. SAP reports a 45 percent energy reduction for HANA workloads on Graviton. Not every application benefits equally, but for many cloud-native workloads the migration is now a straightforward lift-and-shift with no functional changes.
Third, cooling technology. Direct-liquid cooling and immersion cooling can push modern facilities’ PUE below 1.3, versus a legacy average of 1.46. Where waste heat can be economically harnessed-say, in district heating networks-it further improves the balance sheet, though this depends on local demand and isn’t a universal option.
Fourth, software efficiency. Green coding, more efficient algorithms, and a conscious focus on Software Carbon Intensity as a metric can cut individual workload energy use by 10 to over 50 percent, and when combined with hardware levers often proves most effective.
Not every efficiency measure makes economic sense. In practice, waste-heat recovery often stalls because low-grade heat demands expensive heat pumps whose capital costs run five to ten times higher than conventional gas boilers, plus suitable off-takers are rarely nearby. Over half the operators surveyed by industry associations cite poor economics as the main reason heat goes unused.
Similarly, chasing ultra-low PUE targets in new builds with traditional air cooling is technically demanding and capital-intensive without proportional business benefit. And a practical objection carries weight: the biggest CO₂ lever usually lies in the power supply itself-green-energy contracts and power purchase agreements-not in marginal PUE tweaks inside the data centre. Spending scarce capital on a 0.1-point PUE improvement instead of a better electricity deal may well mean optimizing the wrong lever.
Takes effect immediately, independent of legislation
| Chip selection (ARM vs x86) | Up to 60 % less energy, predictable migration effort |
| Workload placement | Up to 41 % lower carbon footprint, minimal latency penalty for batch jobs |
| Waste-heat utilisation | Investment 5–10× higher than gas heating, often uneconomic |
| PUE optimisation with air cooling | High CapEx, business benefit does not scale proportionally |
First: audit your own cloud and server inventory for chip architecture and, regardless of the current status of the EnEfG amendment, set up an ARM pilot for migration-ready workloads. Second: activate your cloud provider’s workload-placement tools and test carbon-aware scheduling for non-time-critical batch processes before reporting obligations make it mandatory. Third: before investing in heat-reuse or PUE improvements, run your own cost-benefit analysis against the alternative of a better electricity contract-don’t pre-empt rules that are still in flux during the legislative process.
It relaxes PUE ceilings for existing data centres, softens heat-reuse mandates and raises the thresholds at which companies must operate a certified energy-management system. The draft is still in the legislative process and may yet change.
Under current law, an average of 7.5 gigawatt-hours of final energy over three years. The amendment proposes lifting this to 23.6 gigawatt-hours, introducing a gentler audit band between 2.77 and 23.6 gigawatt-hours.
For migration-ready cloud workloads, switching to ARM-based chip architectures typically delivers the quickest win: up to 60 % lower energy consumption at comparable performance-no capex required.
No. Without nearby heat off-takers and favourable temperature levels, heat-pump investments can run five to ten times the cost of a gas boiler. It usually pencils out only under the right local conditions.
Read more on Digital Chiefs
Digital ChiefsThe Blind Spot in Fiber-Optic StatisticsDigital ChiefsWhy Startups Lose Billions of Customers Despite Their MillionsDigital ChiefsHow Decision-Makers Will Actually Use AI by 2026More from the MBF Media Network
cloudmagazinHigher PUE limits and longer deadlines: how the EnEfG amendment relaxes data-centre obligations mybusinessfutureInvestment logjam: how AI uncovers hidden budgets securitytodayUptime in the data centre: why smart cooling is a security issueImage source: AI-generated (July 2026)