A 100 MW bridge plant run for five years costs $96 per MWh on solid oxide fuel cells, $104 on gas reciprocating engines and $112 on aeroderivative gas turbines. That is at 85% utilization with gas delivered at $3.78/MMBtu, and the ranking holds at every gas price from $2 to $8 and at every utilization from 40% to 95%. It flips twice, and only twice: take away the 30% fuel-cell investment tax credit and fuel cells go to $128 and lose to engines by $24; price the turbines at their 2023 installed cost of $1,606/kW instead of what GE Vernova charges for an aero package in 2026 and turbines win at $84. The rest of the spread is lead time, which is not in the $/MWh at all: 17 months to first power for fuel cells, 27 for engines, 30 for turbines. The Open Factory Datacenter MEP Cost Table carries all of it, with every input and its source in one CSV.
We cover the three lead times and where they come from, what each technology costs per kW installed after the 2026 repricing, heat rate and fuel at Henry Hub plus basis, fixed and variable O&M, the air-permit route each one takes at 100 MW, resale at year five, the five-year total and $/MWh with sensitivities to gas and utilization, and behind the wall, the technology to buy in Texas, Virginia, Ohio, Arizona and Georgia and the price book of named deals.
Source: Open Factory Lead-Time Monitor (compiled from CPS Energy at PowerGen 2026 via Avanza, EPRI via Utility Dive, GE Vernova and Siemens Energy statements, JLL 2026 outlook; as of September 2026)
Fuel cells at 16 to 18 months, engines at 19 to 30 and aeroderivative turbines at 18 to 36 all beat the 36-month grid connection in Dallas or Phoenix and beat Columbus and Northern Virginia, at 84 months, by four to five years.
Three Lead Times, Three Kinds of Factory
The lead times come from a utility planner, not a vendor. Benjamin Jordan of CPS Energy put the delivery, installation and commissioning timeline at 50 MW-plus scale at 16 to 18 months for solid oxide fuel cells and linear generators, 24 to 30 months for large reciprocating engines and 30 months or more for gas turbines at PowerGen International in early 2026, against 36 to 84 months for a grid connection. Those are whole-project numbers. Bloom Energy’s own headline is faster for one machine: it delivered a fully operational system to Oracle in 55 days, a month ahead of a 90-day plan, in 2025. A hundred megawatts is a different job, with gas interconnection, switchgear and a substation, which is why the Lead-Time Monitor carries the 16 to 18 month figure and not the 55 days.
The three factories behind those numbers are in different states. Bloom is expanding Fremont from 1 GW to 2 GW of annual capacity by the end of 2026 and says each further 1 GW increment takes six to nine months and $100 million to $150 million of capital. Its Q2 2026 revenue was $1.065 billion, with product revenue up 215% to $935 million, and full-year guidance is $3.9 billion to $4.2 billion. Fuel cells are the only one of the three technologies where the supplier’s capacity is growing faster than its backlog.
Engines are the opposite. Caterpillar’s firm backlog reached $72.1 billion at June 30, 2026, and its large reciprocating engine backlog had grown more than 3.5 times since January 2024. Evercore’s David Raso told NGI in May that new large engine orders land in late 2027 at the earliest, with most slotting into 2028; Caterpillar is targeting 65 GW a year of engine capacity, Cummins 55 GW of high-horsepower output by 2030. Our Cummins and Caterpillar Are Sold Out Through 2028 piece has the quarter-by-quarter backlog; the Book-to-Bill League Table updates it each quarter.
Turbines split by size. Wood Mackenzie counted 110 GW of global gas turbine orders in 2025 against 60 to 70 GW of manufacturing capacity. GE Vernova says heavy-duty lead times are about three years, with roughly 10 GW of slots left across 2029 and 2030; Siemens Energy says three years or more; Mitsubishi’s June-quarter orders deliver between 2028 and 2030. A heavy-duty frame is not a bridge. EPRI’s Bobby Noble puts the small turbine at 18 to 36 months, against about five years for a large one, and notes that 70% of the turbine units ordered in 2025 were under 100 MW. The bridge turbine is an aeroderivative: two GE Vernova LM6000 PF+ Sprint units at 56.9 MW and 8,328 Btu/kWh LHV, or three LM2500XPRESS at 34.5 MW each, the package GE Vernova is shipping to Crusoe 29 at a time. The $60 Million Deposit piece covers what it costs to hold a heavy-duty slot; this piece is about not needing one.
What a Kilowatt Costs After the 2026 Repricing
Source: Wood Mackenzie (Apr 2026); BNP Paribas via Utility Dive (Jul 2026); Sargent & Lundy for EIA (Dec 2023); EIA AEO2025; AEP 8-K (Feb 2026); Bloom Energy 10-K (Feb 2026); NRG via Utility Dive (Aug 2026)
The eleven price points run from $600/kW for a heavy-duty turbine as equipment to $7,291/kW for a 10 MW fuel-cell plant installed, and the three we use for the bridge sit at $2,240, $2,670 and $3,003.
Start with the turbine, because it is the one where the 2023 reference cost is no longer true. Sargent & Lundy’s December 2023 case for EIA prices a 4 x 54 MW aeroderivative simple-cycle plant at $339.0 million, or $1,606/kW, of which $155.9 million, $739/kW, is the turbines and their SCR and CO catalysts. BNP Paribas now estimates GE Vernova gets ~$1,800/kW for an aeroderivative and ~$790/kW for a heavy-duty machine (July 2026). Swap the equipment line and hold everything else at 2023 levels and the plant is ~$2,670/kW. The aero turbine’s 2026 price alone is higher than the 2023 cost of the whole plant it sits in. We use $2,670 as the base, $1,606 as the low case and EPRI’s ~$3,000/kW top of the range for turbine plants as the high.
Engines have no public data-center price. Wartsila’s 412 MW order of 40 x 34SG engines for an Ohio hyperscaler, Caterpillar’s 1.5 GW first phase of 636 G3520K gensets for Joule in Utah and INNIO’s 1.25 GW framework with Rehlko all come without a dollar figure. So the base is EIA’s $2,240/kW for a 21 MW internal combustion plant in 2022 dollars, with a range of $1,800 to $2,800 that is ours. If a reader has a 2026 engine quote, the Equipment Price Benchmark is where it belongs; the cell will print when it has eight.
Fuel cells have the cleanest price in the set. AEP’s fourth-quarter 2025 release describes a Wyoming generation project backed by $2.65 billion in fuel cell purchase commitments, and Bloom’s 10-K describes the same agreement as up to 1 GW with an initial 100 MW order: ~$2,650/kW of equipment. Installation is a separate revenue line at Bloom; in 2025 it was $204.1 million against $1,531.3 million of product revenue, 13.3%. Installed is therefore ~$3,003/kW, and after the 30% ITC, ~$2,102/kW, the lowest net capital of the three. The Datacenter MEP Cost Table carries the gross number on the equipment line and the credit on a separate line, because the credit has conditions the equipment does not.
Heat Rate and Fuel: Where Fuel Cells Earn Their Price
Source: Open Factory Datacenter MEP Cost Table (compiled from EIA AEO2025 and Sargent & Lundy cost cases, BNP Paribas via Utility Dive, AEP and Bloom Energy SEC filings, EIA STEO; as of September 2026)
At $3.78 gas the fuel line is $36/MWh for turbines, $31 for engines and $24 for fuel cells; at $6.00 it is $57, $50 and $39.
Heat rates are on a higher heating value basis throughout, because that is how gas is billed. The aero plant runs at 9,447 Btu/kWh net in the Sargent & Lundy case, which includes SCR back-pressure and plant loads; GE Vernova’s 8,328 LHV for the LM6000 PF+ is ~9,230 HHV before those losses. The engine plant is 8,295 Btu/kWh in EIA’s table; an engine-plant developer told NGI new machines run as low as 7,200, which we carry as the low case. Fuel cells are 6,469 in the same EIA table, 52.7% HHV. Bloom’s marketing claim of 60 to 65% fuel-to-electricity, as relayed by the CPS Energy presentation, is an LHV figure that converts to ~6,300 HHV; we run it as the low case only.
Source: EIA Henry Hub spot price via FRED (MHHNGSP) and EIA Short-Term Energy Outlook, September 2026
The monthly average went from $4.26 in December 2025 to $7.72 in January 2026 and back to $2.78 by August; the September STEO puts the 2027 annual average at $3.28/MMBtu, down from $3.43 for 2026.
Our base delivered price is that $3.28 plus $0.50 for firm transport, and the $0.50 is an assumption, not a measurement. The measured number is the state basis. EIA publishes monthly citygate and industrial prices by state; over July 2025 to June 2026 the citygate ran $0.28/MMBtu above Henry Hub in Texas, $2.26 above in Virginia, $2.37 above in Ohio (six months published), $0.59 below in Arizona and $1.26 above in Georgia. Those five numbers, not the $0.50, drive the paid section. A 100 MW plant at 85% and 8,295 Btu/kWh burns ~6.2 million MMBtu a year, so each $1 of basis is $6.2 million a year on engines and $4.8 million on fuel cells. The Genset TCO tool takes the same heat-rate-times-price arithmetic for a single machine.
Permitting: 1,136 Tons of NOx Versus Zero
Source: Open Factory Datacenter MEP Cost Table (compiled from 40 CFR 60 Subparts JJJJ and KKKK, 40 CFR 52.21 and 51.165, Sargent & Lundy for EIA Case 3, Bloom Energy 10-K; as of September 2026)
At the federal floor, 100 MW of lean-burn engines emits 1,136 short tons of NOx a year and 100 MW of aero turbines 447; with SCR the engines fall to ~112 and the turbines to 26, against thresholds of 250, 100 and 50 tons depending on where the plant stands.
The floors are federal new-source performance standards. A non-emergency lean-burn gas engine of 500 HP or more built after July 1, 2010 must meet 1.0 g/HP-hr of NOx, or 82 ppmvd at 15% oxygen under Subpart JJJJ; at 0.746 kW per horsepower and a 97% generator, that is ~3.05 lb/MWh. A new gas turbine between 50 and 850 MMBtu/h must meet 25 ppm, which the rule itself equates to 1.2 lb/MWh, and 15 ppm or 0.43 lb/MWh above 850 MMBtu/h. Sargent & Lundy’s aero case carries SCR and predicts 0.0075 lb/MMBtu, 0.071 lb/MWh at its heat rate. Bloom’s 10-K says its Energy Server has a “negligible impact on air quality”, that most states require permits only for its larger installations, and that New York, New Jersey and North Carolina have specific air-permit exemptions for fuel cells. It publishes no lb/MWh figure in the filing, so the chart shows zero and the notes say why.
The thresholds decide the route. In an attainment area a new plant that is not one of the 28 listed source categories becomes a PSD major source at 250 tons per year of any regulated pollutant; a simple-cycle turbine or engine plant is not a “fossil fuel-fired steam electric plant,” so 250 applies, not 100. In an ozone nonattainment area the major-source threshold drops to 100 tons per year, and to 50 in a serious area. A 100 MW engine plant with 90% SCR is still a major source in Phoenix, Northern Virginia, Dallas and Houston, and a 100 MW aero plant with SCR is a minor source in all four. Major means LAER or BACT, offsets in nonattainment areas, public notice and a review that Sargent & Lundy budgets inside an 18-month development phase. Minor means a permit by rule or a standard permit. Texas has issued minor permits to at least 38 data centers since 2024, covering more than 2,100 diesel generators, including Stargate’s 10 turbines and 62 gensets, and one Vantage site permitted at 99.8 tons against a 100-ton line. The third route is the one xAI took in Southaven, Mississippi: 27 portable turbines totaling 495 MW without an air permit under the state’s reading of the mobile-source provisions, now in federal court with the Department of Justice intervening on xAI’s side. We do not model that route. The Large-Load Policy Tracker follows the case.
Five-Year Cost and Resale
Source: Open Factory Datacenter MEP Cost Table
Over 3.72 million MWh the turbines cost $417 million, the engines $388 million and the fuel cells $357 million; capital net of resale and credits is $46, $42 and $48 per MWh, fuel $36, $31 and $24, and O&M $7, $12 and $5.
The method is deliberately plain. Installed cost times 100,000 kW, less the ITC where it applies, is the capital; we charge interest on it at Moody’s Baa yield of 6.32% (August 2026) for five years and give back a residual at year five. Fuel is MWh times heat rate times $3.78. Fixed O&M is Sargent & Lundy’s $9.56/kW-year for the aero plant and EIA’s $39.57 for engines and $34.65 for fuel cells; variable O&M is $5.70, $6.40 and $0.66 per MWh, the turbine figure including hours-based major maintenance. Bloom sells maintenance agreements of 5 to 20 years separately, and its service gross margin was 13.3% in Q1 2026, so the $34.65 is a floor rather than a quote. Every one of those inputs, with low and high cases and a source URL, is in the assumptions table.
Resale is the assumption we are least sure of and the one we flag loudest. Dealers report that low-hour used gensets now sell at 60 to 75% of new when new delivery is a year out (August 2026), but a bridge engine runs ~37,000 hours in five years, so we take 30%. Aero turbines have a real secondary market: PROENERGY’s 13 x PE6000 packages for Crusoe are rebuilt LM6000-class machines at 50 MW each. We take 35%. Fuel cells have no published secondary market and their stacks are consumed; we take 10%. Moving all three residuals to zero adds $25/MWh to turbines, $18 to engines and $8 to fuel cells and leaves the order unchanged.
Two lines are left out on purpose. We do not add a gas interconnection or a substation, because all three technologies need both and Sargent & Lundy prices them at $3.95 million and $2.4 million for a 211 MW plant with a half-mile lateral, about $30/kW, which does not move the ranking. We do not add the utility’s minimum bill on the bridged load either, because it depends on the tariff; the Large-Load Cost Table carries it by market, and in AEP Ohio it is 85% of contract demand for up to 12 years.
Source: Open Factory Lead-Time Monitor and Datacenter MEP Cost Table
From a September 2026 order, fuel cells run 23 months before 2030 and deliver ~1,430 GWh, engines 13 months and ~810 GWh, turbines 10 months and ~620 GWh; at the Virginia industrial price of 9.31 cents the 810 GWh gap between fuel cells and turbines is ~$75 million of electricity.
That is the number the $/MWh table hides. A bridge exists because the grid is late, and a bridge that arrives 13 months after another bridge is 13 months of a datacenter or a plant not running. The Lead-Time-Adjusted Schedule tool puts the PO date and the gear class in and returns the first-power month, and the Large-Load Interconnection Cost tool prices the utility side of the same wait.
Sensitivity: Gas, Utilization and the Two Flips
Source: Open Factory Datacenter MEP Cost Table
At $2 gas the three options cost $95, $89 and $85 per MWh; at $8 they cost $152, $139 and $123, so the fuel-cell advantage widens from $10 to $29 as gas rises.
Source: Open Factory Datacenter MEP Cost Table
At 40% utilization the costs are $191, $179 and $176 per MWh; at 95% they are $105, $97 and $89, and the fuel-cell margin over engines narrows to $3 at the low end because its net capital per kW is the smallest of the three and utilization punishes capital.
Utilization is the input the buyer controls least. A datacenter campus ramps, so a 100 MW bridge ordered for a 2028 first phase runs at half load until the second hall fills, and every month at 50% costs ~$140/MWh instead of $96 to $112. Engines and fuel cells are modular, forty 2.5 MW gensets or a field of power modules, so the second half of the order can ship a year after the first; Oracle’s 1.2 GW is being deployed through 2027 in tranches for the same reason. Two aero turbines cannot be bought by halves without giving up the second machine’s slot. The right comparison for a phased site is therefore the 60% row for turbines against the 85% row for the other two, which widens the gap by another $30/MWh; the Genset TCO tool runs the phased case unit by unit.
The ranking flips in two cases and both are about capital, not gas. First, the credit. Section 48E after the July 2025 budget act gives fuel cells a 30% ITC without regard to emissions, with no prevailing wage or apprenticeship requirement and no adders, for projects beginning construction after 2025, phasing down only for construction starting in 2034. Without it, fuel cells cost $128/MWh and engines win by $24. The credit is transferable, so a developer without tax appetite sells it; the FEOC supply-chain rules apply to projects beginning construction after 2025 and Bloom manufactures in California and Delaware. Our 30% Fuel-Cell ITC piece walks through the eligibility test.
Second, the turbine price. At Sargent & Lundy’s 2023 installed cost of $1,606/kW the aero plant is $84/MWh and wins outright. At BNP’s 2026 equipment price it is $112 and loses to both. A buyer holding a 2024 aero quote, or a rebuilt PE6000-class package at a rebuilt price, is in the first world; a buyer calling GE Vernova today is in the second. The Quote Check tool prints the band a turbine or engine quote should fall in for its date, and the Bid-Tab Price Book will add public turbine and engine awards as they post.
Source: EIA-860M (July 2026) via Open Factory Lead-Time Monitor
EIA’s July 2026 planned-generator file holds 487 natural-gas units and 68.1 GW, of which 5.6 GW of simple-cycle turbines and 0.34 GW of engines are under construction or complete, and just 1.0 GW of engines in any status, against more than 10 GW of announced US datacenter engine orders at Wartsila, INNIO and Caterpillar alone.
That gap is the behind-the-meter market not yet reporting. The EIA-860M survey catches utility and merchant generators; the engine plants in Utah, West Virginia, Ohio and Texas will appear as they file, or not at all if they never export. Our Behind the Meter, In Front of the Queue forecast tracks the announced GW against the surveyed GW each quarter, and the Datacenter Electrical Roadmap 2026 to 2032 carries the resulting demand for switchgear and transformers, since every one of these plants needs a generator step-up transformer at 144 weeks.
What to Do Monday
Pick the technology by the two things this model cannot average away: the credit and the air permit. If the project entity can use or sell a 30% ITC and the site is in a moderate or serious ozone area, the fuel cell is the cheapest bridge and the fastest, and the PO goes to Bloom or an AEP-style utility channel this month, because Fremont’s 2 GW comes online at the end of 2026 and Oracle has already contracted 1.2 GW of it. If the credit is not usable, the engine is the cheapest bridge, and the order goes in now for a late-2027 or 2028 slot with a written SCR guarantee and a permit strategy that keeps the plant under 100 or 250 tons. The turbine is the right answer only when someone is offering 2023 equipment, rebuilt or otherwise, or when footprint rules out forty engines. Write the delivery month, the heat rate at site conditions and the emissions guarantee into the purchase order, with liquidated damages that bite, and ask for the resale-support clause in the same document; the RFQ for capital equipment template has the language. Then check the state you are in, because the basis and the permit thresholds change the answer in three of the five markets below.
Behind the paywall: the technology to buy in Texas, Virginia, Ohio, Arizona and Georgia given each state’s citygate basis, ozone classification and grid wait, where Ohio gas costs $2.37/MMBtu more than Henry Hub and adds $12 to $18/MWh to every bridge; and the price book of 22 named deals and reference costs from GE Vernova, Wartsila, Caterpillar, INNIO, PROENERGY, AEP and Bloom, in the Datacenter MEP Cost Table.