A data center without diesel: backup power without the nitrogen problem

Why diesel backup is a Dutch siting problem, not just an emissions one

For a tenant or investor evaluating a Dutch data center site, backup power usually reads as a technical line item: N+1 or 2N gensets, fuel contracts, a test schedule. In the Netherlands it is also a permitting variable, worth understanding before it becomes a delay on a project timeline.

The number that decides it is small: 0.005 mol of nitrogen per hectare per year. Under the Dutch Nature Conservation Act (Wet natuurbescherming) and the Environment and Planning Act (Omgevingswet), any project whose calculated nitrogen deposition on a Natura 2000 protected area exceeds that threshold needs a separate nature permit, unless it can offset the deposition through internal or external netting, or commission a supplementary ecological assessment showing conservation goals aren't harmed. AERIUS, the calculation tool the Omgevingswet mandates, rounds a project's calculated contribution to two decimal places for system-technical reasons: a contribution below 0.005 mol/ha/year rounds down to 0.00 and in practice requires no separate nature permit; a contribution of 0.005 mol/ha/year or more rounds up to 0.01, which can trigger a supplementary ecological assessment once a significant negative effect cannot be excluded in advance. 0.005 mol/ha/year is therefore the effective assessment boundary, and 0.01 mol/ha/year the smallest reportable value above it. That rounding is not an ecologically established safe level: the former Ministry of Agriculture, Nature and Food Quality (LNV) had ecologists review in 2019 whether it was ecologically defensible, and they concluded that smaller contributions cannot in any case cause a peak in nitrogen load. There is an active political motion to raise the threshold toward roughly 1 mol/ha/year, based on Wageningen-adjacent research (Petersen); as of this writing it has not been adopted.

The scale gap with neighboring markets is what surprises most international investors and tenants. Germany's general threshold is roughly 21 mol/ha/year; Denmark's is roughly 35 mol/ha/year: four to seven thousand times looser than the Dutch figure. That is a different order of magnitude, and it is the biggest reason a genset configuration that clears permitting without comment in Germany or Denmark can trigger a full AERIUS calculation, a netting exercise, or a multi-year appeal before the Council of State (Raad van State) in the Netherlands. For site selection, this asymmetry is a real variable: the emergency-power design that works elsewhere in Europe is not automatically the design that gets a Dutch permit on schedule.

What gensets actually contribute

Backup generators don't run continuously: they test briefly and irregularly. Dutch environmental rules (Activiteitenbesluit stookinstallaties) make that test regime the pivot point: gensets run under a certain number of hours per year fall under a lighter emissions regime, provided this is demonstrated through hour- and fuel-metering. The moment a genset is also used for peak-shaving or pooled as dispatchable emergency capacity (even while staying under that hour threshold), it is reclassified as a regularly-operating installation subject to periodic emissions measurement. That is exactly why the test regime, not just nameplate capacity, drives the nitrogen calculation.

Modern diesel engines are considerably cleaner than a decade ago. EU Stage emissions standards for generator sets have tightened step by step: Stage II sits around 7.0 g NOx/kWh, Stage IIIA around 4.0 g NOx/kWh, and the current Stage V standard caps out at 0.4 g NOx/kWh: a reduction of more than 90% from Stage II. That is a genuine engineering story. But even at the Stage V limit, using a standard diesel energy content of 10.65 kWh/liter, the ceiling works out to roughly 4.26 grams of NOx per liter of diesel burned at that regulatory limit, a cap calculation, not a measured real-world figure, but enough to show NOx from diesel never reaches zero, however clean the engine.

The precedent that now exists: Eemshaven

Until recently, there was no public, technical study in the Netherlands that modeled nitrogen deposition specifically for a data center project. That changed with Onderzoek Stikstofdepositie (Nitrogen Deposition Study), commissioned from DGMR Industrie, Verkeer en Milieu for a proposed data center at Oostpolder, Eemshaven (final version 23 January 2025, update package 4 March 2025): the first publicly available, primary technical nitrogen study for a Dutch data center project. It is worth understanding how it calculates.

The report models the worst realistic test scenario: every generator run at 100% load, 60 minutes per month, twelve times a year: 12 operating hours per generator per year, chosen because full load produces the highest NOx output per generator (lower load percentages such as 10, 25, or 75% were deliberately excluded). The emissions characteristic used: 1,565 g NOx per Nm³ of exhaust, an exhaust flow of 3.95 Nm³/s, at 446°C, figures specific to one supplier's generator spec for this one project, not a national or EU norm. Multiple generators were combined into a single AERIUS point source, alongside traffic: roughly 30 trucks and 575 passenger cars per day for the full complex, again, a deliberate overestimate.

The result: for most surrounding Natura 2000 areas, calculated deposition rounds to 0.00 mol/ha/year: under the threshold, so no separate ecological assessment is required. One area near Delfzijl, added to AERIUS in October 2024, does receive measurable deposition; a separate ecological assessment (Koolstra Advies, commissioned by the Province of Groningen, October 2024) concluded there is no negative effect on vegetation there, because sedimentation dominates habitat quality at that site. The cross-border picture is notable too: German deposition from the same plan comes out at 0.16 mol/ha/year on the German Natura 2000 area Hund und Paapsand, just 1.8 km from the site, well under that specific area's German threshold of 7.14 mol/ha/year (a site-specific figure, not Germany's general ~21 mol/ha/year norm). The nearest Dutch Natura 2000 area, the Wadden Sea, sits roughly 13 km away. Under the Omgevingswet, a Dutch project within 25 km of the German border must also assess German deposition, and both countries' authorities must coordinate if the German threshold would be exceeded.

The Eemshaven report demonstrates that a carefully modeled test regime (brief, full-load, worst-case) can stay under the Dutch threshold. It does not demonstrate that this outcome is automatic for every project and location; the result depends heavily on distance to Natura 2000 areas and the specific generator configuration chosen.

What happens when it doesn't line up: Wieringermeer

The other side of this story played out at Microsoft's data center complex in Wieringermeer/Middenmeer (municipality of Hollands Kroon, North Holland), built for roughly a year without the required construction and nature permits. The Province of North Holland's provisional authorization (September 2021) was overturned by the Council of State in November 2022, because Microsoft could not finalize the nitrogen emission calculations: heavy diesel construction machinery produced emissions the company had not secured approval for, and it was over a month late submitting the required emissions data to the regional environmental agency (Omgevingsdienst Noordzeekanaalgebied). That agency ultimately issued a construction-and-operations permit on 19 April 2023; Stichting Red de Wieringermeer appealed. (The current status of that appeal was not re-confirmed for this article: treat this as a permitting-risk precedent, not a closed case.)

What this illustrates: nitrogen calculations for the construction phase (heavy machinery, not just the emergency generators themselves) are just as much part of the permitting critical path as the gensets that run after commissioning. LTO vice-chair Henk Geerligs summarized local sentiment at the time: "Voor ons gelden strengere stikstofregels en er wordt geen uitzondering gemaakt" ("stricter nitrogen rules apply to us, and no exception is made"), pointing to the fact that Microsoft was able to build for a year without the same permits local farmers operate under. Fair or not, that perception of a double standard is exactly the conversation a municipality has to be able to have with residents before a data center project breaks ground, and it goes better when the nitrogen math is right from the start, not corrected after the fact.

	Netherlands (general threshold)	Germany (general threshold)	Denmark (general threshold)
Deposition below which no separate nature permit is needed	≤ 0.005 mol/ha/year	≈ 21 mol/ha/year	≈ 35 mol/ha/year
Order of magnitude vs. NL	1×	≈ 4,200× looser	≈ 7,000× looser

(Note: the Eemshaven report separately cites a site-specific German threshold of 7.14 mol/ha/year for the nearest German Natura 2000 area: that is a local figure for one site, not Germany's general ~21 mol/ha/year norm above.)

The alternatives, compared honestly

If diesel is the permitting bottleneck, the obvious next question is what replaces it. Two technologies have moved beyond pilot status, and neither replaces diesel for every scenario. That caveat matters more than the sales pitch.

Battery energy storage (BESS). Microsoft had Saft (a TotalEnergies subsidiary) deliver a turnkey megawatt-scale lithium-ion battery system at its Stackbo, Sweden data center, operational since June 2023, replacing diesel gensets at that site: eight Intensium Max 20 High Energy containers in four groups of 3 MW peak power each, delivering roughly 4 MWh (up to 80 minutes) of backup. It also provides grid frequency support and black-start capability, and sits inside Microsoft's publicly stated 2030 target (announced July 2020) for diesel-free data centers.

Google has run a 2.75 MW / 5.5 MWh lithium-ion system (Fluence Gridstack, sixth generation) at its Saint-Ghislain, Belgium site since 2020, replacing diesel backup for 3 MW of live production load, a first for Google globally at the time. That battery has since successfully carried the full critical load through at least one real utility outage, which is what separates it from a demonstration project.

The real limitation is duration, not capability. Grid-scale short-duration BESS is generally designed for 2–4 hours of discharge; data-center-specific extended designs reach 4–8 hours. A central UPS alone typically bridges only 5–12 minutes before a generator or battery has to take over. Microsoft and Saft's own risk analysis concluded 80 minutes of battery backup covers the vast majority of real-world outage risk at Stackbo, but for a multi-day outage, lithium-ion is not the answer; that requires longer-duration storage technology not yet mainstream at data-center scale.

Cost also deserves an honest framing. A June 2025 NREL estimate (US market) puts diesel gensets at roughly $1,000/kW installed, against roughly $1,300/kW for a 4-hour BESS. Battery storage is not currently cheaper on raw capex: the case for BESS has to rest on avoided fuel logistics and maintenance overhead, permitting speed, and potential grid-services revenue, not a lower sticker price.

There is a genuine supply-chain counterpoint too. The global lithium-ion battery manufacturing supply chain emits an estimated 1.3–1.5 gigatons of CO2 per year today, with a projected path toward roughly 1.0 gigaton within a decade (an optimistic, decarbonized-supply-chain scenario lands around 0.5 gigaton). Per-unit manufacturing footprint figures vary widely across published sources by methodology and material sourcing (nickel refining alone runs 3–7 kg CO2 per kg of refined nickel), so we cite a range here rather than a single number.

Hydrogen fuel cells. NorthC opened what it describes as Europe's first data center with emergency power from green-hydrogen fuel cells (PEM technology, Nedstack) at its Groningen 2 site in June 2022, built in modular 500 kW stages up to a maximum of 1.5 MW. NorthC cites savings of "tens of thousands of liters" of diesel per year and roughly 78 tons of CO2 per year at full utilization (39 tons at partial use), comparable to 24 cars driven roughly 32 km a day for a year, with a service life of 20-plus years and water as the only byproduct. That last point marks the technical distinction from hydrogen combustion engines (such as NorthC's own Jenbacher hydrogen motors for its Eindhoven expansion): a fuel cell converts hydrogen electrochemically, without combustion and therefore without NOx formation; a hydrogen combustion engine burns hydrogen much like a gas engine, and high-temperature combustion typically does form NOx from nitrogen in the intake air, regardless of fuel type. That distinction matters: "zero emission at the point of use" is a defensible fuel-cell claim; "hydrogen backup is emission-free" as a blanket statement is not, once combustion-type systems are included.

The honest limitation sits upstream, in the supply chain: a fuel cell is only as clean as the hydrogen feeding it. Without a reliable, ideally green-electricity-fed local hydrogen production and delivery chain, "zero emission at the point of use" does not guarantee zero emission across the full chain. Hydrogen logistics (storage, delivery, safety procedures for trained personnel) are also not yet as simple to organize everywhere as topping up a diesel tank.

	Diesel (Stage V)	BESS (lithium-ion)	Hydrogen PEM fuel cell
Backup duration	Hours to days (fuel-dependent)	Usually 2–4 hours, up to 8 hours in data-center designs	Hours, resuppliable as long as hydrogen delivery continues
NOx at point of use	Low but not zero (Stage V ≤ 0.4 g/kWh)	None	None for fuel cells; possible for hydrogen combustion engines
NL permitting impact	AERIUS calculation required, possible nature permit	No NOx emission to calculate	No NOx emission to calculate for fuel cells
Installed cost	≈ $1,000/kW (NREL, US, 2025)	≈ $1,300/kW for a 4-hour system (NREL, US, 2025)	Project-dependent; no comparable generic figure found
Biggest limitation	NOx, test regime, permitting	Duration for multi-day outages; supply-chain CO2	Local hydrogen supply chain; logistics and trained personnel

What this means for a new-build

The practical conclusion is not "swap diesel for batteries" as a like-for-like substitution. Backup power needs to be part of the permitting strategy from the design table, not bolted on after the layout is fixed. A hybrid configuration (battery storage covering the first minutes to hours, with a longer-duration layer behind it for the rare extended outage) avoids relying on one technology to cover every scenario, while also limiting how many NOx sources end up in an AERIUS calculation in the first place. For a tenant or investor comparing Dutch sites against alternatives elsewhere in Europe, that is a genuine siting advantage where it is actually engineered in, not a marketing line: a project designed against the 0.005 mol/ha/year reality from day one is less likely to face a Wieringermeer-style permitting delay.

There is upside beyond permitting, too. TenneT, the Dutch transmission system operator, must contract 123 MW of Frequency Containment Reserve (FCR) daily in 2025, with a minimum of 37 MW physically located in the Netherlands: a market batteries suit well given their fast response times. TenneT contracted its first steerable battery specifically as a grid-congestion reliever on the high-voltage network in late 2025/2026: the Sequoia project by Green Energy Storage, roughly 200 MW, in Oosterhout, expected operational in 2027, paired with a transport right guaranteeing at least 85% annual grid access. That is a standalone grid-battery project, not a data center's own system: no confirmed Dutch example exists yet of a data center dispatching its own idle backup battery for FCR or congestion services to TenneT. The model is proven in principle (see Microsoft's own frequency-support and black-start functionality at Stackbo), but in the Netherlands today it remains an opportunity, not established practice.

GreenDatacenters' approach

Our own Clean Resilience & Nitrogen+ page describes a hybrid configuration built from six coordinated elements: battery energy storage for immediate response, hydrogen PEM fuel cells under evaluation for longer-duration backup, intelligent power management coordinating the layers in real time, safe hydrogen storage and logistics assessed per location, renewable energy integration, and heat reuse from the fuel cells where a nearby off-taker exists.

We are deliberately careful with that wording: the fuel cells are under evaluation, not presented as an existing, operating system, and the same page names six unresolved constraints: permitting that varies by jurisdiction, hydrogen logistics requiring a local supply chain, fuel cell reliability at data-center scale still being validated, scalability for extended-duration backup not yet solved industry-wide, a lifecycle carbon footprint depending heavily on hydrogen origin, and a need for trained operators with specialized safety procedures. Our ambition toward nitrogen-positive development is exactly that: a design ambition assessed across ten categories, not a claim of current or achieved performance.

What we can say, based on the record in this article: the Dutch nitrogen framework is strict enough to make diesel backup a genuine permitting risk for a new data center, and thanks to the Eemshaven study, there is now a public precedent for how a project can clear the AERIUS test. A design that accounts for that math from the first sketch, rather than running into it after the fact as happened at Wieringermeer, is not only cleaner, it is also faster to permit. Discuss the backup power and nitrogen strategy for a specific location with our team.