Reciprocating gas engines fit data center sites from a few MW to a few tens of MW; gas turbines fit campuses of 100 MW and up. The deciding factor in 2026 is lead time: large-frame turbine slots are sold out through 2030 to 2031 (StackedAI analysis, Tier-2 conversion timelines, Jul 2026), while engine packages from about 1 MW are marketed as shipping years sooner. Unit size, M+1, and permitting follow.
How do gas turbines and reciprocating engines differ as data center prime movers?
Both burn pipeline gas to drive a generator and both can run continuously as prime power. Reciprocating engines are piston machines: units are typically single-digit MW, reach full output quickly from cold, hold efficiency at part load, and tolerate frequent starts, and their large installed base matters for parts, technicians, and OEM service agreements.
Gas turbines are rotating combustion machines, from a few MW for aeroderivative frames to hundreds of MW for large frames. They favor steady operation near rated output, lose efficiency at part load and in hot ambient conditions, and in combined cycle reach efficiencies simple-cycle engines cannot, an advantage that pays off only at scale. For an energy-first developer the question is which technology can be delivered, permitted, and financed on the date the site needs power, as set out in the energy-first campus development pillar.
Which can a data center actually get delivered in 2026?
Turbines are well documented. Prices are expected to reach about $600 per kW by the end of 2027, up 195% from 2019, with orders of roughly 100 GW against manufacturing capacity of 60 to 70 GW per year (Wood Mackenzie via Bloomberg, Gas turbine prices surge, Apr 2026). GE Vernova’s backlog reached 116 GW in Q2 2026 (GE Vernova, Q2 2026 results, Jul 2026). Large-frame slots are sold out through 2030 to 2031, and secured slots trade at about three times 2019 prices (StackedAI analysis, Tier-2 conversion timelines and Energy-First market review, Jul 2026).
The engine side is less documented. Vendors market prime-power engine and battery packages from about 1 MW as bypassing 3 to 7 year interconnection queues, which StackedAI records as a vendor claim it has not independently verified (StackedAI analysis, Energy-First market review, Jul 2026). The benchmark that matters is the 4.4 year average wait for a new large-load grid connection (Cushman & Wakefield, Global Data Center Market Comparison, May 2026); any prime mover shipping inside that window changes the economics. Fuel cells are the third path: Bloom Energy’s Q2 2026 revenue rose 165.5% year over year (Bloom Energy, Q2 2026 results, Jul 2026).
How do efficiency, part-load behavior, and M+1 granularity compare?
StackedAI’s campus-scale BTM model assumes a heat rate of about 7,400 Btu/kWh, roughly 46% thermal efficiency, which at $2.9 to $3.0 per MMBtu gas gives fuel cost of about 2.2 cents per kWh (StackedAI analysis, Energy-First BTM cost model, Jul 2026). A simple-cycle engine reaches that figure at the unit level; a simple-cycle turbine typically needs a larger frame or combined-cycle operation to match it.
Part-load behavior favors engines. A data center’s load ramps as halls fill and GPU clusters cycle, and a multi-engine plant modulates by switching units on and off, keeping each running unit near its efficient point; a plant of two or three large turbines cannot.
M+1 granularity is the sharpest difference, because the spare is the same size as the working units. A 20 MW critical load on 5 MW engines needs M = 4 and M+1 = 5 units, 25 MW installed and 20% spare; on 10 MW machines it needs M = 2 and M+1 = 3, 30 MW installed and 50% spare. Smaller units mean cheaper spare capacity and growth in steps that track leasing; the worked arithmetic is in M+1 redundancy for on-site power.
How do emissions and permitting compare, and which fits which site?
Both are stationary combustion sources under the Clean Air Act, each with its own New Source Performance Standard. Whether a plant also triggers New Source Review before construction or a Title V operating permit after it depends on potential to emit, a function of unit count, run hours, and controls rather than technology alone. Engines emit more NOx per unit of output before controls and rely on catalytic after-treatment; turbines rely on dry low-NOx combustion and, at scale, selective catalytic reduction. Turbine plants tend to be permitted once at major-source scale; engine plants are often permitted in phases, which helps a Tier-2 site stage capital but draws source-aggregation scrutiny. Thresholds vary by jurisdiction.
| Criterion | Reciprocating gas engines | Gas turbines | Source |
|---|---|---|---|
| Typical unit size | Single-digit MW | Tens of MW; large frames well above | General engineering knowledge |
| Best-fit scale | Few MW to a few tens of MW | 100 MW+ campuses | StackedAI framework |
| Lead time | From ~1 MW, marketed as bypassing 3–7 yr queues (vendor claim, unverified) | Slots sold out through 2030–31 | StackedAI analysis, Jul 2026 |
| Price trend | Not benchmarked in cited sources | ~$600/kW by end-2027, +195% vs 2019 | Wood Mackenzie via Bloomberg, Apr 2026 |
| M+1 spare share, 20 MW load | 5 MW units: 20% | 10 MW units: 50% | Arithmetic |
| Permitting posture | NSPS for engines; catalytic controls; often phased | NSPS for turbines; DLN plus SCR; permitted once | Conceptual |
For the 7 to 25 MW buy box (StackedAI analysis, Tier-2 conversion underwriting framework, Jul 2026) in StackedAI’s Tier-2 brownfield conversion work, engines with battery storage are the default bridge: they arrive inside the grid’s 4.4 year window, give cheap M+1, and become peaking capacity once the utility delivers. Turbines belong to campuses that can absorb a large unit and hold a slot secured years earlier. The cost side is in behind-the-meter power cost for data centers.
Key terms
- Reciprocating engine (RICE): a piston-driven internal combustion engine coupled to a generator; typically single-digit MW per unit in data center service.
- Gas turbine: a rotating combustion engine coupled to a generator; frames range from a few MW to hundreds of MW per unit.
- Heat rate: fuel energy consumed per unit of electricity produced, in Btu per kWh; lower is more efficient.
- NSPS: New Source Performance Standards, federal emission limits for classes of new stationary sources including turbines and engines.
How StackedAI applies this
StackedAI sizes on-site generation for Tier-2 targets from the leasing plan backward: load steps, the utility’s delivery date, and the tariff determine unit size and count, which determine the permit path. For 7 to 25 MW sites the firm models engine-and-storage packages in M+1 as the bridge and treats turbines as an option only where a secured slot exists. Lead times, heat rates, and pricing are benchmarked per engagement.
Frequently asked questions
Are gas turbines or reciprocating engines better for a data center?
Neither in the abstract. Turbines suit campuses of 100 MW and up, where a few large units run efficiently at steady load. Reciprocating engines suit sites from a few MW to a few tens of MW, where single-digit-MW units match load steps, M+1 spare capacity is cheaper, and equipment ships years earlier.
How long is the wait for a gas turbine in 2026?
Large-frame turbine slots are effectively sold out through 2030 to 2031 (StackedAI analysis, Jul 2026). Orders of about 100 GW are running against 60 to 70 GW per year of manufacturing capacity (Wood Mackenzie via Bloomberg, Apr 2026), and GE Vernova’s backlog stood at 116 GW in Q2 2026 (GE Vernova, Jul 2026).
How fast can reciprocating engines be delivered for a data center?
Vendors market prime-power reciprocating engine and battery packages from about 1 MW as bypassing 3 to 7 year interconnection queues. StackedAI records that as a vendor claim it has not independently verified, though it is consistent with the engine market’s larger installed base and shorter production cycle relative to large turbines.
What heat rate should a data center assume for on-site gas generation?
StackedAI’s campus-scale BTM model uses about 7,400 Btu/kWh, roughly 46% thermal efficiency, giving fuel cost of about 2.2 cents per kWh at $2.9 to $3.0 per MMBtu gas (StackedAI analysis, Jul 2026). Actual heat rate depends on unit, load point, and ambient conditions and is benchmarked per engagement.
Why does unit size matter for M+1 redundancy?
Under M+1 the spare is the same size as the working units. A 20 MW load on 5 MW engines carries one 5 MW spare, 20% of installed capacity; on 10 MW machines it carries a 10 MW spare, 50% of installed capacity. Smaller units make redundancy cheaper and let the plant follow load in finer steps.
Sources
- Wood Mackenzie via Bloomberg, Gas turbine prices surge, Apr 2026, https://www.bloomberg.com/news/articles/2026-04-01/gas-turbine-prices-surge-crimping-efforts-to-power-data-centers
- GE Vernova, Second Quarter 2026 Financial Results, Jul 2026, https://www.gevernova.com/news/press-releases/ge-vernova-reports-second-quarter-2026-financial-results-raises-2026-financial
- Bloom Energy, Second Quarter 2026 Financial Results, Jul 2026, https://investor.bloomenergy.com/press-releases/press-release-details/2026/Bloom-Energy-Reports-Record-Second-Quarter-2026-Financial-Results-and-Raises-Full-Year-2026-Guidance/default.aspx
- Cushman & Wakefield, 2026 Global Data Center Market Comparison, May 2026, https://ir.cushmanwakefield.com/news/press-release-details/2026/Dallas-Texas-Ranked-No–1-Primary-Data-Market-in-the-World-as-AI-Demand-Power-Constraints-and-Regulation-Reshape-CRE-Strategy/default.aspx
- StackedAI analysis, Energy-First BTM cost model, Jul 2026 (internal)
- StackedAI analysis, Energy-First market review, Jul 2026 (internal)
- StackedAI analysis, Tier-2 conversion timelines, Jul 2026 (internal)
- StackedAI analysis, Tier-2 conversion underwriting framework, Jul 2026 (internal)