Eaton unveils edge-based monitoring tool to detect and manage AI workload power consumption spikes.

Power management company Eaton has introduced an edge-based solution aimed at identifying sudden spikes in energy demand from artificial intelligence computing, known as AI power bursts. Delivered as a firmware update to the company's Power Xpert quality (PXQ) event analysis system, the new tool allows operators to detect subsynchronous oscillations (SSO) that can disrupt both data centers and the power grid.

AI workloads powered by GPU servers are placing unprecedented pressure on energy infrastructure. These sudden fluctuations, or power bursts, can strain transformers, cause overheating, and damage equipment if not addressed. According to Eaton, the ability to detect SSO in real time provides operators with a preventive measure to reduce the risk of outages and extend infrastructure resilience.

The PXQ system, commonly deployed in switchgear, switchboards, and power distribution units, has been used to track events such as voltage sags, swells, and harmonics. With the new firmware upgrade, the system can identify low-frequency power signals and their spectral signatures, providing operators with early warnings of instability. As JP Buzzell, vice president and chief data center architect at Eaton, noted: "The energy demands of AI workloads surpass anything data centers and the grid have encountered before, with load fluctuations that can exceed the limits of existing infrastructure."

The development is part of Eaton's broader "grid-to-chip" strategy, which explores new methods of managing the rising energy requirements of AI data centers. This strategy includes collaboration with Nvidia on transitioning to 800 VDC power infrastructure to support megawatt-scale racks, and a partnership with Siemens Energy focused on integrating onsite power solutions into new data center designs.