Gas Turbine Shortage: The New Bottleneck for AI Data Center Power
The global race to scale artificial intelligence is hitting a hard physical ceiling as gas turbine manufacturing capacity fails to meet the surging power demands of hyperscale data centers. With order books for heavy-duty turbines now extending to 2031, major utilities and tech firms face a multi-year supply chain bottleneck that threatens to derail projected AI infrastructure timelines through the end of the decade.
The Structural Mismatch in Power Generation
Data center power demand in the United States is projected to climb from 31 gigawatts (GW) in 2025 to 66 GW by 2027, according to Goldman Sachs estimates. This expansion requires a massive influx of baseload power that the current manufacturing sector cannot support. Global turbine manufacturing capacity sits at approximately 60 to 70 GW annually, yet total industry orders have already eclipsed 110 GW. This supply-demand imbalance creates a fiscal environment where capital costs for combined-cycle power plants have roughly doubled in recent years, according to data from Wood Mackenzie.
For hyperscalers attempting to bypass grid congestion with behind-the-meter generation, the situation is equally constrained. These private power projects rely on the same heavy-duty gas turbines as municipal utilities.
Backlog Realities at the Industry’s “Big Three”
The manufacturing crunch is quantified by the record-breaking backlogs held by the world’s three largest turbine producers. GE Vernova reported a gas power equipment backlog and slot reservation total of 116 GW as of the second quarter of 2026, an increase from 100 GW just three months prior. During the July 22 earnings call, GE Vernova leadership confirmed that the firm is already taking reservations for 2031 delivery.
Siemens Energy reported a 69 GW firm backlog as of June 30, with lead times now consistently exceeding three years. Meanwhile, Mitsubishi Heavy Industries recorded a 35 GW large-frame turbine backlog as of its fiscal first quarter, with CFO Hiroshi Nishio noting that the firm is actively becoming “selective in the projects we contract” to manage the delivery window between 2028 and 2030. The discrepancy in how these firms define “backlog”—mixing firm orders with paid slot reservations—suggests the true scarcity of available, near-term production slots may be even more acute than aggregate figures imply.
Grid Reliability and Capital Allocation
The inability to secure equipment is rippling through regional grid operators. The PJM Interconnection, which manages electricity for 13 U.S. states, reported that its July 2026 capacity auction fell 6,831 MW short of its reliability target. This marks the third consecutive year the organization has failed to meet its projected capacity needs.
The competitive landscape is further complicated by international demand. Gulf states, investing heavily in AI clusters and desalination infrastructure, are competing directly with U.S. tech firms for the same finite manufacturing pipeline.
The Path Forward for AI Capital Expenditure
The mismatch between digital growth projections and industrial manufacturing reality suggests a period of significant volatility for data center developers.