Georgia Power is asking regulators to approve a $15 billion plan to add 10,000 MW in five years, largely to feed AI-heavy data centers, with real risk that regular customers carry the cost if demand projections miss.
The proposal leans on new gas plants and other fossil assets, which would lock in emissions and long-lived infrastructure just as data centers and some vendors explore “bring your own generation” models like 90 MW of RICE engines dedicated to a single site.
Regulators and staff warn residential bills could rise $20+ per month under some scenarios, despite Georgia Power’s promise of “downward pressure” and only a non-binding structure targeting an $8.50 benefit years out.
In parallel, FERC is pushing PJM to formalize rules for colocation deals where hyperscalers hook data centers directly to plants (e.g., Amazon–Susquehanna), trying to avoid cost-shifting and grid reliability hits as capacity prices spike ~1,000 percent on data center demand.
This is turning electricity into the next AI bottleneck: utilities want to build ahead of load, while policymakers worry about stranded assets, higher bills, and local air impacts from on-site generation.
States are experimenting with protections like Texas’s ERCOT emergency cut-off rules for data centers, but there is no consistent model yet for aligning AI buildout speed with grid planning reality.
The article is worth a read for anyone mapping AI capacity plans to real-world power, regulatory, and cost constraints.
