Many utilities are pinning their short-term hopes on “demand response” solutions that require companies to curtail operations peak times.
AI model builders typically run data centers at full capacity during “training runs” – where they provide LLMs with massive amounts of data to improve accuracy. These activity increases can collide with the consumption of other customers – including households – during peak consumption, increasing the risk of power outages.
Companies including OpenAI have also asked US regulators to speed up interconnection requests for flexible data centers, arguing this will help “reduce costs” for all users.
“We need to be smarter about using unused capacity on the grid,” said Daniel Eggers, executive vice president at Constellation, an energy company that powers 2 million American homes and businesses.
Researchers at Duke University said earlier this year that if data center operators could limit their consumption 0.25 percent of the time, the electricity grid could accommodate about 76 GW of additional demand. They warned that this would not replace the need to build new capacity.
Brandon Oyer, head of energy and water for the Americas at Amazon Web Services, said the company could tolerate some curtailment on a temporary basis but does not consider it a “smart investment” to do so for an extended period. “Some customers may be able to tolerate that. Some customers may not. It’s going to be a very nuanced decision.”
A white knuckle ride
The concern for hyperscalers is that this patchwork of measures will not be sufficient to power data centers that will come online in the near future few years.
In this scenario, a whole series of projects will no longer be viable because they cannot meet contractual obligations. Others will simply have to wait until grid upgrades and construction of new generation capacity are completed.
In a race between global superpowers, AI could be held back by decades-old network infrastructure and its inaction sufficient capacity.
For some, the power crisis eases concerns about overbuilding. For tech companies and the Trump administration, this could undermine billions of dollars in investments.
“We may not get all of this in the time frame that hyperscalers would like . . . and they won’t be able to connect with each other until we have the resources to meet them,” said Nerc’s Robb. “It’s going to be an exciting ride.”
