AI datacenter boom triples US gas power builds, filling the air with more CO2
Summary
New analysis from Global Energy Monitor (GEM) shows US gas-fired power projects nearly tripled in 2025, driven largely by datacentre and AI demand. The US now leads the world in gas and oil power projects in development, with 251,737 MW planned on top of 561,907 MW already operating. More than a third of the new US projects are directly tied to datacentre builds.
If every project currently in development is built, GEM estimates about 53.2 billion tonnes of lifetime CO2 emissions would be added globally — 12.1 billion tonnes from US projects alone. GEM warns this rapid buildout risks locking grids into fossil fuels for decades, undermining climate targets and creating the potential for stranded assets if AI demand softens.
The piece highlights how policy and private-sector choices are contributing: the US administration has pushed to prioritise natural gas for expediency, and major tech firms have signalled willingness to use gas (sometimes paired with carbon-capture) to meet immediate datacentre power needs. GEM notes 2026 is on course to exceed the record natural gas deployment set in 2002.
Key Points
- GEM data: US gas/oil power projects in development almost tripled in 2025, totalling 251,737 MW.
- The US now accounts for about a quarter of the global gas power pipeline and leads the world ahead of China.
- Over a third of new US gas plants are linked directly to datacentre projects and AI energy demand.
- If all projects proceed, projected lifetime CO2 from new plants is ~53.2 billion tonnes globally, with 12.1 billion tonnes from US projects.
- GEM warns of a potential ‘gas lock-in’ that would tether grids to fossil fuels and hamper the renewable transition.
- Tech companies and government policy choices (including openness to gas + carbon capture) are accelerating builds despite climate pledges.
Context and Relevance
This story matters because it connects the rapid expansion of AI infrastructure to tangible increases in fossil-fuel power development. Datacentres are a major, growing demand sink for electricity; when renewable capacity and grid flexibility lag, developers turn to quick-build gas plants to guarantee power. That dynamic threatens emissions targets, raises the financial risk of stranded assets if AI growth falters, and signals policymakers must reconcile industrial policy with climate commitments.
For readers tracking energy transition, corporate sustainability and tech infrastructure, the article highlights a crucial tension: short-term scale-up of AI vs long-term decarbonisation. It also underscores where regulation, grid planning and corporate procurement strategies will need to shift to avoid locking in emissions.
Why should I read this?
Because it’s a clear, punchy wake-up call: the AI gold rush is helping gas plants boom, and that could wreck climate plans unless someone steps in. If you care about where datacentres get their power, the financial risks of rushed builds, or how policy and tech choices shape emissions, this summary saves you time — it’s the headline facts without the fluff.