AI data centres need more than chips: access to electricity, gas and the equipment to deliver power is becoming a constraint on the build-out. Yahoo Finance reports that Anthropic’s IPO filing warns it may struggle to secure firm natural-gas supply and pipeline capacity for some facilities.
Watch Desk analysis
What happened
Anthropic’s filing says rising reliance on gas-fired generation by data-centre operators could intensify competition for fuel and pipeline capacity, and that shortages could delay or derail planned sites. Yahoo Finance also cites Goldman Sachs research estimating US data-centre electricity demand will double between 2025 and 2027, after roughly 20 years of largely stagnant grid expansion.
The pressure is showing up in equipment orders, too. GE Vernova said its power segment’s orders grew 178% year on year, driven primarily by gas-power equipment such as turbines. The company reported that its gas-power backlog and slot reservations increased from 116 GW to 122 GW, and forecast 130 GW by year-end.
Why it matters
AI infrastructure is often discussed in chips and compute budgets. These figures point to another practical limit: getting dependable power to the places where the computers are meant to run. Fuel, pipelines, turbines and grid capacity do not appear on demand just because a model launch is on the calendar.
The figures are not a neat forecast of which projects will be delayed, or proof that gas is the only answer. They do show that energy supply and equipment orders are becoming part of the AI build-out story, with consequences well beyond the data-centre fence.
Our read
The useful takeaway is not “AI means gas”, a slogan that moves rather faster than the evidence. It is that AI’s appetite for electricity is colliding with infrastructure that takes time to build. If you are tracking the pace and cost of AI expansion, power supply belongs on the dashboard beside chips and capital spending.
What to watch
- Whether Anthropic or other data-centre operators disclose project delays tied to fuel or power access.
- How quickly turbine and grid-equipment backlogs turn into installed capacity.
- Whether data-centre electricity demand grows as Goldman Sachs estimates.
Discussion spark: Should new data centres have to secure and disclose a credible power plan before they are approved, or would that risk slowing useful computing investment?
Sources and evidence
- Natural gas emerges as key AI bottleneck, Anthropic S-1 filing shows – Yahoo Finance (28 October 2026, 14:11 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.