Jigar Shah and Arnab Pal in Utility Dive:
Across the country, a particular brand of conventional wisdom has set in: The only way to power the AI boom is a massive buildout of gas — both on the grid and “behind the meter” at data center campuses. The assumption is that anything else will be too slow, too risky or too complicated.
But “gas first” is simply a reflex, and it comes with two problems that are colliding with politics. First, it’s expensive. New generation and the upgrades to support it cost real money, and the utility model often pushes those costs onto ratepayers. Second, it’s slow, particularly where the grid is constrained. In many regions, equipment timelines are now measured in years rather than months.
If states respond with peak-driven planning — building the grid to serve the hottest hour of the year plus a reserve margin — customers will pay for infrastructure that sits underutilized most of the time. In our analysis, the grid is only 50% used throughout the year. There’s a better way: Build smarter before you build bigger.
Continue reading “Business as Usual Will Not Solve AI Energy Woes”






