A data centre is a machine that turns electricity into computing power and heat. How well it does this can be measured. For electricity, via the PUE: the ratio of total consumption to IT consumption. For water, via the WUE: the litres that evaporate per kilowatt hour of IT load. Both metrics describe a building; we select based on them. The third belongs to us: the useful life of the hardware. We are also allocated the electricity procurement and the IT load, which forms the denominator in the PUE. This page shows all three — PUE and WUE per location, even where the values are poor — and separates what has been achieved from what we are aiming for.
Every environmental statement on this page carries a number, a measurement method and a time period. All values refer to the 2025 calendar year and are updated annually in April; the greenhouse gas inventory is externally audited with limited assurance.
Reporting year 2025
The PUE describes how much total energy a location needs per kilowatt-hour of IT load. A value of 1.08 means: for 1 kWh in the server, an additional 0.08 kWh goes into cooling, losses and building services. The metric therefore measures a building and not a server fleet. We select based on this and require measurement according to EN 50600-4-2 over twelve months. The value itself depends mainly on the building's cooling technology; what we contribute is the selection and specifications such as the cold aisle temperature.
The Location Uusimaa in Helsinki achieves the lowest value because the location is designed for free cooling with a high cold aisle temperature — i.e. cooling with outside air without a chiller starting up. The climate there extends the hours in which this is sufficient. The Location Spree in Berlin has the highest at 1.16: older cooling technology and warm summer hours when free cooling alone is no longer enough and mechanical cooling must be switched on.
A PUE comparison between providers is only reliable if the measurement boundary, period and utilisation match. We therefore name each location individually instead of a smoothed overall value.
What belongs to us in this calculation is the denominator: the IT load. We reduce it per unit of performance delivered by consolidating systems with low utilisation and replacing inefficient series. This does not affect the PUE — but it does affect consumption.
Scale 1.00 – 1.20 · lower is better
Measured according to EN 50600-4-2 over twelve months, including cooling, uninterruptible power supply (UPS) losses and building services. Values from partial load operations are shown but not adjusted — a half-occupied location inherently has a worse PUE.
Five projects, each with an action and what is measured for it today — even where the measured value is still zero. Two of them lie with our own systems and contracts, three with location selection; the “Lever” row states this per project. Goal and status are separate so that one is not read as the other; we do not promise a year.
Target · energy
We select locations based on how well they can be cooled with outside air, and require a cold aisle temperature at the upper end of what the hardware allows. Both extend the hours during which no chiller starts up.
Target · power
We cover consumption via guarantees of origin and power purchase agreements, and are gradually shifting this coverage from tradable certificates to contracts tied to a named generation facility.
Target · hardware
Replacements are made after the efficiency drop, not according to the depreciation period. Decommissioned systems are refurbished, their data carriers verifiably wiped, and the devices reused instead of being scrapped.
Target · waste heat
We check per location whether a heating network is within reach and at what flow temperature it takes heat. Without an off-taker and without a suitable temperature level, extraction remains ineffective. It is building technology itself — we can make it a selection criterion, not promise it.
Target · water
We require that the adiabatic stage remains limited to the hours in which it actually lowers the electricity demand, and calculate for older dry cooler systems what a closed loop costs in electricity. Water and electricity run against each other here, which is why we state WUE and PUE together.
Adiabatic cooling — cooling by evaporating water in the supply air — lowers power consumption and thus the PUE. However, it consumes water. Reporting only the PUE shifts the problem. The WUE indicates how many litres evaporate per kilowatt-hour of IT load; at the locations where the platform runs, it was between 0.09 and 0.34 litres in 2025. The WUE is also building technology: we request it, list it here per location and include it in the selection — we did not design the cooling ourselves.
Around 70% of our emissions are generated not during operations, but during hardware manufacturing. The most effective measure is therefore not increasing efficiency in the data centre, but every additional year of operation for a server that has already been built.
This section belongs entirely to us. We do not decide on the cooling, power and construction of a location — but we do decide on the procurement, useful life, erasure and reuse of our hardware. Each of the six stages is therefore a commitment with procedures and evidence, not a selection criterion.
We only buy hardware for which the manufacturer guarantees spare parts and firmware for at least eight years from delivery. Systems without this commitment do not make it into the catalogue, even if they were cheaper. Power supplies, fans and storage drives must be replaceable without special tools.
8 yearsGuaranteed spare parts supply
Replacements are not made according to the depreciation period, but when the performance per watt of a new generation justifies the replacement — including the manufacturing emissions of the new device in the calculation. Until then, older systems move into less demanding product lines.
6.4 yearsAverage useful life in the inventory
Removed systems always go through the same steps: cleaning, new thermal paste, replacement of fans and power supplies, memory testing, 72-hour stress test. Systems that pass go back into operations or resale; what fails is repaired and retested or discarded.
81%Proportion of removed systems that pass the test
Storage drives are erased before any transfer: SSD and NVMe via sanitize commands — the drive's own erase commands — according to ATA or NVMe with subsequent verification, HDD via multiple overwriting. The basis is BSI TL-03423 in conjunction with measure CON.6 of the IT-Grundschutz. Every process is logged with serial number, method and test result.
100%Storage drives with erasure log and verification
Refurbished systems go into the secondary market via a network of dealers, mainly to educational institutions and smaller businesses. Only what is no longer functional or can no longer be operated securely due to a lack of firmware supply is scrapped.
89%Reuse rate of decommissioned systems
Non-reusable devices go to a certified waste management company. We receive a certificate for the recovered fractions per batch. Storage drives are physically destroyed beforehand, regardless of whether they were previously erased.
11%Proportion that goes into material recycling
Scope 1 covers direct emissions from what we operate ourselves, Scope 2 covers purchased energy, Scope 3 covers everything upstream and downstream — primarily hardware manufacturing and the building services of the locations where the platform runs. Backup power systems and building cooling are therefore not in Scope 1, but in Scope 3: they are included in our inventory, but we do not operate them. Accounted for the 2025 calendar year according to the Greenhouse Gas Protocol, the standard international framework for greenhouse gas inventories. Audited by an external body with limited assurance. We report Scope 2 twice, because the market-based zero would otherwise give a false impression.
Scope 1
Direct emissions
The technical fleet: trips to the locations, transport of hardware and spare parts. That is all that is listed here: backup power systems and building cooling are systems we do not operate; their emissions fall under Scope 3. The decrease is mainly due to the transition of the fleet to battery electric vehicles.
Scope 2 (market-based)
Purchased energy, contractual
Mathematically zero, because all electricity is covered by guarantees of origin or power purchase agreements. This zero does not mean that no CO₂ is generated on the grid.
Scope 2 (location-based)
Purchased energy, physical
Calculated using the actual grid mix at the respective location. The value follows the provisioned capacity — this is the more honest figure, and we therefore include it.
Scope 3.1
Purchased goods (hardware)
Manufacturing of servers, storage media, network equipment. By far the largest item and the reason why we keep hardware running for a long time instead of replacing it after depreciation.
Scope 3.2
Capital goods
Manufacturing of long-lasting equipment that we own: racks, power distribution in the rack, network distributors, measuring and testing equipment in the workshop. The item fluctuates strongly with what is newly procured in a year; in 2025, very little of this occurred.
Scope 3.3 – 3.15
Other upstream emissions
Services, transport, waste, business travel, upstream energy generation — and the building technology of the locations where the platform runs: test runs of the backup power systems and refrigerant losses from building cooling. Fully accounted for, but with greater uncertainty than Scope 1 and 2.
Sustainability reports are full of numbers that are correct but still misleading. This list names the places where this also applies to our own numbers, and assigns each to one of three categories: actually avoided, mathematically balanced or compensated by certificates. Added to this is the limit of our own responsibility: we include building technology in our balance sheet, but we are not responsible for it. ENTRONYX CLOUD therefore never calls itself climate-neutral — we lack the evidence for this statement.
We cover our consumption entirely via guarantees of origin — certificates for electricity generated elsewhere from renewable sources — and via long-term power purchase agreements. We conclude these contracts ourselves; they are the part of the electricity procurement that we can guarantee. Physically, however, the electricity flows from the local grid, with its respective mix. That is why we also list Scope 2 on a location basis: the 118,400 t are the emissions that actually end up in the atmosphere.
Every additional year of use distributes a server's manufacturing emissions over more operating hours. With an average service life of 6.4 years instead of the industry standard of 4 to 5, the manufacturing share per operating hour drops by 22% to 38%. This is the most effective measure in our footprint and at the same time the most unspectacular.
At the locations where the platform runs, no feed-in to a district heating network is currently measured. The dry coolers operate with a flow temperature below that of the surrounding heating networks; raising it via heat pumps has been calculated, but not built. What we have is a calculation and not a meter reading — which is why there is no credit for waste heat in this footprint.
Emergency power systems — the diesel generators for power outages — must run regularly under load, otherwise they are useless in an emergency; cooling systems lose refrigerant over the years. Both are part of the building technology of the locations and therefore not in our Scope 1: we do not operate these systems. They are still accounted for, upstream under Scope 3.3 – 3.15 and calculated with industry-standard factors instead of measured. They are not offset. At one location, HVO fuel (hydrotreated vegetable oil, a diesel substitute) has been tested since May 2026; this is factored into the selection, but does not result in a commitment.
The 152,400 t from hardware manufacturing are not a measured value. They are derived from quantities, weights and industry-standard emission factors, because manufacturers do not provide a footprint per device for most series. The uncertainty of this item is greater than that of Scope 1 and Scope 2 combined, and it impacts the entire footprint accordingly.
Only Scope 1 — the vehicle fleet — and business travel are offset, together 0.4% of the footprint; for this we use certificates from projects according to the Gold Standard, a certification standard for climate protection projects. We do not buy certificates for location-based Scope 2 and the remaining items from Scope 3: an item of 152,400 t cannot be credibly offset, only reduced. That is why there is a footprint at the end of this page and not a zero.
All values on this page refer to the calendar year 2025 and are updated annually in April. The energy management according to ISO 50001 is externally certified at 4 of the 5 locations where the platform runs: Frankfurt am Main (Location Rhein-Main I), Frankfurt am Main (Location Rhein-Main II), Munich (Location Isar) and Helsinki (Location Uusimaa). Where the certificate is missing, we require the same recording, without being able to prove it with a certificate.
PUE, WUE and occupancy are on every location card, the certificates are available in the original. Anyone wishing to include a number from this page in their own inventory will receive, upon request, the measurement boundary, the method and information on whether the number comes from our operations or from the building services of a location.
