For a year, the market treated U.S. power producers as AI stocks. On January 27 it sold them like AI stocks too. Vistra closed nearly 30% lower, erasing its gains for 2025, according to CNBC. Constellation Energy, Talen Energy and GE Vernova each fell more than 20%. Gas producer EQT lost nearly 10%, and pipeline companies Kinder Morgan and Williams fell more than 8%. Advanced nuclear developers Oklo and NuScale dropped more than 20%, Utility Dive reported.
The trigger was a Chinese AI lab. DeepSeek released a model on Christmas Day and followed it last week with a reasoning model, DeepSeek-R1, that competes with OpenAI's o1. By Monday its assistant had overtaken ChatGPT as the most downloaded free app in Apple's store. The question investors asked was simple: if models this good can be trained so cheaply, how much electricity will AI really need?
Why power stocks were exposed
The companies that fell hardest were the ones whose valuations had come to rest most heavily on data center demand. Before Monday, Constellation, Vistra and GE Vernova had led the S&P 500 as investors bet that AI data centers would need enormous amounts of electricity. Constellation has signed a power agreement with Microsoft to restart the Three Mile Island nuclear plant outside Harrisburg. Talen supplies an Amazon data center from its Susquehanna nuclear plant. Vistra has no data center deal yet, but investors see value in its nuclear and gas fleet, CNBC noted. GE Vernova had soared on expectations that its gas turbine and grid businesses would benefit.
Even after the drop, the run-up had been large. Utility Dive noted that Vistra and Talen remained more than twice as expensive as a year earlier, and that Constellation was still up about 127% year on year. NuScale had gained 580% since January 2024.
The demand thesis under question
Analysts put numbers on the risk. Jefferies' power and utilities team wrote that DeepSeek's success "calls into question the significant electric demand projections for the U.S.," because AI represents about 75% of overall U.S. demand forecasts through 2030 to 2035 in most projections. It said the bull case for independent power producers and most integrated utilities is entirely dependent on data centers. A slowdown in data center projections, it added, would hurt utilities that investors expect to grow their rate base. The bank's own $274 price target for Constellation, it said, was premised on 75% of the nuclear portfolio's output being sold at $80 per MWh, with a 50% probability. Assumptions like that are exactly what a cheaper model calls into question.
Bank of America analysts said DeepSeek was challenging the notion of U.S. leadership in AI and raising doubts about high expectations for cloud capital spending, chip growth and power requirements, according to CNBC. Microsoft chief executive Satya Nadella described DeepSeek as "super-compute efficient."
Utility Dive cited a blog post by technology investor Jeffrey Emanuel estimating that DeepSeek's models may have cost 45 times less to train than leading U.S. products. That estimate is not an official figure, but it captured why the market moved: if efficiency gains of that size are real and widely adopted, the compute and power required for each unit of AI capability falls sharply.
The forecasts at stake
The selloff landed on top of a stack of aggressive forecasts. In December, Lawrence Berkeley National Laboratory estimated that U.S. data centers consumed about 176 TWh in 2023, about 4.4% of national electricity, and that the share could reach 6.7% to 12% by 2028. Utility Dive noted that the Electric Power Research Institute said last year that data centers' share of U.S. load could double to 9% by 2030, and that ICF forecast U.S. electricity demand growing by an average of 2% a year through 2033. PJM's preliminary forecast in December showed summer and winter peak load growing by averages of 2% and 3.2% a year through 2045, up from 1.6% and 1.8% in its 2023 forecast.
Those forecasts drive real decisions: capacity procurement, transmission plans, utility resource plans, and power contracts like Constellation's with Microsoft. Less than a week earlier, OpenAI and its partners had announced Stargate, a pledge to invest $500 billion over four years in U.S. AI infrastructure. DeepSeek's release did not change any of those commitments. It changed how much investors were willing to pay for the assumption that they would all be fulfilled.
The other side of the argument
Not every analyst read the news as bearish. J.P. Morgan Wealth Management's strategy team pointed to the Jevons paradox, which holds that greater efficiency can increase total consumption of a resource. Cheaper AI models could lead to faster adoption by companies and households, and therefore to more computing, not less. Other analysts quoted by Utility Dive suggested that DeepSeek could accelerate investment by U.S. technology companies.
ICF's Himali Parmar told Utility Dive that it is too early to tell whether projections account for model efficiency gains. If some of the demand increase is tempered by energy-efficient AI, she said, that could lighten the burden on the grid, supply needs and customer bills.
What it means for the grid
For grid operators, the event is a reminder that data center demand is the least certain part of every load forecast. Data center load comes in large blocks, it depends on corporate decisions that can change quickly, and it rests on assumptions about technology that shift every few months. A model breakthrough can alter the investment case for an entire campus.
But physical systems move more slowly than stock prices. The interconnection requests already filed, the capacity already contracted, and the plants already being restarted will not disappear because of one model release. The more likely effect is greater scrutiny of which projects are firm and which are speculative. That would be healthy. Planners who have been asked to build for very large and uncertain loads now have a fresh argument for requiring stronger financial commitments before they do.
For power producers, the lesson is about concentration. Companies whose value depends on a single source of demand growth inherit that source's volatility. Monday showed how quickly the market can reprice that risk.