Google and Nvidia join an alliance that would fix curtailment terms before a data center connects

"Flexible" has a specific definition inside this coalition. Nvidia's launch post lists the ways a data center can adjust what it draws from the grid: shifting computing workloads, discharging storage, using paired generation, or responding to system contingencies ^[10]. The alliance says it is technology-neutral and performance-based, judging a facility on response speed, duration, predictability and behavior during an emergency rather than on the hardware or software it runs ^[11].

Three of the four principles Nvidia published concern the interconnection process itself ^[16]. Ride-through, curtailment and contingency-response obligations would be defined before a facility connects ^[12]. Customers that make credible and verifiable flexibility commitments would get faster, risk-adjusted pathways ^[14], and interconnection costs would be allocated to reflect actual system impacts and benefits such as avoided upgrades ^[15]. The fourth covers standardized technical requirements, performance metrics and operational data sharing ^[13]. Interconnection was designed around facilities with flat, static electricity demand ^[9], and the alliance is asking for a second queue with conditions attached.

The savings come from peaks. Much of the cost of building and maintaining the grid, including gas-fired plants, is incurred to meet peak hours ^[24]. "It's when you get up to the hottest or coldest day of the year that you need that extra transmission line or need to build a new one," Abraham Silverman told Heatmap News ^[22]. Silverman, a former general counsel at the New Jersey Board of Public Utilities and an assistant research scholar with the Ralph O'Connor Sustainable Energy Institute at Johns Hopkins University, is not involved with the alliance ^[21]. "The savings to consumers if we had effective flexibility is enormous," he said ^[20].

Varun Sivaram, founder and chief executive of Emerald AI, told reporters in a briefing: "There are a lot of AI trade associations, data center trade associations, energy trade associations. This is the only one that is laser focused on flexible AI data centers" ^[6]. The academic case for that focus was published last year by Tyler Norris, then a researcher at Duke University's Nicholas School of the Environment, who argued that much of AI data center electricity demand could be served by the existing grid with modest flexibility ^[25]. Norris is now Google's head of energy market innovation ^[25].

A trade group's principles bind nobody. The terms that matter appear in tariffs and interconnection agreements written by state commissions and grid operators, and those proceedings are already running. "Just about every state has some either legislative or regulatory process for looking at this," Silverman said ^[23], and Heatmap reports that state and local regulators have shown openness to a variety of approaches that could get data centers on the grid faster while minimizing grid impact ^[27]. Nvidia wrote that "The rules governing power for AI are being written now" ^[26].

Neither publication compares power against chip supply or capital, so the record does not settle which constraint binds hardest. But the largest buyers are spending their advocacy on the connection terms.