Data center generator emissions and Tier 4 air permitting

Generator emissions compliance, and why it gates the whole project

Generator emissions compliance is the body of air-quality rules that decides which diesel standby engine you are allowed to install at a data center and how many hours you are allowed to run it. The diesel sets that back up the building emit nitrogen oxides, particulate matter, and carbon monoxide every hour they run, and the rules that cap those emissions reach back to the engine on the order and the permit on the wall. Pick the wrong Tier or skip the permit and the project stops, no matter how good the electrical design is.

Two separate things are in play, and people blur them. The federal engine standard sets a minimum emissions Tier the engine has to meet to be sold and installed as a stationary set. The air permit, issued by the state or the local air district, sets what that specific site is allowed to emit and run. The engine can be perfectly legal as a product and still be unpermittable at a given site, or permittable only with controls the catalog engine does not have.

On a real data center program the emissions and air permit are often the long pole, longer than the gear lead time and sometimes longer than the utility interconnection. Sizing the plant is its own problem, covered in the generator sizing guide, and avoiding the diesel question altogether by building on-site primary power is the subject of the on-site generation guide. This one is about the emissions and the permit: what the engine has to be, how long it can run, and what the air agency makes you prove.

What the diesel actually puts out, and who cares

Diesel exhaust is not a nuisance, it is a regulated mix, and three components drive the whole permitting fight. Nitrogen oxides form in the hot combustion and feed ground-level ozone and regional haze. Particulate matter is the soot you can see and the fine fraction you cannot, the part that lodges in lungs. Carbon monoxide rounds it out. The air agency cares about all three because they roll up into the regional air-quality picture the state has to defend to the EPA.

The community cares for a simpler reason. A row of large diesels firing up to test, or running on a real outage, is loud, it smokes on startup, and it sits next to neighborhoods that increasingly know what a data center is. Air permits draw public comment now in a way they did not five years ago, and a contested permit adds months. Environmental-justice review, where a site sits near an already burdened community, can add conditions or stop a permit outright.

So the emissions number is never just an engineering value. It is the thing the regulator, the neighbor, and the activist all read off the application. Treat the permit as a public document from day one, because it is one.

What is a Tier 4 generator?

A Tier 4 generator is a diesel engine certified to the EPA's most stringent emissions tier, the level that cuts nitrogen oxides and particulate matter far below the earlier tiers. The EPA tiers run from Tier 1 through Tier 4, each step tightening the allowed NOx and PM, with Tier 4 Final the cleanest. Hitting Tier 4 Final on a large engine almost always takes aftertreatment, selective catalytic reduction for NOx and a diesel particulate filter for PM, because the engine alone cannot get there.

The lower tiers are still real engines you will see in the field. A Tier 2 set has no aftertreatment and emits several times the NOx and PM of a Tier 4 Final set of the same size. That gap is exactly why the tier question matters: a Tier 2 engine is cheaper, simpler, and far dirtier, and whether you are allowed to install one comes down to how the engine is classified and where the site sits.

The tier names and the exact gram-per-kilowatt-hour limits are set by the EPA and have shifted across rulemakings and engine sizes, so confirm the current limit for the displacement and power band of the actual engine, against the EPA standard and the engine manufacturer's certification, before you write it into a submittal. Do not quote a tier limit from memory on a permit application.

The two federal rules: NSPS Subpart IIII and RICE NESHAP

Two federal regulations govern the stationary diesel set, and you have to know which one is talking. The New Source Performance Standards for stationary compression-ignition engines, 40 CFR Part 60 Subpart IIII, is the one that ties the engine to the EPA nonroad Tier levels and splits engines into emergency and non-emergency. It is the rule that sets the run-hour structure and the non-resettable hour meter. For new data center standby diesels, Subpart IIII is the rule you live under most of the time.

The second is the RICE NESHAP, the National Emission Standards for Hazardous Air Pollutants for reciprocating internal combustion engines, 40 CFR Part 63 Subpart ZZZZ. It addresses hazardous air pollutants and applies depending on the engine and whether the site is a major or area source of those pollutants. For a new engine that meets Subpart IIII, ZZZZ compliance is often satisfied by meeting the NSPS, but the interaction depends on the engine and the source classification.

The subpart letters and the section numbers are stable enough to cite, but the applicability turns on engine vintage, size, and source status, and the EPA amends these rules. Confirm which subpart and which provisions apply to the specific engine and site against the current CFR and the air district, rather than assuming the standby exemption carries through automatically.

What is the difference between emergency and non-emergency engine use?

Emergency use is the engine running because the utility failed or to maintain readiness for that failure. Non-emergency use is the engine running for any other reason, including selling power, shaving a demand peak, or earning money in a grid program. The distinction is the single most important line in the whole rule, because it decides which Tier the engine has to meet and how long it can run.

An engine classified as emergency standby gets a looser deal. Under the NSPS, emergency engines are generally allowed to meet a lower tier, often Tier 2 on a large set, instead of Tier 4 Final, and there is no hour limit on true emergency operation when the grid is actually down. That exemption is what lets data centers fill a yard with Tier 2 diesels that would never be permittable as non-emergency engines.

The catch is that the exemption is conditional, and it is easy to void. The moment the engine runs for a non-emergency reason beyond the allowance, the classification is in question, and an engine operated as non-emergency has to meet the non-emergency standard, which for new large engines is Tier 4. Run an emergency-rated Tier 2 engine for peak shaving and you have not just broken a paperwork rule, you have installed an engine that is the wrong Tier for what you are actually doing with it. This is the heart of the compliance problem, and it is where sites get caught.

How many hours can a standby generator run per year?

A standby emergency engine can run unlimited hours during a true emergency, but its non-emergency running is capped, commonly at 100 hours per calendar year for maintenance and testing under the NSPS. Within that 100 hours, a smaller slice, often cited at 50 hours per year, is allowed for certain non-emergency situations. Maintenance, testing, and readiness checks are what that budget is for.

The trap is what does not count as allowed non-emergency use. Peak shaving and non-emergency demand response, running the set to cut a utility bill or to sell capacity into a grid program, are generally not permitted within the emergency allowance, and using the engine that way pushes it out of the emergency category. If a site wants the engine in a demand-response or peak-shaving role on any regular basis, it is looking at a non-emergency engine and the Tier 4 obligation that comes with it, not a Tier 2 standby set.

These hour figures are federal defaults, and a site-specific air permit can set them lower. An air district can write a permit that caps testing hours well under 100, or limits total run hours across the fleet, as a condition of staying a minor source. The permit limit always governs over the federal default when it is stricter, so the number to design around is the one written into your permit, confirmed against the air district, not the number in a trade article.

Why emissions became the bottleneck for the AI data center

The AI buildout changed the scale of the diesel question. A large AI campus is not one or two gensets, it is dozens, sometimes scores, of multi-megawatt diesels backing a load measured in hundreds of megawatts toward a gigawatt. Each set is individually a modest source. Together they are a fleet, and the fleet is what the air agency adds up.

That aggregate is why emissions and permitting now gate sites that the grid and the gear could otherwise serve. A fleet of standby diesels large enough to back a hyperscale load can exceed a major-source threshold on paper even though every set sits idle most of the year. Permitting that fleet, defending it through public comment, and accepting the control conditions the agency demands can stretch the schedule past everything else.

Some jurisdictions have responded by raising the floor. Virginia's environmental agency has moved toward a Tier 4-equivalent control baseline for data center generators, and Illinois has set Tier 4 requirements for many new data center diesels, both landing around 2026. Several markets are weighing moratoriums or tighter review tied to diesel emissions. The on-site generation guide covers the cleaner alternatives operators reach for to sidestep this exact wall. Confirm the current rule in your jurisdiction, because this is moving fast.

Selective catalytic reduction for NOx

Selective catalytic reduction, SCR, is the aftertreatment that knocks down nitrogen oxides, and it is the reason a Tier 4 Final diesel can meet a NOx limit the engine alone cannot. SCR injects a urea solution, diesel exhaust fluid, into the hot exhaust ahead of a catalyst. The urea breaks down to ammonia, the ammonia reacts with the NOx across the catalyst, and what leaves the stack is mostly nitrogen and water. The NOx reduction is large, which is why SCR shows up in nearly every Tier 4 NOx control scheme.

SCR is not free to own. It needs a DEF tank, a dosing system, and exhaust hot enough for the catalyst to work, which is a real constraint on an engine that mostly idles or runs lightly loaded for tests. A standby set that rarely reaches temperature can struggle to keep the catalyst in its working window, and that is a known headache on emergency engines pressed into a Tier 4 role. DEF also has a shelf life and freezes, so the storage and the cold-weather plan are part of the design, not an afterthought.

When the air district sets Tier 4-equivalent control as the baseline, SCR for NOx is the piece they mean for the nitrogen oxides. Size the DEF storage and the dosing for the duty the permit actually allows, and confirm the control basis against the manufacturer's certified configuration.

Diesel particulate filter and oxidation catalyst for PM and CO

The diesel particulate filter, DPF, is the soot trap, and it is the PM half of a Tier 4 control package. The DPF captures particulate from the exhaust on a filter substrate, and the trapped soot is burned off periodically in a process called regeneration. On an engine that runs hard, regeneration happens passively from exhaust heat. On a lightly loaded standby set, the exhaust may never get hot enough, so the filter loads up and eventually needs active or forced regeneration, which is one more thing that does not love an engine that mostly idles.

The diesel oxidation catalyst, DOC, handles carbon monoxide and some of the hydrocarbons, oxidizing them across a catalyst, and it often sits in the same train as the DPF. Together the DOC and DPF address CO and PM the way SCR addresses NOx. A full Tier 4-equivalent baseline, like the one some data center jurisdictions now set, typically calls for SCR, DPF, and DOC together, one control for each of the three pollutants the permit tracks.

The recurring theme across all of this aftertreatment is the same: it was engineered for engines that run, and standby diesels do not run much. Plan the regeneration and the exhaust temperature for how the set will actually be used, and confirm the configuration against the manufacturer.

Ultra-low-sulfur diesel is the only fuel that works

Ultra-low-sulfur diesel, ULSD, is required for these engines, and it is not optional once aftertreatment is involved. ULSD caps sulfur at 15 parts per million, far below the older fuels, and the low sulfur is what makes the catalysts survive. Sulfur poisons an SCR catalyst and fouls a DPF, so running higher-sulfur fuel does not just nudge the emissions up, it can wreck the control equipment the permit depends on.

Beyond the catalyst, fuel quality is the quiet failure mode on standby diesels generally. Fuel sits in the tank for months between real runs, and untreated diesel grows microbial contamination and degrades, so the set that has not run since the last test can fail to start or run dirty when it finally has to. Fuel polishing and treatment are part of keeping a standby plant both reliable and compliant.

Specify ULSD, confirm the sulfur spec matches what the engine and aftertreatment are certified for, and keep the fuel-quality program running so the fuel in the tank on outage day is the fuel the emissions numbers assumed.

Do data center generators need an air permit?

Data center generators almost always need an air permit, and the size of the fleet decides which kind. The permit is issued by the state environmental agency or the local air district, not the EPA directly, and it sets the site's allowed emissions and operating conditions. The threshold question is minor source versus major source, because that one classification changes the entire permitting path.

A minor-source permit is the lighter track, available when the site's potential emissions stay below the major-source thresholds. A major source crosses those thresholds and triggers the heavy review: Prevention of Significant Deterioration, PSD, in an area that meets ambient air standards, or the more punishing Nonattainment New Source Review where the air already fails an ozone or particulate standard. Either way, major status pulls in best-available-control-technology requirements and a longer, public, far more expensive process. Above the major track sits Title V operating-permit status, which carries its own ongoing obligations.

The practical move is to find out early which side of the threshold the fleet lands on, because everything downstream, the schedule, the controls, the public comment, depends on it. Whether a site is minor or major is calculated from potential emissions and the local rules, and the determination is the air district's, so confirm it with them before you bank on a minor-source timeline.

How the fleet aggregates and trips a major source

Aggregation is the rule that catches data centers off guard, and it works like this: the air agency adds up the emissions of all the engines at a site, and sometimes across adjacent or commonly owned sites, to decide whether the whole operation is a major source. The fleet is evaluated together, not engine by engine. A genset that looks like a small minor source on its own can become part of a major source the moment it is counted with its forty neighbors.

The number that drives aggregation is potential to emit, the emissions the engines could produce running up to their permitted hours, not what they typically emit sitting idle. That is why a yard of standby diesels that almost never run can still trip a major-source threshold on paper. The agency permits the potential, and the potential of dozens of multi-megawatt engines adds up fast.

The defense is to plan for the aggregate from the start. Operators take federally enforceable permit limits, on hours, on emissions, or on the number of engines that can run at once, to cap the site's potential to emit below the major-source line and stay on the minor track. That is a deliberate strategy, written into the permit, not something you discover after the application comes back as major. Confirm the thresholds and the netting rules with the air district, because they vary by pollutant and by attainment status.

BACT and the controls a major source has to install

Best available control technology, BACT, is the control level a major new source has to meet, and it is determined case by case by the permitting agency. BACT is not a fixed number, it is the agency's finding of the most effective control that is achievable for that source, weighing cost and feasibility. Once a fleet is a major source, BACT is the lever that decides how clean the engines have to be.

For data center diesels, BACT findings have been converging on the Tier 4-equivalent package: SCR for NOx, DPF for PM, and DOC for CO. Some jurisdictions have gone further and set a presumptive BACT, a default control basis the agency expects every data center applicant to meet unless they can argue otherwise, which removes the case-by-case negotiation and just sets Tier 4-equivalent control as the floor.

The takeaway for design is that crossing into major-source territory is what forces the expensive controls. A minor source might permit Tier 2 emergency engines; a major source will be told to install full aftertreatment. That is one more reason the minor-versus-major determination is worth fighting for early, and why some operators cap the fleet specifically to stay minor. Confirm the controlling BACT basis with the air district, since presumptive BACT and its effective dates differ by jurisdiction.

Stack testing, CEMS, and how compliance gets proven

A permit is a set of promises, and the air agency makes you prove them with testing and monitoring. The common requirement is an initial stack test after startup, where a contractor measures actual NOx, PM, and CO out of the stack under load and shows the numbers fall within the permit limits. The permit usually specifies the test method and the conditions, and the engine has to pass it before it can be considered compliant.

Larger or higher-emitting sources can be required to run continuous emissions monitoring, CEMS, which measures the exhaust constantly rather than once. CEMS is more common where the engines run more or where the permit demands ongoing proof, and it carries its own calibration and recordkeeping load. Most standby data center diesels are not CEMS sources, but the permit decides, not the habit.

Underneath all of it is the non-resettable hour meter, which is the monitoring that proves the run-hour limits. The agency cannot verify a 100-hour cap without a tamper-proof count of the hours, so the meter is the foundation the whole emergency classification rests on. Plan the initial stack test into the commissioning schedule, because the engine is not done until it has passed it.

The non-resettable hour meter and the records that back the permit

A non-resettable hour meter is required on an emergency engine that relies on the run-hour limits, and it has to be installed before startup. The meter exists for one reason: to give the agency a count of run hours that cannot be wound back. Without it, the 100-hour and 50-hour limits are unenforceable, so the rule makes the meter a precondition, not an accessory.

The meter alone is not the whole recordkeeping obligation. The permit typically wants a log of every run, the hours, and the reason, split between emergency operation and maintenance, testing, or other non-emergency use, because the categories are capped differently. When an inspector asks how the engine has been used, the answer is the log, and a log that does not separate emergency from non-emergency hours cannot prove the engine stayed inside its limits.

This is exactly the kind of recurring field record FieldOS is built to capture: the run, the hours off the meter, the reason code, and the date, logged at the engine and tied to the asset so the annual total is always current instead of reconstructed from memory at audit time. Keep the running tally against the cap, not a pile of loose run sheets, because the cap is per calendar year and it does not forgive a December surprise.

Load-bank testing counts against the hours and the emissions

Load-bank testing exercises the engine under real load using a resistive load bank, and it is how a standby diesel gets the periodic loaded run it needs to stay healthy and to burn off the wet-stacking that comes from idling. It matters here because that test time runs the engine, which means it counts against the non-emergency hour budget and it produces emissions the permit tracks.

The tension is real. Aftertreatment, the DPF especially, wants hot, loaded running to regenerate and stay clean, so the engine benefits from load-bank time. But every load-bank hour eats into the same capped budget as routine testing, and on an emergency engine that budget is tight. Plan the load-bank schedule against the permit's hour limit, not just the maintenance interval, so a thorough test program does not quietly blow the cap. The load-bank mechanics are their own topic; here the point is only that the hours and emissions count.

Gas, fuel cells, and storage that avoid the diesel permit

The cleanest way to win the diesel emissions fight is to run less diesel, and several alternatives exist precisely to dodge the permit wall. Natural gas engines and turbines emit less particulate and can carry a different, sometimes easier, permitting profile, though they bring their own air and methane considerations. Fuel cells produce power electrochemically with very low criteria-pollutant emissions, which is why they show up as a primary-power option on emissions-constrained sites. Battery energy storage emits nothing at the point of use and can shave the role diesel would otherwise play.

None of these is a clean swap for every site. Gas needs a pipeline and has its own permit questions, fuel cells are capital-heavy and depend on fuel supply, and storage covers minutes to hours, not a multi-day outage, so it complements rather than replaces backup generation. But where the diesel fleet is what trips a major source or stalls in public comment, shifting part of the duty to a cleaner source can be what makes the site permittable at all. The on-site generation and microgrid guide covers these options and the economics in depth; here they are the escape hatch from the diesel permit.

Siting, noise, and the community that reads the permit

Emissions are not the only thing that draws fire at a data center genset yard; noise and siting ride along with them. A bank of large diesels firing for a test is loud, and the sound carries to whatever sits at the property line. Setbacks, enclosures, and exhaust silencers are part of the design where the site is close to homes, and the local noise ordinance can be as binding as the air permit.

Environmental-justice review has become a real gate. Where a proposed site sits near a community already carrying a heavy pollution burden, the agency can apply added scrutiny, attach conditions, or deny the permit on cumulative-impact grounds. Public comment periods give neighbors a formal channel to contest the application, and a well-organized opposition can add months even where the engineering is sound.

Treat siting, noise, and community engagement as part of the permitting work, not a separate problem handed to someone else after the engines are picked. The stack location, the test schedule, and the visible plume are all things the public sees, and the permit is where that gets decided.

Permit in hand before startup, every time

The hard rule at commissioning is that the air permit has to be in hand and its conditions met before the engine starts, including the first test run. Starting up a stationary engine without the required construction or operating permit is a violation in its own right, separate from any emissions exceedance, and it is the kind of mistake that turns a paperwork gap into an enforcement action.

The sequence that keeps a site clean is straightforward. The permit is issued, the non-resettable hour meter and any required monitoring are installed, the engine starts under the permit, the initial stack test is run and passed, and the run-hour log begins from the first hour. Skip the meter or run before the permit clears and the engine's compliance history starts with a violation that follows the site.

Build the permit milestone into the project schedule as a gate ahead of energization, not a parallel task that might catch up later. The commissioning agent should confirm the permit is final and its preconditions are satisfied before the first crank, because once the engine has run, you cannot un-run it.

Where compliance actually goes wrong in the field

The failures on these projects are not exotic, they repeat. The most common is an emergency-rated engine, a Tier 2 set, run in non-emergency use beyond what the exemption allows, which both breaks the hour rule and means the wrong Tier engine is installed for the actual duty. Right behind it is simply exceeding the run-hour cap, usually by not tracking the calendar-year total until it is already blown.

The recordkeeping failures are just as common and just as costly. No non-resettable hour meter, or a meter with no log that separates emergency from non-emergency hours, leaves the site unable to prove compliance even if it actually stayed inside the limits. The aggregate tripping a major source unexpectedly, because nobody added the fleet up against the threshold early, turns a minor-source timeline into a PSD project mid-stream.

The last two are the avoidable ones. Starting up before the air permit is final is a self-inflicted violation. Installing the wrong Tier engine for the local air district, a Tier 2 set where the jurisdiction now demands Tier 4-equivalent control, means buying the engine twice. Every one of these traces back to treating emissions as a back-office formality instead of the constraint that gates the site.

What to document

The emissions record is what answers the inspector, defends the permit, and proves the engine stayed inside its limits. Capture it as you go, because reconstructing a year of run hours from memory at audit time is how sites fail an inspection they would otherwise have passed.

Common mistakes

• Running an emergency-rated engine in non-emergency use beyond the allowance, which voids the exemption and means the wrong Tier engine is installed.
• Exceeding the run-hour cap by not tracking the calendar-year total until it is already blown.
• Installing no non-resettable hour meter, or keeping a log that does not separate emergency from non-emergency hours.
• Letting the aggregate emissions of the fleet trip a major source unexpectedly because nobody added it up against the threshold early.
• Starting up the engine before the air permit is final and its preconditions are met.
• Installing the wrong Tier engine for the local air district, such as a Tier 2 set where Tier 4-equivalent control is now required.

Field checklist

Standards and references

The federal framework lives in two EPA rules. The New Source Performance Standards for stationary compression-ignition engines, 40 CFR Part 60 Subpart IIII, ties the engine to the EPA Tier levels, splits engines into emergency and non-emergency, sets the run-hour structure, and requires the non-resettable hour meter before startup. The RICE NESHAP, 40 CFR Part 63 Subpart ZZZZ, covers hazardous air pollutants and applies depending on the engine and the source's major or area status. The EPA Tier standards themselves set the allowed NOx and PM by engine size and vintage.

The permit and the controls come from the state environmental agency or the local air district, which determine minor versus major source, write the run-hour and emission limits, and set best available control technology for a major source. Several jurisdictions have established a presumptive Tier 4-equivalent BACT for data center generators, with effective dates around 2026, so the controlling basis is local and current. The engine manufacturer's certification establishes the tier and the aftertreatment configuration the engine is approved to run.

The run-hour figures, the tier limits, and the subpart section numbers cited here are the federal defaults and the structure as of this review, and they shift across rulemakings and are overridden by stricter permit conditions. Confirm the run-hour caps against the EPA rule and your specific permit, confirm the controlling tier and BACT against the air district, and remember the two facts that catch sites most often: the aggregate fleet can trip a major source even when the engines rarely run, and the permit must be final before startup.

Units and terms

Emissions and permitting carry their own vocabulary, and the same idea shows up under several names across an engine datasheet, a permit, and an agency letter.

Engine emissions limits are given in grams per kilowatt-hour or grams per horsepower-hour. Pollutants are abbreviated NOx for nitrogen oxides, PM for particulate matter, CO for carbon monoxide, and HAP for hazardous air pollutants. The control equipment goes by its acronyms: SCR, DPF, DOC. The permit terms, PSD, NNSR, BACT, and Title V, each name a specific track or requirement, and potential to emit, PTE, is the calculated basis the major-source determination runs on.

Tier 4 Final

The EPA's most stringent engine emissions tier, typically requiring SCR and a DPF on large engines

Emergency vs non-emergency

The engine classification that decides the required tier and the run-hour limits

NSPS Subpart IIII

40 CFR Part 60 Subpart IIII, the federal standard for stationary CI engines

RICE NESHAP

40 CFR Part 63 Subpart ZZZZ, hazardous-air-pollutant standards for reciprocating engines

SCR / DPF / DOC

Selective catalytic reduction for NOx, diesel particulate filter for PM, oxidation catalyst for CO

PSD / NNSR / Title V

Major-source permitting tracks: deterioration review, nonattainment review, and operating permits

BACT

Best available control technology, the control level a major source must meet

Potential to emit (PTE)

The emissions an engine could produce at its permitted hours, the basis for major-source status

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• How do you size a data center generator?
• Why are data centers building their own power?
• What is an NFPA 110 acceptance test?
• How much fuel does a generator need?
• What are the conditions to parallel two generators?