The grid can’t supply the power AI factories demand. Gas turbine manufacturers can’t keep up either. Some of the larger gas turbine models have waiting lists until the end of the decade. How about nuclear energy?
Great hope is placed in small modular reactors (SMRs) as a means of providing large amounts of power for data centers. But they won’t be broadly available for at least five years. Where else can hyperscalers and data center developers turn for power?
Why developers are turning to gas engines
Enter gas reciprocating engines, which are now finding their way into massive AI data centers.
These engines are essentially the same type as you find in modern cars — internal combustion engines. Traditionally, they are small machines that provide less than 1 MW of power. The bulk of the market continues to produce such machines. But a few models are now in the 3 MW to 20 MW range. By tying together a whole lot of them, some developers hope to provide more than 1 GW of power to their AI factories.
Caterpillar, for example, received a 2 GW order for the Monarch Compute Campus in West Virginia. The company will supply a series of Cat G3516 Fast Response natural gas generator sets to American Intelligence & Power Corporation (AIP) for the facility. Delivery of these units begins in 2026 with the goal of 2 GW coming online by 2027. These engines will work in concert with a battery energy storage system (BESS) that is designed to handle the extreme load swings AI data centers will experience.
But that’s just the first order. Up to an additional 6 GW could be added to the 2,380-acre site near Point Pleasant, West Virginia, over time. These gas engines can ramp from zero to full load in approximately seven seconds. They will operate on natural gas and incorporate emissions controls, including selective catalytic reduction (SCR), to support ultra-low emissions performance and compliance with all relevant air permitting requirements.
Perhaps the biggest gas engine on the market right now comes from Wärtsilä Energy. Its 18V50SG offers more than 18 MW. One data center in Ohio plans to group them together to provide almost 300 MW to the facility.
“Data center developers are facing increasing time-to-power challenges due to grid connection delays,” said Risto Paldanius, vice president of Americas at Wärtsilä Energy in a press release. “Quick access to power combined with high reliability, fuel efficiency and exceptional heat resilience make engines the ideal power supply for data centers.”
Even smaller engines are also getting in on the data center boom. Central Power Systems and Services received an order for 32 of its 420 kW genset packages for a data center in Kentucky. They will run in parallel. This enables one or more to be taken offline for maintenance or repair without seriously impacting total data center capacity.
Similarly, Volvo Penta’s G17 gas engine is targeting the data center market. This 17-liter, 6-cylinder, spark-ignited 450 kW model operates on both conventional natural gas and renewable natural gas.
Lead times reign supreme
The scramble to find power for AI data centers is on. Hyperscalers and data center developers have placed orders for gas turbines and nuclear power, but they aren’t waiting. Many are ordering gas engines with lead times ranging from a month to a year or more.
“Time to power wins,” said Laura Laltrello, COO of data center developer Applied Digital during the recent Data Center World Power conference. “The AI race favors builders who can energize fast — compute demand isn’t going to slow down.”
Also read: In February, high-temperature superconductors emerged as a potential way to move more power in less space for AI-heavy facilities.
