US data center developers, facing grid access wait times of up to seven years, are turning to aeroderivative turbines and diesel generators to power the AI boom
Two industries I follow closely converged in an unexpected way this week: aviation and data centres. … Adil Mohammad : ⚡ What does it actually take to power an AI revolution? ⚡ — While the world focuses on LLMs and neural networks … Bluesky: Dr Heidy Khlaaf / @heidykhlaaf : So not only are energy industries deregulating for AI, risking catastrophic accidents, but they're accelerating the use of the most polluting energy sources during a climate collapse. This is a death spiral. — www.ft.com/content/8deb... Kimmes / @real-me : “The incentives have never been greater for any sort of technology that can supply power,” said Kasparas Spokas, director of the Clean Air Task Force's electricity programme. [embedded post] @climatecentreuk : AI Data Centres Pay Double for Quick Power — To beat years-long grid waits, AI data centres use aircraft engines and generators. BNP Paribas calculates on-site power costs nearly twice normal electricity, raising questions about how long this inefficient workaround can last. — www.ft.com/content/8deb... John Kostyack / @kostyack : Using fossil fuels to power data centers is bad for your lungs & wallets, especially if the fossil fuels are burned behind-the-meter with aeroderivative turbines (jet engines) & diesel generators. But developers can't wait for grid connections & modern turbines, so brace yourselves.
Context & Ripple Effects
The buildout of AI facilities has already been colliding with electricity-system constraints: earlier coverage projected sharply rising data-center consumption, while reporting also found AI demand encouraging fossil-fuel investment where renewable output does not match computing loads. This report shows grid-connection timing becoming an operational constraint rather than a back-office permitting issue.
The move also deepens a conflict visible in coverage of data centers’ strained sustainability commitments and their search for experimental clean-power options: developers need dependable capacity now, while cleaner long-duration alternatives are not yet broadly available at the required pace.
First-order effects
- Developers with long interconnection waits can bring AI facilities online sooner by installing aeroderivative turbines and diesel generation on site, but accept substantially higher power costs than grid supply.
- The immediate power mix for affected facilities becomes more fossil-intensive, increasing local pollution, safety, and climate exposure flagged by experts.
Second-order effects
- Grid access becomes a sharper competitive differentiator: developers able to secure connections or finance interim generation can advance projects while rivals remain delayed.
- Demand shifts toward distributed-generation equipment and fuel supply, reinforcing the fossil investment response to AI energy demand even as developers continue to seek cleaner alternatives.
Third-order effects
- If grid queues remain lengthy, AI-compute deployment will increasingly be shaped by power-delivery timelines and on-site generation economics, not solely by chips, land, or data-center construction.
- This could pressure utilities, regulators, and developers to reconcile faster interconnection and reliability needs with pollution and decarbonization goals; the eventual balance remains uncertain.
The trend: AI infrastructure is becoming utility-constrained infrastructure, with developers using costly interim power to bridge the gap between compute demand and grid capacity.