Power Play: AI Dominance Depends on Energy Resilience

Presented in the I Am The Cavalry track at BSides Las Vegas 2025, this talk argues that AI growth and the electric grid can no longer be treated separately. The speakers describe a cycle in which AI systems depend on power and cooling, while utilities increasingly lean on AI for their own operations. They contrast a grid being built around large data-center loads with one that already struggles to serve ordinary customers reliably.

The middle section explains why large loads matter. Grid operators keep generation and demand in balance to hold frequency steady, and AI training can cause sudden swings of roughly 15 megawatts at a single site. The speakers cited about 200 near-miss events in which data centers disconnected without warning, and a lightning-related fault in Northern Virginia that prompted dozens of facilities to switch to backup generation at once, abruptly removing a large block of load. Small rural cooperatives, often where these sites are being built, have limited cyber staff and planning models that do not yet account for data-center behavior.

The closing part turns to practical responses: demand transparency from data-center operators, consequence-based planning, chaos-style testing of assumptions, and black-start plans that include data centers. The speakers noted that restoration tools relying on data centers could push hospitals further down the queue. Their central point was that planning for failure and recovering quickly matters as much as prevention, with the Iberian Peninsula outage cited as a case where a prepared restart worked.