Product3 publishers2 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
What happened
- Google, Nvidia and AI orchestration firm Emerald AI founded the AI Energy Management Alliance to develop flexible load capability in AI data centers, joined by 18 launch partners including Anthropic and National Grid.
- Tyler Norris of Google wrote that multiple studies find cutting net grid withdrawal for under 100 hours a year can unlock dozens of gigawatts, against 5-7 years or more of waiting for system upgrades.
- Google's 1GW of demand response, first demonstrated with Omaha Public Power District in 2024, now sits inside long-term contracts with utilities including Entergy Arkansas, Minnesota Power and DTE Energy.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- decision A new campus now has two paths to compare, and the alliance exists to make the flexible one a signable arrangement instead of a bespoke negotiation each time.
- constraint Sites whose load is customer-facing inference cannot supply the flexibility by rescheduling work, so they have to buy it as storage or on-site generation and pay for it up front.
- cost Curtailment is paid in delayed work, and the bill lands on whoever owns the job that pauses, not on the utility that asked for the reduction.
- precedent With National Grid, NRG, Constellation, RWE and AES sitting inside the alliance alongside Anthropic, the contract language settled here is what gets offered to operators who had no seat at the table.
The person this lands on is whoever takes the utility's call and decides what stops running. Tyler Norris, Google's head of energy market innovation for AI and infrastructure and the alliance's inaugural board chair, wrote that the capacity can come "via compute flexibility, storage, or generation" [5][6]. Storage and generation are equipment purchases with a delivery date. Compute flexibility means a job that was running is not running, and somebody owns that job.
A hundred hours is 1.1 percent of an 8,760-hour year [14]. Applied to the 1GW Google says it has already integrated, 100 hours of full curtailment comes to 100GWh of consumption dropped or moved [15][9]. The alternative Norris names is 5-7+ years of waiting for traditional system upgrades [5].
The "dozens of gigawatts" belongs to "multiple studies" that Norris cites in the LinkedIn post without naming them [5]. The part with signatures behind it is Google's own: 1GW folded into long-term energy contracts with US utilities as of March [9], starting with a demonstration at Omaha Public Power District in 2024 [10], then agreements with Indiana Michigan Power and the Tennessee Valley Authority [11], and now Entergy Arkansas, Minnesota Power and DTE Energy [12], six counterparties named in all [13]. Google's 1GW is the only capacity figure the DatacenterDynamics account attaches to any member [18].
The two reports Google released alongside the launch point in different directions. The Brattle Group wrote a technical blueprint for turning recent bipartisan direction from the Federal Energy Regulatory Commission into operational reality [7]. Aurora Energy Research modelled flexible data centers in ERCOT paired with front-of-meter resources [8]. In the Aurora case, some of the flexibility sits out in those front-of-meter resources, beyond the scheduling decisions made inside the building.
For an operator, this is a sizing exercise. Pull last year's load curve, mark the hundred hours the utility would most plausibly have called, and look at what was running in them. If it was batch training you can checkpoint and restart, the flexibility costs delay and nothing else. If it was customer-facing inference under a service commitment, the flexibility has to be bought as batteries or on-site generation, and the comparison becomes that capex against 5-7+ years in the interconnection queue [5].
What to watch
- Whether AEMA publishes a member standard, or any curtailment commitment from Nvidia, Emerald AI or Anthropic.
- Whether FERC turns the bipartisan direction the Brattle report addresses into a flexible-load interconnection rule.
- Whether any of Google's six named counterparties files a public tariff that other operators can sign.