The recent announcement from DeepSeek, a Chinese company, has sent shockwaves through the AI and energy sectors. By claiming that its AI model, DeepSeek-V3, requires less than $6 million in computing power from Nvidia H800 chips, the company has ignited a fierce debate about the future energy demands of artificial intelligence. This low-cost model has already outperformed ChatGPT, becoming the top-rated free application in the U.S. Apple app store. The implications of this development are monumental, particularly as it raises questions about the projected electricity demand tied to AI, which has been forecasted to account for approximately 75% of overall U.S. demand through 2030-2035.
The immediate fallout from DeepSeek’s revelation was a significant market rout for Nvidia, which lost a staggering $589 billion in market value on January 28. Investment bank Jefferies pointed out that the efficiency of DeepSeek-V3 could potentially lead to a reevaluation of the exponential power demand growth that many analysts have anticipated. If AI models can operate on significantly less computing power, the energy landscape may shift dramatically, forcing investors and energy companies to reconsider their strategies.
This uncertainty has particularly rattled U.S. utilities and energy firms involved in data centers. Shares of companies, including advanced nuclear technology firms like NuScale and Oklo, experienced declines, as the market digested the potential for reduced demand. Jefferies noted that a slowdown in data center projections could adversely affect premium utilities that investors expect to increase their rate base. However, the bank also cautioned against overreacting to the news, framing DeepSeek’s launch as part of an ongoing evolution rather than a revolutionary shift.
Amid this turbulence, major energy players are still betting on the data center space. Chevron, in partnership with investment firm Engine No. 1, is planning to build natural gas-fired power plants to directly supply data centers. This collaboration, which aims to develop 4 GW of gas-fired generation, underscores the belief that energy is crucial to maintaining America’s AI leadership. Mike Wirth, Chevron’s CEO, emphasized the need for scalable and reliable power solutions to support advancements in AI.
Additionally, other energy companies are making significant moves in the data center sector. Constellation’s acquisition of Calpine for its gas-fired power stations and NextEra Energy’s plans to expand gas-fueled plants highlight a growing trend. The past year has seen a flurry of announcements from technology firms securing power purchase agreements for AI initiatives, spanning various energy sources, including nuclear and renewables.
As the demand for energy grows, the discourse around virtual power plants (VPPs) has gained traction. Mathew Sachs of CPower noted that VPPs could play a pivotal role in addressing the load growth driven by AI and data centers. He characterized AI as a modern-day “Manhattan Project,” suggesting that national security and economic competitiveness hinge on the U.S. leading in this arena. The existing grid infrastructure presents challenges, but VPPs can leverage current resources to meet immediate demand.
The ramifications of DeepSeek’s announcement extend beyond immediate market impacts; they challenge the foundational assumptions about energy consumption in the AI sector. As the industry grapples with these changes, both opportunities and challenges will emerge for electric utilities and grid operators. The energy landscape is poised for transformation, and stakeholders must adapt to this evolving narrative.
