Using Grid-Forming Energy Storage Systems to Provide Dynamic Active Power Support for Hyperscale Data Center

Large artificial intelligence (AI) training data centers are emerging as significant, highly dynamic digital loads that pose numerous challenges to power system stability. Methods to study and assess their impact are still being developed and have been the focus of numerous industry interest groups. Intended to contribute to bridging this gap, this paper develops a credible impact study for data centers, presenting a dynamic data‑center model that includes an uninterruptible power supply (UPS), co‑generator, cooling system, and a grid-forming (GFM) controlled energy storage system (ESS), with explicit converter and control dynamics. Using electromagnetic transient simulations, two stress scenarios are studied: low‑voltage faults and fast load fluctuations characteristic of AI training. The results show that an adequately tuned and controlled GFM ESS can significantly improve the large load response by reducing frequency overshoot, raising the nadir, shortening settling time, and enhancing post‑fault recovery. Under fast fluctuation AI training load profiles, the ESS also damps active‑power oscillations and reduces torsional shaft‑torque oscillations in the generator train to safe levels. These findings support ESS‑based control as an effective tool for integrating large AI data centers without compromising grid stability.