AI is becoming a disaster for the climate
AI is becoming a disaster for the climate Two new studies reveal the exploding carbon emissions of the AI industry, while Silicon Valley pushes to kill communities’ means of protesting data centers and the impacts on human health mount. Anthropic’s initial public offering netted the company over $81 billion on the first day of trading, making it one of, if not the, largest IPOs in history. The leading AI company, just five years old, now officially boasts a net worth in the neighborhood of $1.3-$1.7 trillion, depending on how you do your accounting. I think this is a good time, as such, to point out that, per a recent report, Anthropic also “has one of the worst recent records on fossil fuels, with analysis showing its upcoming data center projects could emit nearly 20 megatons of CO2-e a year.” Despite effectively positioning itself as a uniquely ethical AI company, Anthropic has been doubling down on fossil fuels to power its technology, both by turning to particularly polluting data centers, like xAI’s infamous Colossus plant in Memphis, for its compute needs, and embracing natural gas to power its own rapidly expanding array of in-house data centers. Anthropic, of course, is not alone. Despite all the whipsawing AI news of late—Jensen Huang’s ouster at Nvidia and failed comeback bid, the digital infrastructurapocalpyse of the last week, and the IPO, for starters—I was glad to see another study, on the climate impacts of the AI buildout, get some serious headlines. Even if those headlines were all of the ‘rather alarming’ variety, like the New York Times’, which went “Carbon Emissions of Biggest U.S. Data Centers Already Rival Entire Countries, Study Reveals.” I was less glad to see BAAJA, a permitting reform bill designed to overturn key environmental protections in the name of making it easier for developers to build energy infrastructure and data centers, pass the Senate. We’ll get into all this, and a bit more, including a cameo from a very special guest who might help prove that data center skepticism runs in the family, in today’s edition. As always, this work is made possible by my magnificent paid subscribers—thank you, thank you—who chip in a few bucks a month so I can keep the lights on. If you’re able, and you find value in this work, please consider upgrading, too. Okay! Onwards. First, let’s get to that Harvard University study that made waves right before the Anthropic IPO. The report examined the CO2 emissions of over 400 of the United States’ largest data centers. It found that more than half of the power used to run them, 54%, was generated by fossil fuels, and thus, combined, they emitted as much carbon pollution as a mid-sized nation. Here’s the New York Times: The largest data centers in the United States are responsible for an estimated 25.7 million metric tons of carbon dioxide emissions annually — almost as much as entire countries like Bolivia or Denmark — according to new research. The study, published Monday in the journal Nature Sustainability, delivers the first peer-reviewed estimates of carbon dioxide emissions from the country’s biggest data centers since the start of the artificial intelligence boom. Its authors, a team led by researchers from Harvard University, examined the electricity consumption and carbon emissions of 403 data centers from May 2024 to April 2025. All told, those data centers used almost 2 percent of U.S. electricity, more than New York City and Los Angeles combined, and represented about 0.5 percent of U.S. carbon dioxide emissions. The largest data centers in the United States are responsible for an estimated 25.7 million metric tons of carbon dioxide emissions annually — almost as much as entire countries like Bolivia or Denmark — according to new research. The study, published Monday in the journal Nature Sustainability, delivers the first peer-reviewed estimates of carbon dioxide emissions from the country’s biggest data centers since the start of the artificial intelligence boom. Its authors, a team led by researchers from Harvard University, examined the electricity consumption and carbon emissions of 403 data centers from May 2024 to April 2025. All told, those data centers used almost 2 percent of U.S. electricity, more than New York City and Los Angeles combined, and represented about 0.5 percent of U.S. carbon dioxide emissions. Now, if you think that finding sounds bad, then I hate to break it to you, but the actual emissions picture is probably much, much worse. For one thing, the survey ended in April of last year, and data center development has continued apace since then, with natural gas, and to a somewhat lesser extent, coal, supplying much of the new power demand. Furthermore, as Emily Atkin writes at her great climate newsletter HEATED, the study, for a number of key reasons, “gets nowhere near the actual amount of climate pollution” data centers actually emit. From Atkin’s piece: One [issue] is the way the researchers counted the facilities’ emissions. To do this, they estimated how much electricity each data center used, then looked at how the grid in that part of the country makes its electricity—how much comes from gas, coal, renewables, and so on—to estimate how much CO₂ would be released to generate that much power. But not all data center electricity comes from the grid. Indeed, an ever-increasing number of these massive data centers are also building their own, on-site fossil fuel plants. For example, during the study period (May 2024 through April 2025), Elon Musk’s xAI got some of the electricity for its Colossus data center in Memphis from the grid—but it also got some of it by burning gas in turbines right outside the building. Lead author Gianluca Guidi confirmed that Colossus wasn’t among the 403 facilities studied. One [issue] is the way the researchers counted the facilities’ emissions. To do this, they estimated how much electricity each data center used, then looked at how the grid in that part of the country makes its electricity—how much comes from gas, coal, renewables, and so on—to estimate how much CO₂ would be released to generate that much power. But not all data center electricity comes from the grid. Indeed, an ever-increasing number of these massive data centers are also building their own, on-site fossil fuel plants. For example, during the study period (May 2024 through April 2025), Elon Musk’s xAI got some of the electricity for its Colossus data center in Memphis from the grid—but it also got some of it by burning gas in turbines right outside the building. Lead author Gianluca Guidi confirmed that Colossus wasn’t among the 403 facilities studied. TLDR, the study doesn’t count the growing number of data centers, like Colossus, that use gas turbines on-site to power their plants. Furthermore, “electricity consumption is not the only way these facilities heat the planet. For example, the equipment that keeps the servers from overheating is leaking heat-trapping refrigerants into the atmosphere. And when a data center runs on gas—as so many of them do—methane can escape from wells and pipelines before the gas even reaches the power plant. “electricity consumption is not the only way these facilities heat the planet. For example, the equipment that keeps the servers from overheating is leaking heat-trapping refrigerants into the atmosphere. And when a data center runs on gas—as so many of them do—methane can escape from wells and pipelines before the gas even reaches the power plant. Finally, due to their tracking methodology, the authors may have undercounted the number of data centers in operation at the time the study was carried out by a significant margin. As many as 250 data centers may be absent from the count, as of 2025, Atkin argues. All of which is to say, the study should be viewed as the floor; the minimum for how much greenhouse gas pollution data centers are emitting. And that is more than the entire country of Kenya. The findings of the Harvard study comport rather disconcertingly with those of another recent report, State of AI Climate Impact 2026, by the Australian climate researcher Ketan Joshi. Joshi’s report concludes that “The AI boom is already worsening greenhouse gas emissions,” and even more concerningly, that it is “potentially emitting more across its lifecycle than all international aviation.” And, surprise, it’s only slated to get worse in coming years if current development patterns hold. This is honestly pretty wild to contemplate here. The AI industry is on track, by some counts—including McKinsey’s—to generate more emissions than the coal mining industry, or global shipping and aviation. These are two of the most polluting heavy industries in the world, and AI is poised to blow past them. AI is already emitting twice the greenhouse gases of the global aluminum industry, itself a pretty crazy statistic given that aluminum production is a notoriously polluting endeavor. As the report states, “an unchecked continuation of these trends pushes the AI industry into becoming a major global heavy-emitting industry dwarfing both international aviation and international shipping.” Leading the way is the week’s big IPO winner, Anthropic. As I noted up top, Anthropic rose to prominence by burnishing a reputation as the most ethical purveyor of AI. This notably does not extend to environmental or climate concerns, arenas in which the leading AI firm is notably worse than almost all of its competition. Giants like Amazon, Microsoft, and Google may be reneging on their climate pledges, becoming less transparent in their emissions reporting, and diminishing their efforts to power servers and operations with clean energy. Meta, for instance, one of the nation’s largest data center operators, released its latest sustainability report this week, revealing that its emissions rose 45% since last year. The company said it was rethinking its erstwhile climate goals of producing net zero carbon emissions by 2030 as a result. Yet Anthropic hasn’t even bothered to release any information about its carbon emissions or environmental impacts at all. Here’s a rundown from a special section of Joshi’s report focusing just on Anthropic and its posture towards energy and climate issues. It’s pretty damning. I’ll quote from it at length: Anthropic has never disclosed any energy, emissions or environmental data of any kind. Anthropic has never disclosed any energy, emissions or environmental data of any kind. In July 2025, Anthropic suggested the US government explicitly set a target for the construction of fossil gas power, promising to begin “preconstruction design work for natural gas plants” by the end of 2026. In July 2025, Anthropic suggested the US government explicitly set a target for the construction of fossil gas power, promising to begin “preconstruction design work for natural gas plants” by the end of 2026. In May 2026, it was revealed that Anthropic is leasing capacity from the notoriously polluting SpaceX Memphis facility, itself a major source of local air pollution and climate pollution, and a project that has been strongly contested by affected locals; one analyst estimates (based on payments to SpaceX) that 1 in 2 Anthropic queries is served by this fossil-fueled facility (one of the worst in the world). In May 2026, it was revealed that Anthropic is leasing capacity from the notoriously polluting SpaceX Memphis facility, itself a major source of local air pollution and climate pollution, and a project that has been strongly contested by affected locals; one analyst estimates (based on payments to SpaceX) that 1 in 2 Anthropic queries is served by this fossil-fueled facility (one of the worst in the world). In a recently released set of “data centre community commitments”, Anthropic promises much of the new power generation to serve their data centres will be “natural gas, which can be brought online quickly, ramps up and down as demand changes, and is abundant in the places we’re considering for data centers”. They claim to be considering renewable energy but do not provide any details, and promise to invest in carbon removal to “help address emissions”. In a recently released set of “data centre community commitments”, Anthropic promises much of the new power generation to serve their data centres will be “natural gas, which can be brought online quickly, ramps up and down as demand changes, and is abundant in the places we’re considering for data centers”. They claim to be considering renewable energy but do not provide any details, and promise to invest in carbon removal to “help address emissions”. In the S1 form filed by Anthropic, in advance of their Initial Public Offering (IPO), Anthropic suggests long term high emissions by declaring several fossil-fuel related risks, such as exposure to fossil fuel price volatility or competition with other customers for fossil gas. They also dismiss renewables as too high in cost (while ironically stating that “physical effects of climate change may increase our operating costs over time”). In the S1 form filed by Anthropic, in advance of their Initial Public Offering (IPO), Anthropic suggests long term high emissions by declaring several fossil-fuel related risks, such as exposure to fossil fuel price volatility or competition with other customers for fossil gas. They also dismiss renewables as too high in cost (while ironically stating that “physical effects of climate change may increase our operating costs over time”). In November 2026, Anthropic struck a USD $3.4bn deal with OnePWR, a fossil fuelled turbine manufacturer, for 1.7 gigawatts of generators, with the Financial Times stating that the transaction “is the largest known single purchase of power-generation equipment by an AI lab”. This transaction is more than three times the amount spent by xAI on its notoriously polluting Memphis turbines, and would emit approximately 6 MTCO2-e per year, equivalent to adding 1.8 million fossil-fuelled cars to the road. In November 2026, Anthropic struck a USD $3.4bn deal with OnePWR, a fossil fuelled turbine manufacturer, for 1.7 gigawatts of generators, with the Financial Times stating that the transaction “is the largest known single purchase of power-generation equipment by an AI lab”. This transaction is more than three times the amount spent by xAI on its notoriously polluting Memphis turbines, and would emit approximately 6 MTCO2-e per year, equivalent to adding 1.8 million fossil-fuelled cars to the road. In November 2026, Anthropic posted a job listing paying up to $405,000 USD per year for “a Gas Supply & Transport Lead to own natural gas supply for Anthropic’s non-cloud data center portfolio in the United States and Canada”, adding that “if gas does not arrive, in the volume and at the pressure the plant needs, the compute does not run”. In November 2026, Anthropic posted a job listing paying up to $405,000 USD per year for “a Gas Supply & Transport Lead to own natural gas supply for Anthropic’s non-cloud data center portfolio in the United States and Canada”, adding that “if gas does not arrive, in the volume and at the pressure the plant needs, the compute does not run”. This is a clear portrait of a company that is not concerned with its own ballooning carbon emissions and climate impacts, until, that is, they threaten to hamper its data centers’ capacity to run smoothly. Image by Janusz Walczak via Unsplash The single biggest threat to AI data centers right now, however, is probably that everyone hates them. They hate them so much that public citizens across the United States have mobilized to shut them down, and elected politicians who’ve promised to help them do so. So how has the AI industry responded? By directing its trade groups and lobby to partner with the fossil fuel industry to push the Bipartisan American Affordability and Jobs Act, or BAAJA, through Congress before the data center-skeptical class of 2027 Democrats arrive. The tech industry is not just buying more fossil fuel power, it’s proactively working to enact policies that make it cheaper, easier and less democratic a process to build more fossil fuel power plants. BAAJA is an assemblage of provisions the fossil fuels industry has been after for years: rollbacks on local governments’ legal tools to curtail development projects, curbs on endangered species protections, and a defanging of the Clean Water Act. Here’s Aaron Rugenberg, an environmental lawyer and the director of Public Citizen’s Climate Accountability Project, explaining in the New Republic: BAAJA severely weakens the Clean Water Act, robbing states and tribes of their ability to protect water quality and require adequate safeguards for activities that pollute or destroy water resources. It practically nullifies the Endangered Species Act by delegating its core responsibilities to the states, most of which either oppose the act’s wildlife protections or lack the expertise and resources to adequately fulfill these obligations. And it essentially repeals the National Environmental Policy Act, the bedrock environmental law that gives the public a voice in decisions that impact the environment, by prohibiting the enforcement of NEPA violations. These three legal frameworks are responsible for most the improvements in water quality, green space, and wildlife recovery in our country over the last half-century. And that’s just the beginning. Other provisions of BAAJA go to even greater lengths to radically lock in fossil fuel development far into the future. The “disparate treatment” section of the bill effectively makes the permitting decisions of the Trump administration permanent. In the words of David Bookbinder, an environmental lawyer who has worked for both the left-leaning Sierra Club and the libertarian Niskanen Center, “The result of making these energy permitting policies permanent is that future administrations will be unable to effectively reduce GHG emissions.” BAAJA also guarantees not just the construction of fossil fuel projects but also their “operation at full capacity,” permanently blocking any future administration from regulating energy infrastructure. BAAJA severely weakens the Clean Water Act, robbing states and tribes of their ability to protect water quality and require adequate safeguards for activities that pollute or destroy water resources. It practically nullifies the Endangered Species Act by delegating its core responsibilities to the states, most of which either oppose the act’s wildlife protections or lack the expertise and resources to adequately fulfill these obligations. And it essentially repeals the National Environmental Policy Act, the bedrock environmental law that gives the public a voice in decisions that impact the environment, by prohibiting the enforcement of NEPA violations. These three legal frameworks are responsible for most the improvements in water quality, green space, and wildlife recovery in our country over the last half-century. And that’s just the beginning. Other provisions of BAAJA go to even greater lengths to radically lock in fossil fuel development far into the future. The “disparate treatment” section of the bill effectively makes the permitting decisions of the Trump administration permanent. In the words of David Bookbinder, an environmental lawyer who has worked for both the left-leaning Sierra Club and the libertarian Niskanen Center, “The result of making these energy permitting policies permanent is that future administrations will be unable to effectively reduce GHG emissions.” BAAJA also guarantees not just the construction of fossil fuel projects but also their “operation at full capacity,” permanently blocking any future administration from regulating energy infrastructure. And it’s a similar case for data center projects. My friends at the AI Now Institute put out a statement declaring that a vote for BAAJA is “a vote for a ‘backdoor bailout’ of the AI industry.” Why? Two reasons: BAAJA preempts the few levers that local, state, and Tribal authorities have to hold Big Tech accountable. BAAJA preempts the few levers that local, state, and Tribal authorities have to hold Big Tech accountable. BAAJA supercharges AI data center expansion in exchange for token ratepayer protections. BAAJA supercharges AI data center expansion in exchange for token ratepayer protections. Coming as it is off the heels of the blue wave, anti-AI election, BAAJA feels like a slap in the face to voters. Or as Naomi Klein put it, “The voters just rejected data centers. Why is the Senate trying to fast-track them?” It’s crucial to note that this is not just a Republican/big tech/oil industry-led effort. A bunch of centrist and ‘abundance’ Democrats have joined the GOP in supporting the effort, under the auspices that it will cut red tape for clean energy projects, too. That’s true, to an extent, but this a terrible deal, and an absolutely unnecessary tradeoff at a point when Republicans’ stock could not be lower. After all, a) it’s just as important that we leave fossil fuels in the ground at this late stage of the climate crisis as it is that we build out more clean energy, and b) there are tools to effect desired outcomes without stripping local governments of the power to protect their communities and ecosystems. Regardless, fast-track it the Senate did, with the help of more than a dozen Democratic Senators, including current Majority Leader Chuck Schumer. It now heads to the House of Representatives. I expect this to accelerate calls for a change in Senate leadership. To recap: As of spring in 2025, American data centers were emitting as much greenhouse gas pollution as nations like Kenya, Bolivia, and Denmark. They are almost certainly emitting much more than that today. Worldwide, the AI industry is on a trajectory to outpace the emissions of the heaviest and most-polluting industries, like aviation and coal mining, within the next ten years. And tech firms are fighting to make it easier and faster to build more fossil fuel projects and data centers. In other words, the AI industry is quickly becoming not just a concern, but something more akin to a disaster for the climate. It’s no longer merely prolonging the lifespan of fossil fueled power plants, but actively working to bring more of them online to run their data centers, at a scale that was mostly unthinkable even three years ago. “There’s a kind of dirty ‘energy desperation’ within the industry,” the climate researcher and study author Ketan Joshi tells me. “If you give them 100 wind farms, they’ll take those but still build 5 city-sized gas plants, because the brute force software system they provide has no upper limit on energy consumption.” ”What this means is that regulation that aims to treat the symptoms (eg: 100% renewable mandates, or deregulating permitting to allow for more transmission) just gets added on top of AI consumption,” he adds. “Like fossil fuels in general, we need to limit the source.” There’s one more element to all of this that needs to be underlined, and that’s the fact that the AI industry’s data center and fossil fuel infrastructure buildout won’t just contribute to carbon emissions, but air pollution, too. There’s a real cost to respiratory health, and not just to those who live close to next door to polluting mega-projects like xAI’s Colossus. Dr. Robert Merchant is a pulmonologist in Montana, where a massive 8 square mile data center project has begun construction, built by Quantica Infrastructure, outside of Billings. The project involves bringing online gas-fired power plants to provide 1,800 megawatts of capacity—more capacity than all other power plants currently operating in Montana combined. Merchant submitted a study of the respiratory health impacts a power plant of that size to Yellowstone County, and spoke at a public meeting with the commissioners. Merchant, who I might add is my uncle, and thus perhaps is evidence of a familial predisposition to taking data center interests to task, noted how this level of gas-fired power generation will loose particulate pollution across the state and far beyond. Based on his modeling, the data center project would increase asthma rates, illness rates, as well as the number of heart attacks and strokes. It would cost between $20-50 million in medical costs, dampen productivity in the region as workers take more sick days, and cause 1-2 deaths per year. His report and presentation made news in the local media. Here’s the Montana Free Press: Robert Merchant, a semi-retired Billings Clinic doctor, presented his findings at the commissioners’ Thursday discussion meeting… Fine particles containing nitrogen oxides, sulfur dioxide, methane and other gases can contribute to asthma, heart attacks, cancer, low birthweight, cognitive functioning and other health problems, he wrote. Merchant said the environmental impacts from the data center, as currently planned, would spill well beyond Yellowstone County, stretching as far south as northern Texas… “Inaction will result in health and economic damage to the public.” Merchant thanked the commissioners for their work and said he hoped the data he shared was useful to their analysis. Merchant’s presentation was met with applause. Robert Merchant, a semi-retired Billings Clinic doctor, presented his findings at the commissioners’ Thursday discussion meeting… Fine particles containing nitrogen oxides, sulfur dioxide, methane and other gases can contribute to asthma, heart attacks, cancer, low birthweight, cognitive functioning and other health problems, he wrote. Merchant said the environmental impacts from the data center, as currently planned, would spill well beyond Yellowstone County, stretching as far south as northern Texas… “Inaction will result in health and economic damage to the public.” Merchant thanked the commissioners for their work and said he hoped the data he shared was useful to their analysis. Merchant’s presentation was met with applause. Get ‘em, uncle Bob. I used my inside connections to reach out to Merchant and ask if there was anything about his report or the situation in Montana he wanted to share with readers. He responded in kind: What frustrates me to no end is that this type of data isn’t part of every single proposal for a new data center. What a complete lack of transparency. Even if they don’t do the modeling of health effects they should be required to state their expected emissions. I had to reach out to acquaintances at an excellent organization, Montana Environmental Information Center, to get the needed estimates. They gathered it up for public, but not easily found, information. A reporter from our local paper had also sourced some of the information. It wasn’t that hard for me to do the analysis and report but why should the region be relying on a retired pulmonologist to figure out the health impacts of this huge new data center? Where is the EPA? Where is the Montana Department of Environmental Quality? Where is the Yellowstone County Public Health Department? Or even more idealistically, shouldn’t the company itself care about the region and do the analysis themselves? And share it with the community? What frustrates me to no end is that this type of data isn’t part of every single proposal for a new data center. What a complete lack of transparency. Even if they don’t do the modeling of health effects they should be required to state their expected emissions. I had to reach out to acquaintances at an excellent organization, Montana Environmental Information Center, to get the needed estimates. They gathered it up for public, but not easily found, information. A reporter from our local paper had also sourced some of the information. It wasn’t that hard for me to do the analysis and report but why should the region be relying on a retired pulmonologist to figure out the health impacts of this huge new data center? Where is the EPA? Where is the Montana Department of Environmental Quality? Where is the Yellowstone County Public Health Department? Or even more idealistically, shouldn’t the company itself care about the region and do the analysis themselves? And share it with the community? I could not agree more. Speaking of, this kind of transparency would be some very low-hanging fruit that the new wave of data center-critical politicians could demand when they take office. The AI industry and its data center buildout are on track to become a genuine crisis factor for the climate, especially given its limitless appetite for expansion. And it’s funny, you never hear the AI execs promise that their products are going to “solve” climate change anymore. Now they stick to saying it will cure cancer. Saying AI is good for climate now apparently creates too much cognitive dissonance for the tech elite to stomach, and that’s saying something. Some fun media stuff this week: I joined KQED’s Forum with Washington Post tech reporter Shira Ovide and former Biden AI policy advisor Ben Buchanan. We talked Nvidia, Anthropic, the global internet outages, and more. It’s always a good time joining Alexis’s show. I was interviewed for the BBC’s Global Story, for an episode about AI and Luddites, which was particularly cool. It’s n+1 Bookmatch season, which longtime readers know is one of my favorite holiday traditions. If you haven’t taken it yet, get to it; donate to a great literary mag and find some stellar reads in the process. I got at least one home run already, which I somehow missed until now: Matthew Wolfe’s Fires in the Night, a history of the Earth Liberation Front, a group with some strong Luddite vibes indeed. Finally, if you didn’t catch it, for the show this week I sat down with Cory Doctorow to talk about how AI is breaking the internet. This is already one of our biggest shows on YouTube, for perhaps the obvious reasons. No one does this better than Cory. Alright, that’s it for now. As always, thanks for reading. And keep those hammers up.