arXiv:2611.23771v1 Announce Type: new Abstract: Gigawatt-scale artificial intelligence data centers (AIDCs) are being connected at a single point of interconnection (POI) faster than transmission can be reinforced, yet studies of their flexibility rely on DC power flow and simplified load envelopes that omit voltage, cooling dynamics, and workload deadlines. This paper addresses this gap by embedding a high-fidelity AIDC model, with separate inference and training pools, hall thermal mass, PV, a battery energy storage system (BESS) with reactive capability, and a STATCOM, in a five-minute multi-period AC optimal power flow. Simulations on the RTS-GMLC test system show that active-power flexibility mitigates local price spikes and relieves congestion, whereas reactive support from the BESS inverter or a STATCOM is needed to keep the POI voltage away from sagging around its 0.95~pu limit. Sensitivity analyses show that cost-driven charging of large batteries can induce new morning congestion and that storage largely substitutes for wider workload deferral.