A shortage of gas turbines has emerged as a new bottleneck in AI development. Demand to secure power for data centres has surged, but the supply of gas turbines for natural gas generation is failing to keep pace, raising the likelihood that power procurement will be delayed, online media outlet Gigazine reported on Aug. 24 (local time).

In AI infrastructure, power supply systems for data centres have been cited as a bottleneck along with semiconductors, memory and storage. Recently, some AI companies have been building natural gas power generation facilities on their own sites to cut long-term power costs. BloombergNEF estimated that about 100 data centres have already built such facilities or are under construction. PwC forecast natural gas consumption in the AI industry will rise more than fivefold by 2035.

Gas turbine manufacturers are already signalling tight supply. GE Vernova said at an earnings briefing on July 22, 2026 that even if large gas turbines are ordered now, they would not be delivered until 2031. It also said more than half of the production capacity reserved for 2031 output would be contracted by the end of 2026.

Siemens Energy said delivery takes more than three years. Mitsubishi Heavy Industries said projects it won from March to June 2026 are scheduled for delivery from 2028 to 2030, adding it is selecting projects to contract. The overall price of power generation equipment also doubled between 2023 and 2025.

The pace of growth in power demand is steeper. Goldman Sachs estimated U.S. data centre power demand will rise to 41 GW in 2026 and 66 GW in 2027 from 31 GW in 2025. On the supply side, it forecast capacity additions of 8.5 GW in 2025, 13.6 GW in 2026 and 36.3 GW in 2027.

Still, it is uncertain whether the outlook for expanding supply will materialise. GE Vernova's annual gas turbine production is about 20 GW, and only one-fifth of that is allocated for data centres.