Many utility companies are placing their short-term bets on “demand response” solutions that require companies to reduce activity during peak times.
AI model builders typically operate data centers at full capacity during “training runs” — where they provide LLMs with massive amounts of data to enhance accuracy. This increased activity can clash with consumption from other customers — including households — during peak usage, raising the risk of blackouts.
Companies like OpenAI have also urged US regulators to expedite interconnection requests for flexible data centers, arguing that it will help “lower costs” for all users.
“We need to be more strategic in utilizing unused capacity on the grid,” said Daniel Eggers, executive vice president at Constellation, a power company that serves 2 million US homes and businesses.
Researchers at Duke University stated earlier this year that if data center operators could limit their consumption by 0.25 percent of the time, the grid could accommodate approximately 76GW of additional demand. They cautioned that this would not eliminate the necessity to construct new capacity.
Brandon Oyer, head of energy and water for the Americas at Amazon Web Services, mentioned that the company could handle some curtailment temporarily, but did not view it as a “wise investment” for an extended period. “Some customers may be able to endure that. Some customers may not. It will be a very nuanced decision.”
An intense journey
The concern for hyperscalers is that this combination of measures may not suffice to power data centers that will be operational in the next few years.
In this scenario, numerous projects will become unfeasible because they cannot fulfill contractual obligations. Others will have to wait for upgrades to the electricity grid and the construction of new generation capacity to be finalized.
In a competition among global superpowers, AI could be hindered by outdated grid infrastructure and a failure to provide adequate capacity.
For some, the power shortage alleviates worries of excessive construction. However, for tech companies and the Trump administration, it could undermine billions of dollars in investments.
“We may not achieve all of this within the timeframe that hyperscalers desire . . . and they will not be able to connect until we have the resources to accommodate them,” stated Nerc’s Robb. “It’s going to be an intense journey.”