During a World Cup soccer match simulation, millions of Brits switching on electric kettles created a sudden electricity demand spike. Emerald AI's Conductor software responded by instructing a London data center to reduce power consumption from its chips, preventing potential blackouts. This marked a departure from traditional data centers that typically consume power without grid coordination.

Conductor enables data centers to scale back energy use during peak demand while prioritizing critical workloads. Nvidia and Digital Realty back the system as one of the first 'power-flexible AI factories.' The technology addresses a major bottleneck: PJM, America's largest grid operator, requires eight years to bring new generation online, far longer than data center construction timelines.

Research supports the flexible approach. A Duke University study found the US grid could provide 76 gigawatts of additional capacity—about 5% of total capacity—to data centers willing to reduce usage just 0.25% of the time. Meanwhile, $150 billion in data center projects stalled in 2025 due to community opposition over power consumption and environmental concerns.

Despite benefits, adoption faces resistance from utilities comfortable with rigid grid management and companies preferring dedicated power generation. The GRID Act proposes severing new data centers from public grids entirely. However, gas turbine supply cannot meet hyperscaler demand, making flexibility a practical necessity rather than optional optimization.