Skip to content

Product1 publisher3 min readPublished

Google and Nvidia ask grid operators to fast-track data centres that can power down

Google, Nvidia and 18 launch partners want grid operators to give faster connections to data centres that can cut power on request. Emerald AI's chief executive says the flexibility promise has to be enforceable first.

The Product Desk · Product desk

Photograph accompanying Google and Nvidia ask grid operators to fast-track data centres that can power down
Photo: thenextweb.com

What happened

  • Google, Nvidia and the startup Emerald AI launched the AI Energy Management Alliance on 16 September, with 18 launch partners including Anthropic, National Grid, Constellation, AES, RWE and NRG.
  • AEMA says power, not capital or chips, is now the main limit on building data centres in the US, and that today's connection rules suit facilities with flat, steady demand.
  • The alliance wants faster connection routes for credible flexibility commitments, and it plans to press the case with federal and state regulators, regional grid operators and utilities.
  • Nvidia, Digital Realty and Emerald AI plan to switch on a power-flexible AI factory in Virginia later this year that will draw nearly 100MW.
  • Polling from AP-NORC and the University of Chicago's Energy Policy Institute found 84% of Americans worry about data centres' impact on local electricity prices.

Compiled by The Product DeskSomething wrong?How this is made

Why it matters

  • constraint Agreeing curtailment and emergency response rules before a facility connects makes the shed auditable, so a missed call becomes a compliance matter for the site operator instead of a phone conversation.
  • decision Anyone sitting in an interconnection queue now has to price queue position against the cheapest flexibility they can actually deliver, and the three routes AEMA names do not cost the same.
  • exposure AEMA's own principle asks that connection costs follow a facility's real impact on the grid. Adopt that and the upgrade bill lands on developers, which is the part of the package that already polls with 79% of Democrats and 76% of Republicans.
  • precedent If regulators adopt performance-based standards, the metrics chosen become the ones every operator is measured against, and the numbers available to set them so far come from six demonstrations.

Data centres should get faster connections only if their ability to cut demand is "verifiable and enforceable", Emerald AI chief executive Varun Sivaram said on a call with reporters [15]. His pitch for the alliance is that flexible facilities can connect faster and at greater scale, and that host communities would get a more reliable grid and protection from rising bills [27]. "The most effective way to accelerate American AI is to make every data center a good citizen of the grid," Sivaram said in the launch announcement [12].

Google supplied the only hyperscaler number in the launch. It has committed 1GW of power demand it can reduce when needed, through utility agreements across the US, a figure Tyler Norris, its head of advanced energy market innovation, gave and Axios reported [14]. AEMA's website says flexible AI could unlock an extra 100GW from the existing grid [17]. The Google commitment is 1% of that [1].

AEMA's website puts the backlog for connecting new AI facilities at 5 to 10 years [18]. Sivaram wrote in a Fortune commentary that the US grid runs at only 50% of its capacity on average, citing a Brattle Group estimate that each 10% gain in utilisation lowers electricity rates by about 3.4% [19]. Emerald AI and Nvidia have completed six demonstrations of flexible data centres around the world, he wrote [20]. Emerald AI raised $150m in August [22].

Josh Parker, Nvidia's head of sustainability, said the company wants "technology-neutral, performance-based standards" that would judge flexible data centres on measurable grid reliability metrics [13]. AEMA says it is technology-neutral and wants facilities judged on what they deliver, such as how fast they can cut demand, Nvidia said in a blog post [10].

AEMA is a relaunch of the Advanced Energy Management Alliance, founded in 2014 to work on demand flexibility and electricity policy, which Sivaram said had become largely inactive, according to Axios [23]. Frank Lacey, an energy industry executive, is its executive director [3]. GridUnity, which makes grid connection and planning software, will sit on the board, and its chief executive Brian Fitzsimons said the company wants "to make the coalition's ambitions actionable" [24].

For anyone holding a queue position, the split worth making is between the flexibility in the filing and the flexibility deliverable at the hour the utility calls. AEMA names three routes: shifting or pausing computing work, using batteries, or running generation on site [6]. Batteries and on-site generation cost capital up front and let the racks keep computing. Pausing work costs the job itself, and a commitment built on deferrable training runs is a different document from one that reaches inference with customers waiting. The queue jump pays only if the route usable on the worst day is one the budget can carry, and if the person signing the enforcement clause knows they signed it. "Power-flexible data centers are a wonky topic," Sivaram said [16].

What to watch

  • Whether any state regulator or regional grid operator writes a flexibility fast lane into interconnection rules, and what verification it requires.
  • Whether the Virginia AI factory with Digital Realty and Emerald AI energises this year at the stated size.
  • Whether a second hyperscaler publishes a reducible-demand figure that can be compared with Google's 1GW.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories