A Silicon Valley AI data center automatically reduced its electricity draw from four megawatts to three in response to a signal from local utility Silicon Valley Power. Higher-priority computing continued while flexible jobs were slowed or rescheduled.
Emerald AI’s Conductor software coordinated the adjustment across thousands of Nvidia GPUs without an operator intervening. Nvidia describes it as an early example of the flexibility that its DSX Flex infrastructure is intended to support. The utility has since sent more than 200 demand signals to the facility, and Emerald says the system responded successfully each time.
The approach treats some AI computing as schedulable industrial demand. Training, batch inference or other work that can wait may be shifted away from moments when air-conditioning and other loads strain the grid, while time-sensitive services retain power.
This does not eliminate the electricity required by the data center or replace new generation and transmission. It changes when part of that demand reaches the grid. The reported deployment is one facility and comes from the participating companies, but it demonstrates that a large GPU cluster can provide demand-response capacity without shutting down all customer work.