AI Data Center Power Demand Forces Utilities to Rethink Multi-Decade Grid Planning Timelines

US electric utilities are compressing grid capacity planning timelines that traditionally operated on decade-plus horizons, as power capacity requests tied to AI data center construction from hyperscalers — Microsoft, Google, Amazon, and Meta chief among them — have outpaced every prior demand forecasting model utilities used to plan transmission and generation investment. Several regional utilities have publicly acknowledged that current AI-driven demand growth projections, if realized, would require grid capacity additions at a pace not seen since the post-World War II industrial buildout. The practical bottleneck utilities describe most consistently isn't generation capacity itself but transmission infrastructure and interconnection queue processing time — building a new power plant or securing a long-term power purchase agreement is often faster than getting the transmission lines and grid interconnection approvals needed to actually deliver that power to a new data center site, a mismatch that has left some announced hyperscaler data center projects facing multi-year delays waiting on grid connection rather than on data center construction itself. The financing and cost-allocation question dividing utilities, regulators, and hyperscalers is who bears the cost of the accelerated grid buildout: hyperscalers have increasingly agreed to direct infrastructure cost-sharing arrangements and, in some cases, financing dedicated power generation — including nuclear power purchase agreements with plant operators like Constellation Energy and Talen Energy — rather than relying entirely on traditional utility rate-base financing that would spread the cost across all ratepayers, a shift regulators in several states have encouraged specifically to avoid AI infrastructure costs landing on residential electricity bills. The unresolved long-term risk utilities are quietly planning around is demand volatility: AI data center power demand forecasts assume continued AI infrastructure investment growth at something close to the current pace, and utilities that commit to multi-decade generation and transmission investment based on current hyperscaler demand signals are exposed if AI infrastructure spending growth slows meaningfully before that capacity investment is paid off, a scenario several utility risk assessments now explicitly model as a planning scenario rather than dismissing as unlikely.