Co-Located AI data centers, hydrogen production, and battery storage: A Multi-Timescale Grid-Connection assessment

Converter-dense campuses that combine large digital loads, hydrogen production, and battery storage at one point of interconnection (POI) require coordinated assessment across planning timescales. This paper presents a reproducible workflow that links steady-state and N-1 power flow, electromechanical frequency dynamics, frequency-dependent network impedance, harmonic screening, and electromagnetic-transient (EMT) fault recovery. The principal benchmark adds a 500 MW data center and a 500 MW electrolyzer load to the IEEE 39-bus system, with a 100 MW/400 MWh battery energy storage system (BESS) in the coordinated topology. The 1 GW load is a severe hosting-capacity stress case, not a proposed rating for this network; generality is tested at 500, 750, and 1000 MW, at three POIs, for three generator contingencies, and in a separate WECC 179-bus portability case. At the IEEE 39 reference POI, passive co-location reduces voltage to 0.936 p.u. Coordinated reactive support raises it to 0.947 p.u., but neither that case nor 29 additional local mitigation variants satisfies the benchmark-relative N-1 target. In WECC, where the same facility is 1.65% of demand, the corresponding voltages are 0.956 and 0.963 p.u. Post-contingency electrolyzer curtailment improves frequency nadir in both networks. The IEEE 39 prescribed-spectrum screening gives 6.84% and 7.10% voltage total harmonic distortion (THD-V) for passive and coordinated co-location; the WECC values are lower but remain above the 1.5% limit. ParaEMT provides a time-domain consistency check for harmonic propagation and exposes load-model sensitivity in low-voltage ride-through (LVRT) results. The output is therefore a planning synthesis with explicit evidence levels and validation gates, rather than universal connection requirements.