Multistage Power Grid Capacity Expansion Planning under Uncertain AI Data Center Demands

Rapid growth of AI data centers creates large and difficult-to-forecast electricity demand. We formulate a multistage adaptive robust model in which utilities revise generation-capacity investment as demand is revealed, while generation and data-center flexible operation provide operational recourse. A tractable affine-decision-rule approximation is solved by constraint generation. Experiments show that multistage planning lowers total cost by about 3.5%, reduces capacity overbuilding, and decreases the largest absolute stage-wise expansion relative to a two-stage robust benchmark.