How AI data centers are changing grid design

The growing energy demands of AI data centers are exposing a less visible issue than power supply: grid architecture. A sponsored MIT Technology Review article from ON.energy argues that recent outages in Virginia show how large AI campuses can create reliability problems when their loads change too quickly for conventional grid and data center designs. The article says traditional data center power systems were built for steadier, smallerscale loads, not AI facilities that can swing power use sharply or disconnect during disturbances. It argues that legacy UPS and protection systems can react in ways that worsen grid instability when many facilities behave similarly at the same time. To address this, the piece describes a different approach: moving power systems to medium voltage, placing equipment closer to substations, and using inline systems that smooth load changes continuously. It says a 2026 laboratory test found the design could handle both AI load swings and grid faults, and that it may simplify interconnection, improve space use, and support grid services such as demand response.