The Cheapest Power Plant for AI Might Be the One Nobody Builds

Updated: 10 hours ago
Part 3 of 3 in the series Who Pays for the AI Grid.
PJM's scarcest resource isn't energy. It's capacity at 5 p.m. on a hot August day.
Google just expanded its Voltus virtual power plant in PJM tenfold, to roughly 1 GW. Homes and businesses get paid to trim usage at peak, with commitments running through 2040.
Barron's calls utilities the lower-risk way to own AI demand. Flexibility makes that case stronger:
Speed to revenue: new load connects years sooner than waiting on new peakers
Affordability: serving more load on the same wires helps hold down rates, a big win in rate cases
Less stranded asset risk: fewer plants built for a handful of peak hours a year
Better capex mix: dollars shift toward grid, storage and utility-run programs like Xcel's Capacity*Connect
Strategic finance takeaway: Rate base growth matters. The quality of that growth matters more. Flexibility trades some peaker capex for lower risk and stronger regulatory goodwill.
My call: 60% odds another hyperscaler signs a VPP capacity deal of 100 MW or more by mid 2027.
Utility leaders: is flexibility a threat to rate base, or the smartest capacity you'll never have to build?
Background: Canary Media
Originally posted on LinkedIn on October 25, 2026. Join the discussion on LinkedIn.


Comments