Gas turbine supply constraints emerge as primary barrier to AI data center expansion

The rapid expansion of artificial intelligence infrastructure is increasingly constrained by a severe shortage of heavy-duty gas turbines, with major manufacturers reporting production schedules that extend into the early 2030s. GE Vernova has confirmed that orders placed today will not be delivered until 2031, a timeline that stands in stark contrast to the aggressive power demand projections for data centers. This supply bottleneck is becoming a defining factor in the energy sector, overshadowing the numerous announcements regarding new AI facilities made over the past two years.

Demand for electricity in the United States is rising sharply due to data center growth. Goldman Sachs estimates that U.S. data center power demand will increase from 31 gigawatts in 2025 to 41 gigawatts in 2026, reaching 66 gigawatts by 2027. This trajectory implies that data centers will account for 8.5% of total U.S. peak summer demand by 2027, up from 4.1% currently. However, the manufacturing capacity for the gas turbines required to meet this demand is operating on a significantly slower timeline, creating a widening gap between projected consumption and available generation equipment.

Major turbine manufacturers are fully booked through the 2030s. GE Vernova reported a second-quarter backlog of 116 gigawatts, including slot reservation agreements, up from 83 gigawatts at the end of 2025. The company expects to have at least 125 gigawatts under contract by December. Siemens Energy ended its fiscal third quarter with a 69 gigawatt backlog, while Mitsubishi Heavy Industries reported a 35 gigawatt backlog for large-frame turbines. Although these figures are not directly comparable due to differing definitions of backlog and reservation agreements, the collective industry direction indicates record-high order volumes. Global orders in the second quarter reached a record 38 gigawatts, a 71% year-on-year increase, with the United States accounting for half of that total.

The disparity between demand and supply is stark. Goldman Sachs projects 36.3 gigawatts of new capacity additions in 2027 alone, a figure that exceeds the plausible annual output of the entire global turbine industry. GE Vernova’s annualized production is approximately 20 gigawatts, and only a fraction of its contracted gigawatts are earmarked for data centers. The remainder is allocated to utilities replacing coal, industrial loads, and grid reliability projects. Consequently, even if a manufacturer dedicated its entire global output to AI campuses, it would not cover one year of the projected demand curve, as most production slots were sold years ago.

The supply chain faces structural challenges beyond simple production volume. Specialized components such as hot-section castings rely on a limited number of foundries, and the skilled labor required for assembly has been depleted by previous industry downturns. Average lead times for new combined-cycle plants have increased from three and a half years in 2023 to roughly five years, with some heavy-duty frames requiring seven years. This scarcity has driven up prices, with BloombergNEF estimating the average cost of a combined-cycle project at $2,157 per kilowatt last year, up from under $1,500 in 2023. Wood Mackenzie expects turbine prices alone to reach $600 per kilowatt by the end of 2027.

Market mechanisms are already reflecting this shortage. PJM’s capacity auction for the 2028/2029 delivery year cleared at the regulatory cap of $325 per megawatt-day, yet the system remained 6,831 megawatts below its reliability requirement. PJM has proposed a one-time procurement targeting 14.9 gigawatts of new capacity to address the deficit. Meanwhile, manufacturers are adopting risk-mitigation strategies, such as slot reservation agreements, which function as paid options on future production. This allows companies like GE Vernova to collect down payments on turbines that may never be converted to firm orders, improving cash flow while managing inventory risk.

Counterarguments suggest that actual demand may be lower than projected. Exelon recently reduced its high-probability data center load estimate by nearly 40%, and Texas has ordered an audit of its interconnection queue after requests swelled to 474 gigawatts. BloombergNEF estimates that this pause puts nearly 50 gigawatts of potential capacity at risk of delay. Goldman Sachs itself assumes only 50% to 60% of scheduled data center capacity will be delivered on time. Regardless of whether the demand is real or speculative, the 2027 timeline remains constrained by the physical limits of turbine manufacturing, forcing reliance on alternative power sources such as reciprocating engines and fuel cells in the interim.