SMRs and AI as the New ‘Pump and Dump’: Hyped, Extractive, Exploitative, and Toxic

This title is “doing a lot of work” as we often say in poetry workshops. Still, it was necessary to pull out the main takeaway: the threat that overhyped small modular reactors (SMRs), which are actually more toxic than conventional nuclear reactors but which are being widely embraced as the solution to generative AI’s massive energy costs, pose to people and planet. This is the key point I worry will get buried under all the other information I had to share in order to counter the usual responses from (a) climate change deniers; (b) otherwise concerned environmentalists who are AI users and/or techno-optimists and who, therefore, make exceptions for generative AI; and (c) pro-nuclear progressives. Characteristically-scary-and-yet-techno-optimistic ex-Google CEO Eric Schmidt’s assertion that we should focus on ramping up energy- and water-guzzling generative AI because “we are never going to meet our climate goals anyway” and the steady uptick in generative AI use even as people and planet are being hit with extreme, increasingly frequent environmental disasters in the form of (a) stronger-than-normal hurricanes and floods here in the United States, (b) catastrophic floods in northern Africa, Nigeria, South Africa, southeast Asia, central Europe and France, and (c) historic droughts in southern Africa convinced me of the urgency of this piece. For those who want to reverse this regressive tide, I hope this deep dive, which is chock-full of links to reputable sources, will serve as a handy resource that can be shared with others.

The Emerging Consensus on Generative AI’s Environmental Costs

In 2020 Microsoft pledged to not just be carbon neutral but carbon negative by 2030, a time period the United Nation’s Intergovernmental Panel on Climate Change identified as our last chance to keep global warming to within 1.5 degrees Celsius above preindustrial levels or face significantly worse environmental destruction that may also be irreversible. Google committed to being carbon neutral by 2030. Meta and Amazon also committed to net-zero-by-2030 goals. Yet every one of these companies has dramatically increased its emissions to fuel generative AI. This at a time when 64% of people in the world (in particular, younger people) perceive climate change as a global emergency.

Generative AI, the technology many use to output text, images, audio, and video for fun, is incredibly energy and resource intensive. Despite Bill Gates’ self-serving admonition that we should “not go overboard on” worrying about generative AI’s energy use (which sounds eerily similar to Andreessen Horowitz’s proclamation that we should, as Business Insider’s Kali Hays put it, “stop talking about AI’s copyright issues”), a consensus that generative AI is massively extractive, exploiting not just people—namely, artists and data workers—but also the planet, is emerging.

Generative AI’s Energy Demands

According to France 24, Microsoft and Google “saw their greenhouse gas emissions soar in 2023 because of AI: up 48 percent for Google compared to 2019 and 29 percent for Microsoft compared to 2020.” These figures, sourced from the companies’ own environmental reports, are published in numerous reputable news sources.

The LA Times reports that, according to Goldman Sachs, “a query on OpenAI’s ChatGPT, on average, requires nearly 10 times as much electricity to process as a Google search” (emphasis added).

According to the International Energy Agency (IEA), which released a 170-page analysis and forecast of energy use that spans two years (up to 2026):

Electricity consumption from data centres, artificial intelligence (AI) and the cryptocurrency sector could double by 2026. Data centres are significant drivers of growth in electricity demand in many regions. After globally consuming an estimated 460 terawatt-hours (TWh) in 2022, data centres’ total electricity consumption could reach more than 1 000 TWh in 2026. This demand is roughly equivalent to the electricity consumption of Japan. (8, original emphasis in bold, emphasis added in italics)

Vox reports that “training a large language model like OpenAI’s GPT-3, for example, uses nearly 1,300 megawatt-hours (MWh) of electricity, the annual consumption of about 130 US homes.” Citing the IEA, Vox (like Goldman Sachs) notes, “a single Google search takes 0.3 watt-hours of electricity, while a ChatGPT request takes 2.9 watt-hours.”

In that same Vox article, Sasha Luccioni, lead climate researcher at Hugging Face, states, “From my own research, what I’ve found is that switching from a nongenerative, good old-fashioned quote-unquote AI approach to a generative one can use 30 to 40 times more energy for the exact same task” (emphasis added).

Generative AI’s Water Demands

All of the statistics above don’t even address another pressing concern that will become increasingly urgent: generative AI’s massive water use. This is a topic that, as with questions about what data is used to train AI and about how much energy goes into both training generative AI models and generating outputs, AI companies shroud in secrecy. As ridiculous as it sounds, AI companies like Microsoft now appear to treat all data that could call into question the necessity of their product—information about copyrighted artwork stolen from artists to train generative AI and information about the environmental impacts of generative AI—as “proprietary,” preventing the media, public, and impacted parties, such as artists, from accessing this crucial information. “In AI Is Taking Water from the Desert,” Karen Hao writes:

Exactly how much power does this Goodyear data center use, and how much of it is renewable? Neither Microsoft nor the local utility company would say. As for water use, a records request to the city returned documents with all of the numbers redacted; a representative for the city said the numbers were ‘considered proprietary by Microsoft.’

Nonetheless, researchers have managed to pry the lid off the black box and reveal some of the AI industry’s dirty secrets. According to the Newsweek article “Why AI Is So Thirsty: Data Centers Use Massive Amounts of Water” by Jeff Young, Shaolei Ren, who is an associate professor in the Electrical & Computer Engineering Department at the University of California, Riverside, calculates that “global AI demand could result in as much as 6.6 billion cubic meters, or 8.6 billion cubic yards, of water withdrawal by 2027. . . [which] ‘will be roughly equivalent to four to six Denmarks of national water withdrawal’ and is ‘quite a lot’ (emphasis added).

Moreover, collaborative work between The Washington Post and researchers at the University of California, Riverside—research in which Ren was involved—revealed that using ChatGPT-4 to generate just one 100-word email requires a little more than one 519-milliliter or 16-ounce bottle of water.

Lest you assume that this is all reused or reusable water, think again. In “How much water does AI consume? The public deserves to know,” Ren writes:

AI models consume fresh water in two ways. . . . [1] To dissipate the heat into the outside environment and avoid server overheating, data centres commonly use cooling towers and/or outside air, which need a staggering amount of clean, fresh water. . . . [2] Generating electricity also consumes a lot of [fresh] water through cooling at thermal power and nuclear plants and expedited water evaporation caused by hydropower plants. (emphasis added)

Not only does this deprive a community of its drinking water, but it also facilitates or exacerbates drought. In September 2023, Matt O’Brien and Hannah Fingerhut of The Hill reported that Microsoft, planned to open its fourth and fifth data centers in Des Moines, Iowa, that year and was, according to Des Moines Mayor Steve Gaer, “building them [data centers] as fast as they can.” In March 2024, Newsweek’s Jeff Young reported that the National Oceanic and Atmospheric Administration had designated 85% of Iowa as in drought “with 56 percent of the state . . . in extreme or severe drought.” State Geologist Keith Schilling listed "data centers requiring vast quantities of cooling water" as one of the contributors to declining groundwater levels in the state’s aquifers.

Ren offers one way to mitigate generative AI’s environmental costs. The LA Times’ Melody Petersen writes:

Ren said users should be told on the websites where they are asked to type in their queries how much energy and water their requests will require. He said this would be similar to how Google now tells people searching for airline flights the amount of carbon emissions the trip will generate.

Such transparency is a step in the right direction. However, given the United States’ consistent ranking as one of the top-two greatest polluters on Earth, this is far from sufficient. There is no way to ensure that generative AI improves life for rather than devastates people all over the world at all socioeconomic levels (but, especially, from the bottom up) except government regulations that are truly independent from corporate influence.

This article is mostly focused on the nuclear power “solution” to generative AI’s energy costs—a “solution” that is currently as fake as the industry it is intended to power (that is as fake as generative AI itself)—so I won’t go in depth into the topic of how generative AI is robbing communities of fresh drinking water and pushing them toward drought.

However, as recent hurricanes and floods in the US and Europe should have convinced all of us, we are now (like the rest of the world, which usually suffers most from our government’s unwillingness to strictly regulate industries in order to fight climate change) at an environmental tipping point. Generative AI is tipping the scales rapidly toward climate disaster.

Toxic Love: Generative AI and Nuclear Power

Desperate, apparently, to fend off criticism, Microsoft funded “independent researcher” Jonathon Koomey, who cast doubt on the emerging consensus from experts that generative AI is accelerating climate change in this unscientific-and-not-at-all-peer-reviewed Nature article (written by Koomey, another Microsoft-funded writer, three Microsoft employees, and two scholars) that has been touted by generative AI supporters (who often usually cite obscure sources or blogs as proof that, despite ample reporting from various reputable sources, generative AI is not causing much environmental harm). (See, also, this MIT Technology Review article in which Koomey downplays generative AI’s environmental impact.)

The tech industry, as a whole, has also turned to the “solution” that it appears Bill Gates and protégé apparent Sam Altman anticipated would be necessary years ago when each invested in it: nuclear power. Gates cofounded heavily government-subsidized nuclear fission company TerraPower in 2008. (According to Business Insider, the Department of Energy paid $1.9 billion or nearly half the expected cost for TerraPower’s first experimental reactor and covered 80% of the cost for another experimental reactor Terra Power partnered with Southern Company and a few other companies to build.) Business Insider’s Britney Nguyen describes how Altman courted nuclear fission company Oklo, which makes small modular reactors, over a decade ago (in 2013). According to Business Insider’s Darius Rafieyan, in 2021 Altman also invested heavily in nuclear fusion company Helion Energy, stating, "The obvious reason is of course AI." Note that both TerraPower and Oklo build small modular reactors or SMRs. This is important for obvious reasons this article makes apparent. Nuclear energy, which was so dead in the water that in 2018 Ahmed Abdulla, an assistant research scientist at UC San Diego’s Center for Energy Research, wrote, “There seem to be no innovations or policy changes on the horizon that could provide relief for the future of nuclear technology,” is experiencing a renaissance due largely to government subsidies and the tech industry’s embrace of it as the panacea to the problem of generative AI’s energy burden.

Years ago, I wrote about the nuclear industry’s brazen attempts to ride the coattails of the green revolution and rebrand nuclear energy as “clean and green.” This rebranding effort continues. A more recent example of this is Microsoft-supplier Constellation Energy’s “ComeClean” campaign, which goes so far as to tell cashiers that it is safer to work in nuclear power plants than at grocery stores! Meanwhile, as I note below, thousands of nuclear industry workers are seeking compensation from the federal government for job-related, debilitating illnesses like thyroid and brain cancer. Nonetheless and among other tactics, the nuclear industry’s massive, aggressive effort to convince the public that nuclear is on par with renewables like solar and wind in terms of its “cleanliness” has included:

1. repeating the claim that it is safe to live near and work in nuclear power plants despite findings in 2024 from researchers that people who live near nuclear power plants have higher cancer rates, findings from the International Nuclear Workers Study (or INWORKS) of “a positive association between prolonged low-dose exposure to ionizing radiation and mortality from . . . hematological cancers,” findings in 2021 that “brain cancer incidence rates are positively associated with the number of nuclear research reactors per state,” and previous conflicting conclusions in 2012 from studies on the impact that proximity to nuclear power plants has on childhood cancer rates
2. insisting that a pro-nuclear scientific consensus exists even though there is a lot of disagreement between/no consensus amongst environmental scientists and other experts that it is wise to continue to use or even upscale nuclear power in order to fight climate change
3. funneling defense money to civilian universities and thereby producing young, predictably pro-nuclear scientists
4. dismissing the fact that nuclear power plants usually cost approximately twice what is projected and, as this excellent piece by floodlight’s Kristi E. Swartz notes, this has been true even for SMRs, with SMR-maker NuScale already canceling a major contract due to the exorbitant increase in what it would ultimately cost a community-owned utility’s customers

Generative AI and the Nuclear Renaissance

While anti-nuclear activists are often accused of fear mongering, an unacknowledged irony is that the nuclear industry plays into the public’s fears about climate change, claiming that it offers “clean” energy while downplaying how actually dangerous academic researchers and government regulators admit nuclear power remains. Unfortunately, the nuclear industry has been successful at this opportunistic and predatory greenwashing.

However, I strongly suspect most Americans are unaware of how poorly regulated our country’s nuclear industry is and of the extent to which negligence has characterized it. In addition to the information in the previous section, consider the following:

1. Despite a study conducted by Stanford and University of British Columbia researchers which showed that the “next generation,” more aesthetically pleasing, and widely hyped small modular reactors (SMRs) that AI investors are championing “actually generate more radioactive waste than conventional nuclear plants,” this massive effort to project outsized confidence in nuclear power, including SMRs, is coming from the top (from our own government). The romanticization of SMRs is especially troubling when we consider that, unlike conventional nuclear power plants, which are usually built with a lot of attention to issues like how prone an area is to natural disasters and the site’s proximity to drinking water, SMRs are being marketed as nuclear plants that can be prefabricated in a cookie-cutter fashion (which is, in and of itself, troubling) and then relocated/assembled at sites that normally would not be suitable for conventional nuclear reactors. This will (a) spread nuclear risks, such as radioactive spills and storage leaks, across wider areas; (b) likely result in one-size-fits-all SMRs that are not actually suitable for diverse environments; and (c) make it even more difficult for regulators to inspect, investigate, monitor, and otherwise hold such plants accountable for violations. Given the internal and external criticisms of how well the Nuclear Regulatory Commission has regulated the industry to date (see below), a significant increase in the number of nuclear plants across a wider geographic range poses serious regulation hurdles. Moreover, since there is no guarantee that just one SMR would be placed at a site, nuclear plants that operate SMRs might not necessarily produce less nuclear waste than conventional plants.
2. Although both the nuclear industry’s negligence in keeping records and the federal government’s negligence in monitoring plants, as well as a history of either outright lying or bad science on the part of the U.S. Agency for Toxic Substances and Disease Registry (ATSDR), have, unsurprisingly, made it difficult to get accurate estimates of deaths—a twisted reality that serves pro-nuclear advocates well—it is now well established that waste from nuclear power plants or uranium mining has caused cancer in communities in (a) New Mexico (see, also, here) and, (b) due to the Boeing-Rocketdyne Nuclear Facility accident, which was actually the worst nuclear accident in American history and was initially covered up, Los Angeles, as well as (c) cancer in nuclear plant workers around the country. The latter is a problem so great that ProPublica reports “[Bill] Richardson, the former energy secretary who established the employees’ compensation program in 2000, said work at nuclear facilities is ‘still dangerous’” and “estimate[s] that hundreds of thousands of workers could be eligible for benefits but lack the records to support their claims” (emphasis added).
3. ProPublica also reports that “for years, [investigator George Mulley] documented how the [Nuclear Regulatory Commission] dropped the ball on the handling of nuclear fuel and security in nuclear plants.” This issue appears to have gotten worse as the NRC’s Office of the Inspector General, which ProPublica describes as “an independent agency that serves as watchdog to the watchdog [that is the NRC]” has become even more reticent to, essentially, do its job.
4. Although many would expect decommissioning a nuclear plant to be undertaken by experienced professionals, it seems that our government is fine with entrusting this delicate work to a company like Holtec, which despite (a) having no experience in decommissioning plants, (b) having been caught, disbarred, and fined for bribing a federal employee in order to secure a contract to manage radioactive waste, and, (c) as Bloomberg Business Week’s Will Wade notes, being compelled to pay another fine of a few million dollars “to avoid [criminal] prosecution in New Jersey after it was accused of falsifying records,” still secured potentially highly lucrative contracts to decommission no less than four nuclear power plants. Holtec is also now reopening and operating one of these plants and building SMRs.

Pro-nuclear propaganda also appears to be winning out despite the very literally persistent (centuries to millennia long) problem of nuclear waste storage, which the United States has never handled well and which must now be contended with against the foreground (not backdrop) of a far less predictable climate, and despite the aforementioned and equally persistent problem of not just human error but outright negligence. The latter has resulted, for example, in spills of radioactive waste into waterways in Philadelphia, Pennsylvania, where there is (surprise?) a high incidence of childhood cancer—at a plant run by Constellation Energy, the Microsoft energy supplier that I noted is aggressively pushing the message that nuclear power is clean through its “ComeClean” campaign.

Constellation appears to be in the process of rebranding the plant at which the spill occurred (Limerick Generating Station) as Limerick Clean Energy Center (emphasis, again, on clean). Crucially, Constellation will also be managing the reopening and operation of one of the units—TMI-1—at the Three Mile Island nuclear power plant, which is where a different unit—TMI-2—had a partial meltdown due to both, according to History.com, “mechanical and human errors” that resulted in “the worst [or at least most well-known] commercial nuclear accident in U.S. history.” The TMI-1 unit is being reopened to fuel Microsoft’s data center needs—to power generative AI and, says Microsoft, homes.

Conclusion

Today I’m alarmed by the extent to which the US populace has swallowed the propaganda that nuclear energy, which Bill Gates, Sam Altman, and other AI tycoons have marketed as a “solution” to generative AI’s massive energy consumption, is now “clean” and should be embraced to fight climate change. I’m also disturbed by our government’s blatant, simping encouragement of this dangerous technology. To be clear, I support upgrading and using existing nuclear power plants to meet necessary rather than extraneous (generative AI-induced) energy demands. I’m also very cautiously optimistic about nuclear fusion as a future energy source but strongly believe experimental nuclear testing should be confined to remote locations. New investments in nuclear fission, including SMRs, however, are misguided, regressive, and unnecessary.

If generative AI’s future depends on harnessing such a dangerous form of energy, it simply is not worth it. This is especially true when many use generative AI for fun or for tasks that could be done better and with lower social costs and higher societal benefits by actual human beings, such as creative workers (or artists).

BONUS RESOURCES

Below are a few straightforward or easy-to-understand resources that didn’t make it into the main piece above (or weren’t adequately highlighted) but which I found compelling. These are all videos, and so you can watch or listen to them at your leisure.

1. In the article above, I don’t discuss some of nuclear power’s other shortcomings, including the fact that (as Greenpeace activists who have repeatedly broken into heavily guarded nuclear facilities have shown), nuclear power plants can become targets for terrorists or, in times of war, adversaries. Greenpeace has an excellent short video that sums up these additional arguments against nuclear power. Watch it here.
2. Bill Nye remains objective, consistent, and trustworthy in his critiques of nuclear power. Check out his videos!
3. Theoretical physicist Sabine Hossenfelder offers a fairly objective, dispassionate analysis of nuclear power. I arrive at a different conclusion than Hossenfelder’s mostly neutral stance and attribute this to the following: (a) she seems to be unaware of the fact that SMRs produce more radioactive waste than conventional nuclear power plants, (b) she doesn’t seem to be aware of the nuclear industry’s history of negligence and deception in the United States, (c) she seems to also be unaware of studies that link proximity to or employment in nuclear plants with cancer, (d) for no apparent reason and despite climate change making extreme weather less predictable, she dismisses valid concerns about the problem of nuclear waste storage, and (e) she avoids discussing the national security risks associated with nuclear power plants.