Recent power disruptions in Ashburn, Virginia, have underscored a growing vulnerability in the electrical grid caused by the rapid expansion of AI data centers. In July 2026, a transmission line fault resulted in a loss of over 3 gigawatts of load, following a similar incident two years prior that disconnected approximately 1,500 megawatts. Industry analysts suggest these failures stem from outdated power architecture rather than a lack of electricity supply.
Traditional data center power systems, which rely on low-voltage uninterruptible power supply (UPS) units located inside buildings, are struggling to manage the volatile energy demands of AI infrastructure. AI campuses often experience rapid load swings of up to 70% in milliseconds, which can trigger protective disconnects when the grid experiences minor disturbances. Because existing protection logic was designed for smaller, more predictable industrial loads, these facilities often shut down during grid fluctuations, exacerbating instability.
To address these challenges, experts are advocating for a transition to a medium-voltage, inline power architecture. This proposed model involves moving power conditioning equipment outside the data center into modular enclosures near substations. By routing power through a system that operates at medium voltage—typically 13.8 kilovolts or higher—data centers can absorb load swings before they impact the grid. This approach allows the facility to present a flat load profile to utility providers, potentially reducing permitting timelines and allowing backup systems to participate in grid-stabilization programs.
Testing conducted in early 2026 at the National Laboratory of the Rockies demonstrated that this medium-voltage configuration could maintain stability during both simulated AI load spikes and full zero-voltage grid faults. By shifting the architecture, operators may be able to transform data centers from a strain on the grid into a reliable asset, while simultaneously increasing construction density and improving the economic viability of backup power systems.
Source: MIT Technology Review
មិនទាន់មានមតិយោបល់នៅឡើយទេ។ សូមធ្វើជាអ្នកដំបូងដែលចែករំលែកគំនិតរបស់អ្នក។