A recent sell-off of energy and infrastructure stocks was triggered this week by developments in artificial intelligence from the Chinese start-up DeepSeek, highlighting a lack of understanding regarding AI’s power demands, according to the International Energy Agency (IEA).
Investors sharply reduced the valuations of various companies, including European gas turbine manufacturers and US power generation firms, following DeepSeek’s unexpected breakthrough that suggested AI might require significantly less power than earlier estimates indicated.
Affected companies included Siemens Energy, GE Vernova, Mitsubishi Heavy Industries, Constellation Energy, Vistra Energy, Schneider Electric, ABB, and Legrand. Schneider Electric, which had previously surpassed TotalEnergies in market valuation as investors anticipated the rising energy demands of AI, has now fallen behind the French oil giant once again in market capitalization.
Thomas Spencer, an analyst at the IEA focusing on energy and AI, indicated that the market currently lacks the necessary tools and information to correctly evaluate the implications of AI on electricity demand. He noted that fluctuations in the energy sector were expected due to the uncertainty about how AI impacts energy consumption in data centers.
DeepSeek’s model reportedly required less than 10% of the computational resources necessary for Meta’s Llama model, which suggests that its energy consumption is significantly lower compared to its western counterparts. Analysts from Rystad commented that since electricity use in data centers is linked to their computational power, the veracity of DeepSeek’s claims could mean a substantial reduction in overall energy consumption.
Analysts from Citi indicated that if DeepSeek’s figures hold true, it could substantially impact companies involved in energy infrastructure. Data center suppliers had previously enjoyed positive growth forecasts, but more efficient AI could alter these expectations.
The sell-off raised concerns within the energy sector, with a senior Siemens Energy executive stating that the company’s 21% drop in share price was “not reasonable.” Siemens Energy, which spun off from the gas and power division of Siemens four years ago, had seen its stock increase by 336% over the previous year prior to the decline. The company reported €9 billion in revenues in the first quarter, exceeding forecasts, and projected an 18% year-on-year organic growth rate. Siemens Energy noted that the share decline was exaggerated, as it expects only a small portion of its future growth to come from data centers.
The executives of an electricity infrastructure firm held discussions to gauge the impact of the DeepSeek news, concluding that stability would come gradually if financial results remained positive. Others argued that while DeepSeek’s model might consume less power, it could accelerate AI adoption, leading to increased energy demand overall.
This phenomenon is known as the Jevons paradox, which describes how advancements in efficiency can lead to increased consumption of a resource. Microsoft’s CEO remarked on this, suggesting that as AI becomes more efficient and widely available, its usage is likely to surge.
Rystad also observed that a decrease in power-intensive AI could influence the broader energy transition, as major tech companies have been some of the largest purchasers of zero-carbon electricity. For instance, Amazon, the world’s biggest corporate buyer of renewable energy, has signed agreements enabling it to procure 77 terawatt hours of electricity annually, equivalent to Belgium’s total electricity consumption. With reduced renewable energy needs for AI, more could be directed towards other sectors, potentially accelerating the shift away from fossil fuels.
photo credit: www.ft.com
