Gas turbine supply constraints and limitations in the US electricity grid could slow the expansion of power supplies for artificial intelligence data centres, according to analysts interviewed by OilPrice.com. The report identifies natural gas as the main prospective fuel for meeting that demand, while warning that equipment delivery times could hinder new generating capacity.
Mehrdad Emadi, head of risk analysis at London-based consultancy Betamatrix, told the outlet that electricity accounts for up to half of US AI firms’ costs. He said competition for generating equipment was creating bottlenecks in a turbine market dominated by GE Vernova, Siemens and Mitsubishi Heavy Industries. According to Emadi, order books are full, waits extend to seven years and are lengthening, while prices are heading towards a threefold increase.
The potential supply squeeze comes against forecasts of sharply higher demand. OilPrice.com cited Goldman Sachs Research projections that US data centre capacity requirements would rise from roughly 42 GW currently to 118-134 GW by 2030. Longer-term modelling from BloombergNEF puts the figure at about 194 GW by 2035. Meeting those requirements would entail a substantial expansion in reliable electricity supply.
Alternative sources face their own constraints, the report said. Wind and solar generation are intermittent, while nuclear projects carry lengthy development schedules and potentially substantial additional costs. Microsoft, Amazon and Google are considering investments worth tens of billions of dollars in nuclear projects, according to OilPrice.com. However, the outlet said conventional US nuclear developments have experienced 15-year timelines, while small modular reactor construction windows can reach seven years.
Steve Keen, honorary professor at University College London, told OilPrice.com that China benefits from a more centralised transmission system capable of moving electricity over greater distances. The report contrasted China’s ultra-high-voltage direct-current network with the US system’s three separate interconnections: Eastern, Western and Texas/ERCOT. Keen argued that US transmission constraints make it harder for American technology companies to expand AI operations at competitive costs.
Both analysts linked those energy constraints to risks for US AI company valuations. Their warnings centre on whether firms can secure enough steady power and generate sufficient revenue to support their costs, rather than on any announced change in fuel purchases or generating capacity.