- AI use is surging astronomically around the globe, requiring vastly more energy to make AI-friendly semiconductor chips and causing a gigantic explosion in data center construction. So large and rapid is this expansion that Sam Altman, the boss of OpenAI, has warned that AI is driving humanity toward a “catastrophic energy crisis.”
- Altman’s solution is an audacious plan to spend up to $7 trillion to produce energy from nuclear fusion. But even if this investment, the biggest in all of history, occurred, its impact wouldn’t be felt until mid-century, and do little to end the energy and water crises triggered by AI manufacture and use, while having huge mining and toxic waste impacts.
- Data centers are mushrooming worldwide to meet AI demand, but particularly in Latin America, seen as strategically located by Big Tech. One of the largest data center hubs is in Querétaro, a Mexican state with high risk of intensifying climate change-induced drought. Farmers are already protesting their risk of losing water access.
- As Latin American protests rise over the environmental and social harm done by AI, activists and academics are calling for a halt to government rubber-stamping of approvals for new data centers, for a full assessment of AI life-cycle impacts, and for new regulations to curb the growing social harm caused by AI.
“We walk for the water we need. If we don’t walk, who will give it to us?” asks Juan, a middle-aged man born in the Indigenous rural community of Maconí, Mexico. “It’s a four-hour journey each day to fetch water … Since last year, there hasn’t been rain, and this year it’s the same.” The bean crop has withered and there’s no corn to make tortillas, he told Ana Valdivia, an AI expert from the U.K.’s Oxford Internet Institute.
Adding to Maconí’s climate change-driven water crisis is the extreme suck placed on its water supply by an influx of new internet data centers, whose already astronomical consumption of limited community water and electricity is about to surge again as AI drives an exponential leap in global demand for computer chips and data capacity.
Valdivia, who is about to publish a paper titled “The supply chain capitalism of AI,” met Juan during an October 2023 community protest in Querétaro state, north of Mexico City, where rural people demanded fundamental water access rights.
Querétaro is the only Mexican state with its entirety at high risk of drought. It is also where Big Tech has chosen to develop what could become the largest data center hub in Latin America.
“Querétaro is already hosting 10 functioning data centers, and planning to install 18 more, some of them to support the growing demand of ChatGPT,” says Valdivia. “Data centers are extracting drinking water for their economic businesses, whereas Juan has to walk almost a day to water his beans and nixtamal [used to make tortillas].” Mexico is ideally situated for the mushrooming data industry, explains Valdivia, due to its location between North America and Central and South America via undersea cables.
Mexico’s government has welcomed the data processing and storage industry even as local communities, who seem unlikely to benefit much from the boom, demonstrate against it.
But Mexico’s Secretariat of Environment and Natural Resources told Mongabay that the government has not authorized new concessions for data centers in Querétaro state and that the data centers already established buy the energy they need following the regulations drawn up by Querétaro’s State Water Commission.
The driving force behind rapid data center growth in Mexico and throughout the developing world is now artificial intelligence. Data centers and chip manufacturing facilities, which are already sucking up water and energy at unsustainable rates, are about to multiply across the world due largely to AI, likely triggering energy and water wars between corporations and communities.
Artificial intelligence boom, environmental bane
AI demand is growing at lighting speed, with ChatGPT gaining 1 million users within the first five days of its release in November 2023. AI is expected to grow by 37% from 2023 to 2030, according to Grand View Research, a market intelligence firm. But that’s a prediction some see as a major underestimate. One preprint suggests the global demand for water alone to make chips and cool AI data centers could equal half that of the U.K.’s consumption by as early as 2027.
Producing an AI chip requires 10-15 times more energy than making a standard chip. This is because AI machine learning requires a different kind of computer processor, called a graphics processing unit (GPU), which uses models to carry out increasingly complex tasks. GPUs devour huge amounts of energy: Whereas in 2020 it took about 27 kilowatt-hours of energy to train an AI model, by 2022 this rose to 1 million kWh, a mind-boggling 37,000-fold increase. All this computing power requires huge sums of electricity and water for cooling.
AI will be the key driver in an expected doubling of global data center electricity demand by 2026. Along with this comes a huge increase in water use. Mainstream media are already reporting that AI energy demand could overwhelm and even break the U.S. power grid.
So how is Big Tech responding? Google told Mongabay that it has become more efficient. A spokesperson said: “Google data centers are designed, built and operated to maximize efficiency — they are more than 1.5x as energy efficient as a typical enterprise data center and, compared with five years ago, we now deliver approximately three times as much computing power with the same amount of electrical power.”
Moreover, Big Tech firms like Microsoft, Meta, Amazon, Apple and Google are confident that the gargantuan computing power they offer will be good for all. “Increased investment and a policy focus on AI technologies can unlock new opportunities, from health care and sustainable agriculture to financial services and more,” Google CEO Sundar Pichai promised on the company’s “Google in Latin America” website.
But as the industry roars forward, with its PR machine trumpeting its green credentials, such as this amusing video from Apple, critics remain unconvinced. They note, for example, that the primary purpose of AI today is advertising, which could further accelerate the global overconsumption crisis destroying Earth’s environment.
AI has other concerning features. Hundreds of chemicals are involved in the manufacture of chips, including the highly toxic PFAS, a family of about 12,000 chemicals that won’t break down in the environment for tens of thousands of years, earning them the moniker “forever chemicals.” Described as essential for semiconductor manufacturing, these synthetics “bioaccumulate,” that is, they get absorbed by organisms faster than they can be excreted, so they build up inside the body over time. PFAS have been linked to serious diseases, including cancers. And while the semiconductor industry has phased out some PFAS use, the industry continues relying on forever chemicals.
In addition, the menu of needed minerals to make chips has soared from 11 to more than 60, including gallium and germanium, which lack federal quality standards in the U.S. and elsewhere, and for which knowledge of health and ecological effects are “limited,” according to the U.S. Environmental Protection Agency. If this chemical cocktail were to pollute Latin America’s water supply, governments there would have few resources for major cleanups.
A history of chip manufacturing
Chipmaking first blossomed in the U.S., mainly in California’s Silicon Valley, but companies soon relocated most of their production to other countries to reduce costs and possibly deflect criticism at home over rising levels of pollution.
For Big Tech giants, it was a sensible business decision. “You keep all the design and IP [intellectual property] at home and you do all your sales, marketing and services at home too, and that’s where you make the money,” said William Alan Reinsch, a senior adviser at the Center for Strategic and International Studies, a national security think tank. But he added: “You build a big factory [abroad] and you crank these [chips] out by the thousands, and you do it in a low-wage, non-union country that probably doesn’t have environmental requirements.”
Taiwan was an early destination of what Valdivia describes as a colonial approach. The Taiwan Semiconductor Manufacturing Company (TSMC) is now the world’s largest chipmaker. Among its customers is Apple, which it supplies with the chips that power iPhones. But it didn’t take long for environmental problems to emerge in Taiwan, with high concentrations of PFAS found in local rivers as far back as 2009. Recent severe droughts have also pitted local farmers against chip manufacturers there.
As relations between China and the U.S. soured, manufacturers felt vulnerable. “If China would invade Taiwan, that would be the biggest impact we’ve seen to the global economy,” Glenn O’Donnell, research director at Forrester Research, told Business Insider.
“Semiconductors have become almost like the oxygen of the global economy. Without the chips, you can’t breathe,” O’Donnell added. And sooner or later this invasion seems likely, with controlling Taiwan being central to Chinese President Xi Jinping’s goal of achieving a “great rejuvenation of the Chinese nation” by 2049, the 100th anniversary of the People’s Republic of China.
Tech firms have since sought to reduce their dependence on Taiwan. Some are adding U.S. facilities: Intel, one of the world’s largest semiconductor chip manufacturers, is building factories at a cost of more than $20 billion in Ohio; Micron, a producer of computer memory and computer data, will spend $100 billion on a chip factory in New York state; South Korea’s Samsung Electronics, the world’s largest information tech firm and chipmaker, is building a $25 billion plant in Texas; and TSMC is investing $40 billion for two plants in Arizona. Two of these locales, Arizona and Texas, already suffer from severe climate change-driven water shortages, and that crisis could collide sharply with chip manufacturing. In 2022, for example, TSMC alone consumed about 97 million metric tons of water for its business.
Latin America as AI manufacturing hub
Today, high-tech companies are searching for chip factory and data center locations outside the U.S., particularly in Latin America. “Latin American countries tend to have lower environmental regulations than the U.S. and Europe. And energy and water are cheaper,” Sebastián Lehuedé, a lecturer in ethics, AI and society in the Department of Digital Humanities at King’s College, London, told Mongabay.
In 2023, Costa Rica and Panama announced the intention to create the first chip hub in Central America. In 2024, the Dominican Republic made clear that it wanted to play a key role in chipmaking. Brazilian and Taiwanese authorities have been in discussion about “’unlocking the ‘potential’ in Brazil’s semiconductor market.” In October 2023, the U.S. Dallas News reported that as “the U.S. and China vie for dominance in the global semiconductor industry, Latin America has become a key battleground.”
The volume of data processed in Latin American centers has doubled since 2020, according to CBRE. Processing volume is expected to grow by more than 9% a year between 2024 and 2029, according to Mordor Intelligence, with the volume of data processed in Mexico even expected to outpace that frenetic rate, exceeding 11% annual growth between 2021 and 2026. What we’re seeing is the “meteoric rise of data centers in Latin America,” according to Layer 9, a data center provider focused on the region.
Costa Rica: A Big Tech hotspot
Because Big Tech presents itself as clean, green and sustainable, the industry was particularly pleased with its plans for a manufacturing hub in Costa Rica — the country widely regarded as the world’s trendsetter in environmental sustainability, paired with wise economic and socially conscious development.
“The United States views Costa Rica as a partner in ensuring the semiconductor supply chain can keep pace with the digital transformation underway,” said Cynthia Telles, the U.S. ambassador to Costa Rica, in July 2023.
“Costa Rica’s gross domestic product cannot be understood without the operations of Intel in the country,” Pablo Gámez Cersosimo, a Costa Rican researcher specializing in technology and biodiversity, told Mongabay. “Since 2020, Intel has exported 28 million product units to 128 clients in 44 countries. It operates a Research and Development Center dedicated to the design, prototyping, testing, and validation of integrated circuit and platform solutions; as well as an Assembly and Processor Testing Plant. It also operates a Global Services Center. Intel employs 3,300 people directly in Costa Rica and is currently undergoing a $1.2 billion expansion.”
But its role as a high-tech hub poses real environmental risks for Costa Rica. Almost all the country’s energy comes from renewables, mainly hydropower and some wind. But in 2023, Costa Rica suffered a severe drought, and that came with just a 1.2° Celsius (2.2° Fahrenheit) increase in the average annual global temperature since the preindustrial era.
“The lack of rain led to a decrease in Costa Rica’s capacity for hydroelectric power generation,” Gámez Cersosimo explains. During 2023’s drought, the government was forced to import oil to run its thermal power stations.
“Costa Rica is at risk from … natural disasters,” according to the Global Facility for Disaster Reduction and Recovery: “The trend over the last 40 years suggests a strengthening of the hydrological cycle, with more intense rain occurring during shorter periods.” That trend is expected to worsen in future as climate change brings, “a greater frequency or intensity of extreme events such as floods and droughts.”
According to former environment vice minister Rafael Gutiérrez, the government’s commitment to environmental sustainability has flagged over the years. “We did very good things decades ago, and we still have good results from that,” he said. “But controls have been weakened, and budgets have been reduced. In some sectors, the idea that the environment is an obstacle to development has grown.”
Gámez Cersosimo fears that water- and energy-intensive chip manufacturing, along with escalating climate change, will make everything worse. “In this context, we witness the arrival of the semiconductor industry, which is dependent on millions of liters of water per day,” he said.
Environmental and social impact
Even before AI begins to make its epic power and water demands, reports are already coming in from around the world of the impact data centers are having on supplies. The number of centers is growing both in Nigeria — where they sit next door to citizens who struggle every day to get enough clean drinking water — and in Uruguay, which is still pressing ahead with data center expansion despite suffering from a record drought. Even in India, a country where rivers are drying up and groundwater is almost depleted, there is robust data center growth.
According to Valdivia, the serious socioenvironmental problems these centers cause stem from inadequate evaluation during planning because authorities, when they give the go-ahead for a data center, fail to look at the facility’s complete life cycle. “The supply chain starts with natural resource extractivism, proceeds to chip manufacturing, continues with data centers’ operating data and AI, and concludes in e-waste dumping,” she says. The impacts are greater, and stretch further, than what’s estimated up front.
Many analysts expect the situation to worsen. A February 2024 study by Josh Lepawsky, a geography professor at Canada’s Memorial University of Newfoundland and Labrador, found that 40% of existing semiconductor facilities are already located in river basins with high or extremely high risk of water stress by 2030 to 2040.
The extreme desire by many developing world governments to attract foreign investment causes other problems, as officials tend to give priority to the demands of Big Tech over the daily, ongoing needs of the local population.
“When Hurricane Maria and Hurricane Irma devastated Puerto Rico, the data centers on the island did not go hungry for power or thirsty for water,” Steven Gonzalez, an expert on data centers, whose family comes from Puerto Rico, told Mongabay. “They had no downtime, even as citizens went without water and power for months.” (Google and its employees did donate $1.5 million to Hurricane Maria recovery efforts.)
There are signs resistance is growing. In Chile, a local community activist asked: “With Google as my neighbor, will there still be water?” In February 2024, a Chilean court listened, and it partially reversed a permit for a new data center, saying that Google must submit a new application taking proper account of climate change.
Outsourcing AI’s dirty work
Big Tech often locates in countries where workers are already poorly paid and regulations are lax. One reason for this arises from a little-known fact about AI: It requires an army of manual labor to produce these seemingly magical, smart, automated cyber products. AI text, images and videos need to be classified, categorized, cleaned and detoxified by humans, all according to U.S. cultural norms — including U.S. standards for hate speech, porn and extreme violence. And it is generally people from other nations who carry out this repetitive, often boring, and emotionally stressful work.
Julián Posada, assistant professor for the American Studies program at Yale University, told Mongabay that historically much of this work was done in India, Pakistan and Bangladesh. AI data annotation sweatshops, worker exploitation and abuse have also been reported on in the Philippines. More recently, Latin America has become a hotspot.
Oskarina Fuentes, an oil engineer by training, told El País that the economic crisis that erupted in 2016 in Venezuela forced her to leave the country and become an invisible AI worker in neighboring Colombia. Fuentes works for Appen, an Australian virtual platform that compiles data for Big Tech. Her role is “to tag data, to improve the performance of online bots,” for which she earns $200-$300 a month, close to the minimum wage.
Mongabay contacted Amazon because one of its subsidiaries, Amazon MTurk, was named in an article published by Reuters for employing Indians in “precarious,” “emotionally taxing” and “low wage work” related to AI. Amazon replied: “We don’t have anything further to add on the Reuters story.”
Wherever there is hyperinflation and an economic crisis but good Internet infrastructure, AI companies come in to offer data work, Posada said. Because of its dire economic situation, he says, Argentina is now following the same path as Venezuela and Colombia.
“It’s a very attractive model for Big Tech because you can pay people in cents of a dollar,” Posada states. Workers are isolated, nonunionized and under constant surveillance “so if their productivity is deemed inadequate by the algorithm, they can be summarily fired.”
However, the companies, he says, are often not so vigilant in ensuring that existing labor regulations are upheld. “There’s a lot of evidence of children doing this work,” Posada says. “One young person told me that his grandma gave her verification details to the system, so that he could start working.”
Much of the work is clearly not suitable for adults, let alone children. Posada showed Mongabay a platform called Hive, where there was a range of data jobs offered, some of them with a knife symbol indicating violent content. The knife symbol was added recently, he explained, as the only indication that children should not do this work. Not surprisingly, carrying out these tasks day after day can impact workers’ mental health.
Lehuedé told Mongabay that European and North American tech firms have excelled at “outsourcing damage.” This sort of demoralizing AI work has been shown to be sapping the mental health of workers throughout the world.
Buying its way out of trouble
Big Tech knows that AI consumption of energy and water is growing very rapidly and becoming a political and public concern. But Google told Mongabay that some of the most alarmist predictions may not be accurate.
“With AI at an inflection point, predicting the future growth of energy use and emissions from AI computing in our data centers is challenging. Historically, data center energy consumption has grown much more slowly than demand for computing power,” Google’s spokesperson said. “We have used tested practices to reduce the carbon footprint of workloads by large margins; together these principles can reduce the energy of training a model by up to 100x and emissions by up to 1,000x. We plan to continue applying these tested practices and to keep developing new ways to make AI computing more efficient.”
However, other AI industry leaders recognize that the current high-tech surge very likely isn’t sustainable, that the world cannot generate enough electricity to support the burgeoning industry, even as they avoid discussing industry environmental impacts.
OpenAI CEO Sam Altman warned at the World Economic Forum in January that the AI industry, and the world, is heading toward a catastrophic energy crisis. “There’s no way [forward] without a breakthrough,” he said. The solution, he added, is nuclear fusion.
Altman is currently in talks with investors, including the government of the United Arab Emirates, to secure funding for an audacious plan that could require raising $5 trillion to $7 trillion — which would be the largest single investment in world history — to fund a crash research program for a planetary energy transformation. (As insurance, Altman also admits he is prepping for disaster: “I have guns, gold, potassium iodide, antibiotics, batteries, water, gas masks from the Israeli Defense Force and a big patch of land in Big Sur I can fly to,” he says.)
But even if nuclear fusion on this vast scale proves viable, the results would only be felt by mid-century, say critics, and even then the solution would not solve the looming water crisis, or pollution crisis, but greatly add to it.
Altman’s idea is also simplistic, say analysts, because it doesn’t take into account the adverse impacts that such a gargantuan energy infrastructure project would have on the nine planetary boundaries, of which humanity has already dangerously transgressed six.
Earth system scientist Johan Rockström recently argued that what is urgently needed today is “a global sustainability transformation,” and that any large-scale human endeavor undertaken now must carefully consider its positive and negative impacts on all nine planetary boundaries, of which climate change is just one.
Without doing this encompassing environmental evaluation, Altman’s energy breakthrough might only speed up growth, creating nightmarish problems. It would, for example, be important before forging ahead on such a project to evaluate the pollution by rapidly intensified AI and data center e-waste (already the fastest growing and most toxic waste stream in the world), not to mention the risk posed by new nuclear fusion power plants.
Others are calling for a wholly different approach. They note that much of the data being stored at data centers is irrelevant trash, with 80% coming from videos, and that it’s time to clean up our vast electronic attic. These critics want much tighter quality controls and stricter safety standards for AI. They also want a far broader consultation process before the go-ahead is given to build new data centers.
Valdivia argues that “the supply chain capitalism of AI should be examined ‘from Below and on the Left,’” a slogan coined by Mexico’s revolutionary Zapatista movement circa 1985, which contends that economies need to be looked at from the perspective of the historically oppressed.
She says this “brings a radical and critical perspective, unveiling algorithmic harms related to extractivism, dispossession and capitalism’s destructive capacity. Given the current context of climate emergency, the AI community should embrace this perspective.”
Some Earth System scientists are thinking along the same lines, arguing that the nine environmental planetary boundaries for maintaining a “safe operating space for humanity,” need to be joined by a 10th social boundary evaluating human justice.
“We need to have a world-scale democratic conversation about the allocation of data centers, one that gives voice and rights to the great sweep of humanity and to the huge diversity of Nature,” Lehuedé said. “I feel that the core of the problem is the imposition of a technological ecological vision by the Global North on the Global South.”
This imposition isn’t new with AI; it has a 500-year history in Latin America. According to Valdivia, Spanish colonizers diverted water to their elite settlements and industries, leading “to a destabilization of natural water balances and the dispossession of Indigenous communities in Querétaro,” Mexico.
Latin American communities are growing increasingly wary of what they see as Big Tech’s imposed environmental and social injustices. Valdivia recalls how recently, as El Día de Los Muertos approached (the Mexican Day of the Dead), she encountered a protest in front of the Querétaro municipal council building.
Demonstrators carried posters reading: “It’s not drought but plunder,” and “Water is not a business,” and “Water for the people, not for the companies.” Maybe it’s time that Google and the rest listened, she said.
Banner image: Producing an AI chip requires 10-15 times more energy than making a standard chip. Image by analogicus via Pixabay (Public domain).
The Cloud vs. drought: Water hog data centers threaten Latin America, critics say
Citations:
Wu. C.-J., Raghavendra, R., Gupta, U., Acun, B., Ardalani, N., Maeng, K., … Hazelwood, K. (2022). Sustainable AI: Environmental implications, challenges and opportunities. arXiv. doi:10.48550/arXiv.2111.00364
Lin, A. Y., Panchangam, S. C., & Lo, C. (2009). The impact of semiconductor, electronics and optoelectronic industries on downstream perfluorinated chemical contamination in Taiwanese rivers. Environmental Pollution, 157(4), 1365-1372. doi:10.1016/j.envpol.2008.11.033
Lepawsky, J. (2024). Climate change induced water stress and future semiconductor supply chain risk. iScience, 27(2), 108791. doi:10.1016/j.isci.2024.108791
Crawford, K. (2024). Generative AI’s environmental costs are soaring — and mostly secret. Nature, 626(8000), 693-693. doi:10.1038/d41586-024-00478-x
Martel, A. J. (2018). Science from below and to the left: A processualist understanding of the Zapatista movement’s counter-hegemonic practices (Master’s thesis, Norwegian University of Life Sciences, Ås, Norway). Retrieved from http://hdl.handle.net/11250/2584191
Rockström, J., Gupta, J., Qin, D., Lade, S. J., Abrams, J. F., Andersen, L. S., … Zhang, X. (2023). Safe and just Earth system boundaries. Nature, 619(7968), 102-111. doi:10.1038/s41586-023-06083-8
FEEDBACK: Use this form to send a message to the author of this post. If you want to post a public comment, you can do that at the bottom of the page.