‘Game changer’? What ‘DeepSeek’ AI means for electricity.

By Christa Marshall | 01/29/2025 06:53 AM EST

The Chinese company says its AI model is cheaper and uses less energy than those of its competitors, driving questions about predictions of enormous power demand.

The DeepSeek app is displayed on an iPhone screen.

The DeepSeek app is displayed on an iPhone screen. Justin Sullivan/Getty Images

The emergence of Chinese artificial intelligence company DeepSeek is challenging conclusions about future electricity demand because of data centers, a debate with implications for climate change and the future of fossil fuels.

The company caused a mass stock sell off Monday of U.S. utilities and computer chip company Nvidia after DeepSeek produced an AI model that appeared to compete with those from American firms and use a much smaller amount of energy at less cost. That prompted some analysts to say that surging predictions of electricity demand from AI may be overblown, or at least need a reset. Since AI is slated to drive the majority of electricity demand growth in the next decade, those predictions could affect how many power plants come online and how much they emit.

“It could be a game changer and reset expectations as to how the sector progresses from here,” said Jesse Jenkins, a Princeton University professor who helped advise Democratic lawmakers on crafting the Inflation Reduction Act, about DeepSeek.

Advertisement

Similarly, Morningstar analyst Travis Miller wrote in a note Monday that “data center growth will result in more electricity demand, but not as much as market valuations suggested.”

DeepSeek, which is owned by the Chinese stock trading firm High-Flyer, upended the tech world after releasing an app that rose to the top of the download charts of the Apple store. It appeared to have similar functionality as OpenAI’s ChatGPT chatbot, which can do things like write poetry when queried.

DeepSeek says its model uses roughly 10 to 40 times less energy than similar U.S. AI technology — a reduction that seemingly would sharply cut the need for energy-gobbling data centers. A Nature paper this month also reported that DeepSeek required about 11 times less computing resources than a similar one from Meta.

That indicates “it may be an order of magnitude more efficient,” said Jenkins.

Currently, Lawrence Berkeley National Laboratory predicts that AI-driven data centers could account for 12 percent of U.S. electricity demand by 2028. But projections vary, depending on assumptions made about the underlying technology being used.

Ayse Coskun, a computer expert at Boston University, said she expected DeepSeek’s open source data and energy-saving predictions to be validated. The company claims it uses about an eighth of the computer chips needed to power other AI systems.

Wall Street continues to see DeepSeek as a threat to U.S. companies building out AI, as stocks for Nvidia — which supplies computer chips fueling the AI boom — and Vistra — which is looking to support gas-fired data centers — remained down Tuesday from their earlier highs before Monday’s sell-off.

Yet for DeepSeek to cause a significant change in future electricity demand, there would have to be mass switching away from existing AI models, including by major corporations, said Betsy Soehren Jones, managing director at West Monroe, a consulting firm that supports electric, gas and water utilities.

She said she was not convinced large companies, which are some of the biggest drivers of AI demand, would be willing to tie their private data to a Chinese company.

“Look at what happened to TikTok,” she said. “I don’t know of a single company that is going to put their proprietary information into a Chinese-owned company that hasn’t shown us what they are doing with a that information.”

Overall electricity demand is still going to surge because other major drivers — particularly U.S. manufacturing and cryptocurrency — are not going to be affected by DeepSeek, she said. Policies such as President Donald Trump’s planned tariffs are likely to boost U.S. manufacturing and keep power demand high regardless of what happens with AI, she said. The same is true with an ongoing push for more electrification of appliances and use of electric vehicles, according to Jones.

Another factor is a paradox that when technology becomes more efficient and cheaper, people tend to want more of it, keeping overall demand stable. Coskun pointed to computer chips — which became more plentiful and thus used more energy overall — when they could make more computations per minute.

“Efficiency will come, but whether this is going to drop substantially the demand for AI energy, is very questionable,” Coskun said. The amount of AI power demand in the next few years “will still be a high number” even if it shifts slightly, she said.

She noted that while DeepSeek’s computer system appears to use less energy than other models, it still uses similar amounts of energy as competitors when the chatbot is queried. That suggests that the energy reductions could be minimized, she said.

Eric Gimon, a senior fellow at the clean energy think tank Energy Innovation, said uncertainty about future electricity demand suggests public utility commissions need to be asking many more questions about utilities’ potential projects and should not assume that demand they are planning for will be there.

There have been similar “land rushes” in the technology world before, where people overestimated how much infrastructure was needed, Gimon said.

“I’m concerned about utility planning,” he said.

A hit to Trump’s Stargate?

A lingering question is how DeepSeek might affect Stargate, a company involving several U.S. technology giants that was touted by Trump at the White House earlier this month. Stargate plans to invest roughly $500 billion in AI in data centers across the country over four years. Trump is looking to the project as a route to build more fossil fuel sources, vowing to do everything in his power to help bring company projects online.

White House press secretary Karoline Leavitt said at a press briefing Tuesday that the president believes that DeepSeek is a “wake-up call” to the U.S. AI industry. “President Trump believes in restoring AI dominance,” she said, referring to executive orders from the president last week undoing former President Joe Biden’s plans for AI. The White House is looking into the national security implications of DeepSeek, she said.

According to a research note from Morgan Stanley on Monday, the market reaction to DeepSeek was “overdone,” and there will continue to be a large number of U.S. data center announcements outside of DeepSeek tied to nuclear, gas and renewables. Among those that have been announced are a deal between Microsoft and Constellation Energy to restart a unit at the Three Mile Island nuclear reactor in Pennsylvania.

Part of the reason for the industry optimism is that a large number of U.S. data center projects are already in the pipeline, the note said.

The Stargate announcement is indicative of a U.S. approach that is “highly capital and power intensive,” Morgan Stanley analysts wrote.

Nvidia, which saw its stock rebound 9 percent Tuesday after a record plunge Monday, called DeepSeek “an excellent AI advancement” in a statement, noting it uses “significant numbers” of the company’s chips.

Ultimately, the effect of AI and data centers on emissions could hinge more on where projects are built, as the fuel mixes firing regional grids vary widely. Gimon said he thought a more competitive AI playing field could give a boost to clean energy projects in areas like West Texas, which has a lot of wind and solar.

“As there is more uncertainty in the field, [companies] are less willing to throw money around on expensive energy, … and the West Texas projects are some of the better deals,” he said.

This reporter can be reached on Signal at ChristaMars.31.