Nvidia and Google's new energy alliance offers curtailment in exchange for faster grid connections

Google's contribution to the pitch is 1 gigawatt of demand it can cut through utility agreements, according to Tyler Norris, its head of advanced energy market innovation and the alliance's inaugural board chair ^[7][20]. Varun Sivaram, the chief executive of Emerald AI, says modest flexibility during the grid's worst hours could free up 100 gigawatts on a system that runs only about 50% utilized on average ^[8]. The one contracted commitment named at launch is 1% of the headroom the coalition wants regulators to price ^[14].

The demonstrated base is thinner. Emerald AI and Nvidia say they have run six flexibility demonstrations globally, and the facility that Nvidia, Digital Realty and Emerald AI plan to switch on in Virginia later this year is nearly 100 megawatts ^[9][10]. That is a tenth of Google's pledge and a thousandth of the 100 gigawatt figure ^[15].

The coalition reused an existing vehicle. The Advanced Energy Management Alliance, set up in 2014 to lobby for demand-response firms, had gone dormant, Sivaram told reporters ^[21]. (Cryptopolitan counted 20 companies and organizations at the open, including Anthropic, National Grid, AES, Constellation, NRG and RWE; Emerald AI said 18 launch partners alongside the three founders, which comes to 21 ^[23][24][25].)

Curtailment can leave the chips running. The flexibility can come from pausing or shifting AI computing jobs, drawing on batteries, or leaning on nearby generation ^[2]. The alliance says it is technology-neutral and performance-based, judging the measurable service a facility delivers instead of the hardware or software behind it ^[3]. A developer with enough storage or paired generation can meet a curtailment requirement without stopping a training run. That turns the obligation into capital spending, and capital spending is what the developer would otherwise hand the utility for the upgrades that flexible facilities can in some cases avoid ^[6].

The item on AEMA's agenda with money in it is the last one. It allocates interconnection costs to reflect actual system impacts and benefits, alongside faster, risk-adjusted pathways for customers that make credible and verifiable flexibility commitments ^[5]. How much faster those pathways run is not specified. "It's really a win-win-win solution," Nvidia's head of sustainability, Josh Parker, said before the launch ^[12]. Emerald AI said in a LinkedIn post: "What we agree on is specific: The grid was built for flat, predictable load, AI is not that, and the answer is not to ask the grid to simply absorb us" ^[13].

That pitch lands in an unfriendly room. AP-NORC and the University of Chicago Energy Policy Institute found 84% of Americans worried about data centers' effect on local electricity prices. It also found 53% extremely or very concerned about AI's environmental impacts, up 12 points in a year ^[16][22]. Ben Green, a University of Michigan researcher, said communities have organized well enough to block or pause many projects and called the local economic benefits a "significant false promise" ^[18]. Greg Abbott paused new Texas data centers in early August pending an audit of everyone queuing to connect ^[19].

FERC's June instruction to regional grid operators to weigh new ways of connecting large, flexible customers gives the coalition a docket, and a standard flexible-interconnection service follows ^[11]. Or utilities take the curtailment and state commissions still bill developers for the upgrades, which is where 79% of Democrats and 76% of Republicans sit ^[17]. Or developers buy batteries, satisfy the metric, and the grid gets a customer that flexes on paper. I expect the fight to land on cost allocation, not on the performance metrics. Sivaram himself set verification as the qualifying condition, and the members writing the metrics are the ones who will be measured ^[4]. If a commission grants the Virginia site both a faster connection and a reduced upgrade charge, that is wrong.