Google, Nvidia and Emerald AI have launched the AI Energy Management Alliance, a trade group that will advocate for data centers able to reduce their electricity demand when power grids are under strain. The group brings together AI developers and energy companies to seek policies supporting flexible computing loads.
Energy policy veteran Frank Lacey will lead the organization, which evolves from the Advanced Energy Management Alliance founded in 2014. Members include Anthropic, NRG and Constellation Energy. Emerald AI founder and CEO Varun Sivaram said the group’s focus would be flexible AI data centers.
Google’s head of energy market innovation, Tyler Norris, said at the AI Infra Summit that the alliance was formed to submit comments in regulatory proceedings. Six federal proceedings opened in June address large-load grid connections and the allocation of transmission costs, with responses from the named grid operators due during the fall.
Google says its existing utility contracts include one gigawatt of demand flexibility. In remarks to reporters, Norris said studies indicate that loads willing to reduce consumption for fewer than 100 peak hours a year could allow dozens of gigawatts of additional demand on the existing US power system. Those are estimates of potential capacity, rather than completed grid connections.
Nvidia separately described an operating example at its Eos AI facility in Santa Clara, California. Silicon Valley Power sends demand signals to Emerald AI’s Conductor software, which slows or reschedules lower-priority computing while keeping higher-priority inference running. Nvidia says the facility has responded successfully to more than 200 signals.
In one August response described by the company, consumption fell from four megawatts to three without manual intervention. Conductor follows a predefined workload hierarchy and responds in under a minute, Nvidia says. The facility participates in Silicon Valley Power’s Flexible Load Interconnect Program.
The Santa Clara installation uses Conductor, with integration into Nvidia’s DSX Flex platform planned as that platform matures. Nvidia identifies a 96-megawatt Vera Rubin facility at its AI Factory Research Center in Manassas, Virginia, as the first dedicated commercial DSX Flex deployment. The operating results from Eos are company-reported.
The alliance’s formation follows large power commitments by its technology members. Google has tied a €13 billion Finnish infrastructure buildout to a 22-year nuclear contract. Utilities have also begun requiring bank guarantees for proposed data center grid connections.
Norris said the alliance’s comments would address proceedings that the industry’s existing association does not serve. The six grid operators’ fall filings are the next scheduled step in the federal connection and transmission-cost proceedings.
Sources: Heatmap, Nvidia, Broadband Breakfast
–
By the Control Plane Editorial Team