Three years ago, Meta, Microsoft, and Google all set ambitious climate targets. Now all three are building or planning natural gas power plants to feed their AI data centers.
The push is driven by a straightforward math problem: AI training and inference (the process of running models to answer your questions) require enormous amounts of electricity, and the grid can't supply it fast enough. Renewable energy projects take years to permit and connect. Gas plants can come online faster and deliver reliable baseload power around the clock.
How Much Power AI Actually Needs
AI workloads are power-hungry in ways that traditional computing never was. Training a single large language model can consume as much electricity as thousands of homes use in a year. Running inference across hundreds of millions of users multiplies that demand continuously.
US data center electricity consumption is projected to roughly double by 2030, with AI as the primary driver. In many regions, wait times for new grid connections stretch to several years. For companies racing to build AI infrastructure, that timeline is unacceptable.
The 30-Year Lock-In Problem
Here's where it gets uncomfortable: natural gas plants are expensive to build and typically operate for 25-30 years to recoup their investment. Build a gas plant in 2026, and you're committing to burning fossil fuels through the 2050s - the same decade these companies have pledged to hit net-zero.
The "bridge fuel" argument sounds reasonable in isolation. But bridges are supposed to lead somewhere, and none of these companies have shown a credible plan for decommissioning brand-new gas infrastructure that still has decades of useful life remaining.
Beyond CO2, natural gas extraction and transport involve methane leaks - and methane is roughly 80 times more potent than carbon dioxide as a greenhouse gas over a 20-year period. The full emissions picture is worse than the headline numbers suggest.
What This Means for AI Tool Costs
For daily AI tool users, this infrastructure buildout matters more than you might think. Electricity is one of the largest operating costs for AI services. Companies that lock in cheap, reliable power gain a structural cost advantage - one that eventually shows up in subscription prices and API rates.
It also puts pressure on sustainability marketing. Many AI companies position themselves as environmentally responsible. Building dedicated fossil fuel plants makes that claim harder to defend, especially as enterprise customers with their own climate commitments start scrutinizing the carbon footprint of their AI vendors.
The uncomfortable irony: the same companies building AI tools to optimize energy use in other industries are dramatically increasing their own energy consumption to do it.