Fuel-Supply & Gas-Process Engineering

Chapter 4.8 engineered the electron side of on-site generation: paralleling, synchronization, grid-forming inverters and the BESS bridge for transient acceptance. This chapter covers the molecule side: getting fuel to the prime movers reliably, in spec, and firmly enough to underwrite the availability the campus is sold on. By mid-2026 Cleanview counted 59 US data centers with behind-the-meter generation announced, about 90 GW in all; about 2 GW was operating, and 2.8–3.2 GW is expected by year-end. Announcing 500 MW of aeroderivative turbines is easy because the turbines are a known quantity; the gas is the schedule-defining work. The midstream operator must approve and build the pipeline tie-in, the gas utility must certify the custody-transfer meter, a let-down station must bring transmission pressure down to something the plant can use, and a conditioning skid must hold the fuel inside the turbine's inlet spec. The decision most often deferred is the firm-vs-interruptible delivery contract, and it decides whether the plant still runs on the coldest day of the year.

Every fork in the fuel chain has its own downstream cost: pipeline tie-in or virtual pipeline; firm, interruptible or synthetic-firm supply; boost compression or adequate inlet pressure; conditioning to the OEM-approved operating map, an approved alternate fuel, or stopping outside it; single-fuel or dual-fuel. The gas supply itself moves on other parties' clocks: the midstream operator, the air-permit authority and the local distribution company. Treat fuel as a commodity to sort out after the turbines arrive and you have mis-sequenced the critical path. → energy-supply strategy in Chapter 3.5; the electrical integration this chapter feeds in Chapter 4.8.

Trace the molecule from the transmission main to the combustor, and each link has its own lead time, failure mode and permitting hook. Pipeline tie-in: a hot tap or new lateral off an interstate or intrastate transmission line, which the pipeline operator must engineer, approve and schedule. It is often the longest-lead non-equipment item in the chain, because you are queuing for the midstream operator's construction calendar, not your own. Custody-transfer metering: the fiscal boundary where ownership and money change hands. It is built to AGA/API accuracy standards (orifice or ultrasonic meters, a gas chromatograph for composition, a flow computer for energy reconciliation) because every Dth is bought and billed. Pressure regulation / let-down: transmission gas arrives at 200–1,000+ psig, while the plant wants tens of psig at an engine and a controlled, higher pressure at a turbine. A regulator station with monitor-and-worker runs and slam-shut over-pressure protection does the job, and, because expansion chills gas, line heating keeps it above the hydrocarbon dew point. Conditioning: filtration, liquids knockout, and dew-point and Wobbe control hold the fuel inside the prime mover's inlet spec. Boost compression: required when the lowest pressure you can count on at the plant, once metering, regulation and conditioning have taken their drop, still sits under what the prime mover needs at full flow. Turbines hit this far more often than engines. Wärtsilä's own engine-versus-aeroderivative comparison shows the gap: 6 bar (87 psi) at the engine, nearly 50 bar (725 psi) at the aeroderivative turbine. Treat that as one maker's data, since other models differ. A compressor train is a rotating, redundant subsystem in its own right. Storage and changeover: on-site LNG/CNG or distillate for firmness, with dual-fuel logic to switch on a curtailment. Safety overlay: gas detection, relief, venting and ESD interlocks wrap the whole chain.

The tie-in is where the schedule risk lives. A gigawatt campus burning gas needs a transmission-grade interconnect, and that interconnect is the midstream operator's project to permit and build, gated by its own engineering queue, right-of-way and, depending on the jurisdiction, environmental review. The 'dirty little secret' that makes on-site gas attractive is that connecting a load to an existing, underused pipeline is often far faster and cheaper than connecting generation to a congested electric grid (Natural Gas Intelligence, 2025). But 'faster than a 5–7 year grid queue' is not 'fast', so sequence the tie-in as a critical-path item alongside the turbine order, not after it. The INGAA Foundation (March 2026) projects that US and Canadian gas pipeline capacity must grow about 39% from 2022 levels by 2052 to serve load growth, gas-fired generation included. That midstream buildout is itself a multi-year constraint, and data-center gas demand, which RBC Capital Markets forecast in May 2026 at about 6 Bcf/d by 2030, is competing for it.

Custody transfer is the fiscal and contractual boundary. The meter run, typically dual ultrasonic or orifice meters with an inline gas chromatograph and an AGA-compliant flow computer, establishes the volume and the energy content (Dth) that everything downstream is billed against. It is also the reference point for the Wobbe and heating-value spec the operator must hold. Get the metering wrong and you are mis-billed, and you also lose the trustworthy basis for the energy accounting that the air permit and the efficiency guarantees depend on.

The delivered price is a stack, and the hub price is only its first layer. The regional basis between Henry Hub and your receipt point sits on top, and it can run either way with pipeline takeaway and weather. Firm transport also carries a monthly reservation charge, due whether or not you burn any gas. Usage charges and fuel retention apply to every Dth shipped, and on-site let-down, conditioning and compression add their own parasitic power and upkeep. EIA's Short-Term Energy Outlook of October 6, 2026 forecasts Henry Hub at an average $3.48/MMBtu for 2026 and $3.16/MMBtu for 2027. Fuel is the largest operating cost of a gas plant. At a 9.0 MMBtu/MWh heat rate, every $1/MMBtu of delivered price adds $9/MWh; for a 500 MW plant running at an 85% capacity factor, that is about $34 million a year. Price the firm-or-interruptible choice on the delivered stack, not on the hub forecast.

Pressure let-down carries an underrated hazard. Transmission gas at several hundred psig must be regulated down in stages, with a monitor regulator backing the worker regulator and a slam-shut valve as the last defense against over-pressuring the plant piping. Gas cools as it expands (the Joule–Thomson effect, on the order of ~7 °F per 100 psi dropped), so a large let-down can push the stream below its hydrocarbon and water dew points. That drops liquids and can form hydrates that foul the conditioning train downstream. Large plants therefore usually get a heated let-down station, with a line heater sized to keep the regulated gas above its dew point. Skip the heating analysis and the cheap regulator station becomes the source of the liquids that damage a nine-figure turbine.

Prime movers do not burn 'natural gas' in the abstract. They burn fuel inside a defined window, and the conditioning skid holds it there. The governing parameter is the Wobbe index, the heating value divided by the square root of specific gravity, because it is the quantity a fixed combustor 'sees' as constant energy delivery through a fixed orifice at a fixed pressure. A gas turbine's combustor is tuned to a turbine-specific Modified Wobbe Index target (conventional volumetric Wobbe bands run ~40–55 MJ/m³), and manufacturers hold the allowable swing tight. The fuel specification for the combustion system you are offered must state its allowable Modified Wobbe Index range and change rate. A generic ±5% outer bound or a ±1–3% target does not authorize you to run the machine. Outside that window the consequences are not gradual: combustion dynamics, higher NOx, overfiring, flame instability and hardware distress that shortens hot-section life. A dry-low-NOx (DLN/DLE) combustor trades a narrow stability margin for low emissions, so its Wobbe tolerance is tighter still, and an out-of-spec excursion can trip the machine or violate the air permit at the same instant.

The conditioning train therefore does four things. Filtration / coalescing removes particulates and aerosols that erode and foul the fuel nozzles. Liquids knockout / scrubbing drops free liquids and, where supply is rich or variable, removes or controls the heavier hydrocarbons that raise the hydrocarbon dew point. Dew-point control keeps the fuel a margin above its hydrocarbon dew point at the burner, so nothing condenses as the gas expands across the fuel valve. That is why upstream let-down heating matters. The margin is whatever your OEM guarantees at your actual pressure and composition; a quoted 20–28 °C or 50 °F figure is not a universal acceptance criterion. Wobbe / heating-value trim blends or trims on sites with variable supply to hold the index, and on aeroderivatives fuel-gas heating brings the inlet to the temperature the combustor map specifies. The reciprocating-engine fleets (Wärtsilä, INNIO Jenbacher, Caterpillar, Bergen Engines, and Everllence, formerly MAN Energy Solutions) tolerate more fuel variability and lower inlet pressure than turbines. That is one reason engines have won so much of the data-center bridge-power market: they need less conditioning and compression. Linear generators are a newer option with the same low-pressure, fuel-flexible appetite. Mainspring's 2.5 MW block takes any blend of natural gas, biogas, hydrogen or propane at 5–20 psig and lists 46% electrical efficiency (LHV, net AC). In October 2026 AEP signed for up to 1.6 GW of them through 2034, starting with 190 MW for delivery from 2027 (Mainspring, 13 October 2026). → prime-mover selection and the engine-versus-turbine tradeoff in Chapter 3.5; their electrical sizing in Chapter 4.8.

Reciprocating engines accept gas at modest pressure, often within the let-down station's output, so a particular engine offer may need no on-site compression, provided the minimum dynamic supply pressure still leaves enough inlet pressure after conditioning losses. Aeroderivative and industrial turbines are the opposite. Their fuel systems demand a high, stable inlet pressure (frequently several hundred psig, machine-dependent). If pipeline pressure at the meter is below that figure, the operator has acquired a boost-compressor train and all the consequences of rotating machinery on the critical fuel path. A boost compressor is not a passive skid. It is a screw or reciprocating compressor with its own driver, lube system, aftercooler, pulsation control and controls. Compression heats the gas, so the aftercooler outlet must be checked against the pressure-dependent hydrocarbon and water dew-point limits and reheated as the turbine OEM requires. The train draws parasitic power, has its own maintenance and overhaul cycle, and, most important, sits on the path between the pipeline and every turbine on site.

That makes redundancy a first-order decision. A single boost-compressor train is a campus-scale single point of failure: lose it and the turbines starve however much gas is in the pipeline. So the train runs N+1 (or 2N for the most availability-critical sites), with the spare sized to carry full load and a fast, tested changeover. That costs capex, footprint and lead time. Treat the fuel-gas booster package as its own procurement milestone and get a dated quote from the packager. As a market marker, Kodiak Gas Services' August 2026 investor presentation puts new Caterpillar 3600 inline gas-compression engines at over 180 weeks. That figure covers the driver only; the booster package you are offered carries its own schedule. The siting choice is stark. Where line pressure already clears the turbine inlet spec, you avoid the compressor entirely. Where it does not, you inherit a redundant, parasitic, long-lead rotating subsystem. Some operators therefore favor reciprocating-engine fleets despite their lower per-unit output: the engine choice can delete boost compression where the guaranteed inlet pressure allows it, at the cost of more units to install.

When firm pipeline transport is unavailable or insufficient, fuel arrives by truck. This 'virtual pipeline' of trucked CNG and LNG, and the on-site storage that buffers it, is the physical form of the availability target. The two media trade energy density against logistics radius. CNG stores natural gas as compressed gas in carbon-fiber-wrapped cylinders, a 'battery of gas'. Its economic haul radius depends on each module's usable energy, round-trip time, unloading rate and fleet size, so size it with the burn-and-inventory arithmetic used below. At an assumed 9.0 MMBtu/MWh (HHV), 100 MW for 24 h needs 21,600 MMBtu of usable inventory, or 24,000 MMBtu gross at 90% deliverability. Divide the 21,600 MMBtu usable requirement by the HHV energy one offered module is guaranteed to deliver, and you have the on-site module count. Modules being loaded, in transit or turning around come on top of that. LNG is far denser. Before you choose a haul radius or a multi-day inventory, compare each offered LNG or CNG trailer’s usable HHV energy, unloading rate, round-trip time and delivered cost. LNG buys that density with a cryogenic subsystem: an insulated storage tank near −162 °C, a vaporization train (ambient-air or fired vaporizers) sized to the plant's peak burn, boil-off management, and an odorization and gas-detection basis confirmed at the battery limit. LNG is odorless, while pipeline gas may arrive odorized or unodorized depending on its service and the governing rules. Vaporizer capacity, not tank volume, is often the limit: a tank holding ten days of gas is useless if the vaporizers cannot regasify fast enough to feed full plant load.

How much storage you build is how much availability you buy. Hours of autonomy ride out a momentary pipeline upset; days ride out a multi-day winter curtailment, the scenario that actually threatens a synthetic-firm design. Size the usable stored energy to cover the correlated-curtailment case you selected. Then run a separate check: the sendout your vaporizers can deliver has to keep up with the plant's fuel flow for the whole event. Autonomy sets tank volume, which sets the permitted footprint and process-safety classification, which sets the EHS and commissioning burden. You can add tanks and sendout equipment later, but whether that is feasible, and at what cost and schedule, turns on spare land, separation distances, containment, permit amendments and how the new tanks tie in. Hold footprint and expansion connections for the load ramp, and plan vaporizer capacity and tank volume as one expansion. → density-ramp and reserve-the-headroom discipline in Chapter 1.1.

Stored energy covers the missed delivery only while the surviving fuel train can deliver it. Fuel demand is 23 MW × 9.0 MMBtu/MWh = about 210 MMBtu/h. The 72 h event burns about 15,000 MMBtu, which at 90% deliverability needs about 17,000 MMBtu gross. The 18,000 MMBtu store holds about 16,000 MMBtu usable, enough for 78 h autonomy, and either 250 MMBtu/h train alone carries the burn. Two trains that share one pump, power supply or regulator count as one train.

The margin is thin. The load crossover is usable inventory ÷ (72 h × 9.0) = 25 MW: at 26 MW the inventory lasts only 69 h, even though one train still covers the roughly 230 MMBtu/h burn. A less efficient offer at 11 MMBtu/MWh burns about 253 MMBtu/h at 23 MW, so the store lasts 64 h and one train no longer covers the burn. If only 80% of the store is deliverable, it lasts about 70 h. The fix is more usable inventory or less retained load, never an uncontracted truck expected to arrive during a regional curtailment.

A 24 h resupply lead time consumes about 5,000 MMBtu usable before any margin, and that is the reorder point the fuel operator must turn into a contract and an alarm. Commissioning inherits the OEM fuel spec, the heel and boil-off calculation, the cold-ambient flow guarantee and the common-cause assessment. Fuel quality, changeover stability and delivered flow stay unproven until someone witnesses them. The calculation does not predict an availability percentage, and it does not license out-of-spec fuel.

The last line of fuel firmness is the ability to switch fuels. Dual-fuel prime movers, both reciprocating engines and many turbines, run on natural gas as the primary fuel and fall back to distillate (ULSD / No. 2 fuel oil) from on-site tanks when gas is curtailed. Changeover is a non-trivial engineering event. On a turbine it is a transfer that must hold flame stability and stay inside emissions limits through the swing, and the machine derates and emits differently on liquid fuel, so the air permit must cover distillate hours explicitly. Distillate storage is the logistics problem the diesel-standby world has always solved: tanks, containment, fuel polishing to keep stored fuel from degrading, and a haul contract for replenishment. It is the firmness backstop because liquid fuel in a tank on the property is the one supply no upstream counterparty can curtail.

It is a cost-and-emissions tradeoff. Single-fuel (gas only) is cheaper, simpler and cleaner, but its worst-case availability is only as good as the gas supply behind it. Dual-fuel buys firmness at the cost of distillate storage, changeover engineering, fuel-quality maintenance, and an air permit that must now cover a dirtier secondary fuel and the transient of switching. For a campus whose revenue rides on uptime, the distillate backstop is often the difference between 'firm on paper' and 'firm in a February cold snap.' → EHS for on-site generation and fuel-process safety in Chapter 6.9.

Every link above moves a flammable, asphyxiant, high-pressure (and, for LNG, cryogenic) fluid, so a process-safety overlay wraps the whole chain. Gas detection uses point and open-path methane detectors voting into the control system. They trigger alarms and, at high-high concentration, an emergency shutdown (ESD) that isolates the affected supply zone. Isolation zones and trip scope come from the hazard analysis and the cause-and-effect matrix: a localized gas or fire event need not isolate every zone, while a fault whose consequences cross a skid boundary can require broader isolation. Relief and venting: every blockable section that can be over-pressured needs a relief path (pressure-safety valves discharging to a safe vent or, on larger plants, a flare or vent stack designed for dispersion), with the let-down station's slam-shut as the primary over-pressure defense. Interlocks: a cause-and-effect matrix ties detection, fire-and-gas and process upsets to the ESD logic, fail-safe by design, so any credible upset drives the system to a safe state rather than venting fuel toward an ignition source. LNG adds cryogenic hazards: brittle fracture and rapid phase transition. Oxygen displacement is a hazard with natural gas in its gaseous form as well as with LNG vapor, most of all where a release can collect in an enclosed space.

The decision that can erase the schedule advantage is regulatory. A large on-site gas plant frequently crosses the threshold of a major stationary source under the Clean Air Act (PSD/Title V, GHG reporting, BACT/LAER). Federal OSHA PSM / EPA RMP applicability depends on material use and jurisdiction, not inventory alone: PSM excludes hydrocarbon fuel used solely for workplace consumption unless it is part of another covered process, and RMP excludes listed flammables used as fuel. Crossing those lines pulls in air permitting, a formal process-hazard analysis, mechanical-integrity programs and a commissioning regime that can take longer than the turbine procurement it was meant to outrun. So settle two questions before you order equipment. The first is source classification, based on source-wide potential emissions with aggregation. The second is PSM/RMP applicability, based on material use, the exclusions and the defined process. Lower-emitting prime movers such as fuel cells can change the air classification; dividing one connected process among smaller vessels does not remove PSM coverage.

Codes, not only permits, govern how the fuel train is built and commissioned. NFPA 56 covers cleaning and purging flammable-gas piping, new or existing, and it forbids using flammable gas to clean pipe. It exists because a natural-gas blow to clean new piping at the Kleen Energy power plant in Middletown, Connecticut, exploded in February 2010 and killed six workers (U.S. Chemical Safety Board). Cleaning and first purge of a new plant's gas piping is when that hazard returns, so the purge procedure belongs in the commissioning plan. On-site LNG storage falls under NFPA 59A as the state or local fire code adopts it, and its thermal-radiation and vapor-dispersion siting distances set how much land the tanks need. PHMSA's 49 CFR Part 193 excludes LNG facilities used by an ultimate consumer, so it applies only where the LNG plant is part of a gas pipeline system, such as one a utility owns. Engine and turbine installations fall under NFPA 37, and the lateral and meter station the pipeline operator or utility runs usually under 49 CFR Part 192. Name the governing edition of each in the basis of design. → air permitting and the critical path in Chapter 3.9; EHS and PSM in Chapter 6.9; commissioning the integrated plant in Chapter 13.4.