By M.V. Ramana
Image by Johannes Plenio.
One bright spot amidst all the terrible news last couple of months was the market’s reaction to DeepSeek, with BigTech firms like Nvidia and Microsoft and Google taking major hits in their capitalizations. Billionaires Nvidia’s Jensen Huang and Oracle’s Larry Ellison—who had, just a few days back, been part of Donald Trump’s first news conference—lost a combined 48 billion dollars in paper money. As a good friend of mine, who shall go unnamed because of their use of an expletive, said “I hate all AI, but it’s hard to not feel joy that these asshats are losing a lot of money.”
Another set of companies lost large fractions of their stock valuations: U.S. power, utility and natural gas companies. Electric utilities like Constellation, Vistra and Talen had gained stock value on the basis of the argument that there would be a major increase in demand for energy due to data centers and AI, allowing them to invest in new power plants and expensive nuclear projects (such as small modular reactor), and profit from this process. [The other source of revenue, at least in the case of Constellation, was government largesse.] The much lower energy demand from DeepSeek, at least as reported, renders these plans questionable at best.
Remembering Past Ranfare
But we have been here before. Consider, for example, the arguments made for building the V. C. Summer nuclear project in South Carolina. That project came out of the hype cycle during the first decade of this century, during one of the many so-called nuclear renaissances that have been regularly announced since the 1980s. [In 1985, for example, Oak Ridge National Laboratory Director Alvin Weinberg predicted such a renaissance and a second nuclear era—that is yet to materialize.] During the hype cycle in the first decade of this century, utility companies proposed constructing more than 30 reactors, of which only four proceeded to construction. Two of these reactors were in South Carolina.
As with most nuclear projects, public funding was critical. The funding came through the 2005 Energy Policy Act, the main legislative outcome from President George W. Bush’s push for nuclear power, which offered several incentives, including production tax credits that were valued at approximately $2.2 billion for V. C. Summer.
The justification offered by the CEO of the South Carolina Electric & Gas Company to the state’s Public Service Commission was the expectation that the company’s energy sales would increase by 22 percent between 2006 and 2016, and by nearly 30 percent by 2019. In fact, South Carolina Electric & Gas Company’s energy sales declined by 3 percent by the time 2016 rolled in. [Such mistakes are standard in the history of nuclear power. In the 1970s, the U.S. Atomic Energy Commission and utility companies were projecting that “about one thousand large nuclear power reactors” would be built “by the year 2000 and about two thousand, mostly breeder reactors, by 2010” on the basis of the grossly exaggerated estimates of how rapidly electricity production would grow during the same period. It turned out that “utilities were projecting four to nine times more electric power would be produced in the United States by nuclear power in 2000 than actually happened”.] In the case of South Carolina, the wrong projection about energy sales was the basis of the $9 billion plus spent on the abandoned V. C. Summer project.
The Racket Continues
With no sense of shame for that failure, one of the two companies involved in that fiasco recently expressed an interest in selling this project. On January 22, Santee Cooper’s President and CEO wrote, “We are seeing renewed interest in nuclear energy, fueled by advanced manufacturing investments, AI-driven data center demand, and the tech industry’s zero-carbon targets…Considering the long timelines required to bring new nuclear units online, Santee Cooper has a unique opportunity to explore options for Summer Units 2 and 3 and their related assets that could allow someone to generate reliable, carbon emissions-free electricity on a meaningfully shortened timeline”.
A couple of numbers to put those claims about timelines in perspective: the average nuclear reactor takes about 10 years to go from the beginning of construction—usually marked by when concrete is poured into the ground—to when it starts generating electricity. But one cannot go from deciding to build a reactor to pouring concrete in the ground overnight. It takes about five to ten years needed before the physical activities involved in building a reactor to obtain the environmental permits, and the safety evaluations, carry out public hearings (at least where they are held), and, most importantly, raise the tens of billions of dollars needed. Thus, even the “meaningfully shortened timeline” will mean upwards of a decade.
Going by the aftermath of the Deepseek, the AI and data center driven energy demand bubble seems to have crashed on a timeline far shorter than even that supposedly “meaningfully shortened timeline”. There is good reason to expect that this AI bubble wasn’t going to last, for there was no real business case to allow for the investment of billions. What DeepSeek did was to also show that the billions weren’t needed. As Emily Bender, a computer scientist who co-authored the famous paper about large language models that coined the term stochastic parrots, put it: “The emperor still has no clothes, but it’s very distressing to the emperor that their non-clothes can be made so much more cheaply.”
But utility companies are not giving up. At a recent meeting organized by the Nuclear Energy Institute, the lobbying organization for the nuclear industry, the Chief Financial Officer of Constellation Energy, the company owning the most nuclear reactors in the United States, admitted that the DeepSeek announcement “wasn’t a fun day” but maintained that it does not “change the demand outlook for power from the data economy. It’s going to come.” Likewise, during an “earnings call” earlier in February, Duke Energy President Harry Sideris maintained that data center hyperscalers are “full speed ahead”.
Looking Deeper
Such repetition, even in the face of profound questions about whether such a growth will occur, is to be expected, for it is key to the stock price evaluations and market capitalizations of these companies. The constant reiteration of the need for more and more electricity and other resources also adopts other narrative devices shown to be effective in a wide variety of settings, for example, pointing to the possibility that China would take the lead in some technological field or the other, and explicitly or implicitly arguing how utterly unacceptable that state of affairs would be. Never asking whether it even matters who wins this race for AI. These tropes and assertions about running out of power contribute to creating the economic equivalent of what Stuart Hall termed “moral panic”, thus allowing possible opposition to be overruled.
One effect of this slew of propaganda has been the near silence on the question of whether such growth of data centers or AI is desirable, even though there is ample evidence of the enormous environmental impacts of developing AI and building hyperscale data centers. Or for that matter the desirability of nuclear power.
As Lewis Mumford once despaired: “our technocrats are so committed to the worship of the sacred cow of technology that they say in effect: Let the machine prevail, though the earth be poisoned, the air be polluted, the food and water be contaminated, and mankind itself be condemned to a dreary and useless life, on a planet no more fit to support life than the sterile surface of the moon”.
But, of course, we live in a time of monsters. At a time when the levers of power are wielded by a megalomaniac who would like to colonize Mars, and despoil its already sterile environment.
M. V. Ramana is the Simons Chair in Disarmament, Global and Human Security at the School of Public Policy and Global Affairs, University of British Columbia and the author of The Power of Promise: Examining Nuclear Energy in India.
This article was published on March 13, 2025 at CounterPunch.