Hours after his inauguration, Donald Trump declared a “national energy emergency.” The US, he warned in an govt order, suffered from “a precariously inadequate and intermittent energy supply, and an increasingly unreliable grid.” If it did not broaden its vitality infrastructure and enhance provide, Trump warned, “this situation will dramatically deteriorate in the near future due to a high demand for energy and natural resources to power the next generation of technology.”
Trump was characteristically fast guilty excessive vitality costs on “the policies of the previous administration” and “our Nation’s diminished capacity to insulate itself from hostile foreign actors.” The nation was not but going through something that could possibly be referred to as an vitality emergency (though Trump’s personal insurance policies would possibly nicely result in one). However there have been certainly steep challenges going through the facility sector. After remaining largely flat for the final twenty years, demand for electrical energy within the US has begun to surge. In 2024 electrical utilities predicted they would wish virtually twice as a lot further energy by 2028 as they thought they’d in 2023. Main energy firms in some components of the nation are already struggling to keep away from blackouts, particularly throughout excessive climate occasions. The demand is simply going to extend: in spite of everything, as most of the world’s nations agreed on the COP28 summit, we have to roughly triple our renewable capability to maintain the Paris Settlement goal of 1.5 levels of warming inside attain, not simply producing electrical energy from sources apart from soiled fossil fuels but in addition utilizing clear vitality to energy all transit and to warmth and funky all buildings.
This surging demand has a spread of sources, notably the unfold of electrical autos and the electrification of buildings and business. However a major contribution to the pressure on the grid comes from the energy-hungry knowledge facilities that energy synthetic intelligence. Meta, as an illustration, is at present creating a facility in Louisiana that Mark Zuckerberg has promised to broaden into a knowledge heart “supercluster” that may use virtually twice as a lot vitality as all the metropolis of New Orleans. In the meantime, knowledge facilities in Virginia—residence to Knowledge Middle Alley, which has one of many densest concentrations of such amenities on the earth—devour greater than 1 / 4 of the electrical energy generated within the state. Researchers estimate that the diesel-powered turbines used as backups there might already be inflicting 14,000 instances of bronchial asthma signs and imposing public well being prices of $220 to $300 million per yr. And in Memphis, Elon Musk’s synthetic intelligence firm, xAI, is powered by thirty-five methane fuel generators that belch smog-forming air pollution.
Lately AI’s environmental impacts have come beneath shut scrutiny. Public studies concerning the monumental portions of vitality it makes use of and the prodigious quantity of water needed to chill the gear in its knowledge facilities began appearing after the launch of the ChatGPT chatbot in 2022, and state-level efforts to mandate reporting of AI’s environmental influence adopted shortly thereafter. Up to now they haven’t managed to indicate many outcomes. The Biden administration issued an govt order on the “Safe, Secure and Trustworthy Development and Use of Artificial Intelligence” in 2023, however it prevented any remark concerning the know-how’s voracious vitality use, focusing as a substitute on selling competitors inside the AI business and stopping AI-enabled threats to civil liberties and nationwide safety. Early final yr Senator Edward Markey and others launched a invoice that required the federal authorities to evaluate AI’s environmental footprint and create a system via which firms might volunteer to report future impacts. It died in Congress and not using a vote, though the European Union did undertake a stronger model of the measure, which went into impact in August 2024.
On the primary day of his second time period Trump jettisoned Biden’s modest govt order. The subsequent day he introduced Undertaking Stargate, an initiative via which a number of main know-how giants will make investments as a lot as $500 billion—greater than the price range for the Apollo house program—in synthetic intelligence and knowledge facilities throughout the US over the subsequent 4 years. On Tuesday OpenAI introduced that it will accomplice with the corporations SoftBank and Oracle—with which it’s already constructing a large knowledge heart in Texas—to assemble 5 new facilities in Texas, Ohio, New Mexico, and an as-yet-unannounced state within the Midwest.
Some analysts have predicted that synthetic intelligence will assist cut back humanity’s environmental footprint by permitting customers to make energy-saving decisions and by optimizing the applied sciences on which a lot of society now relies upon. Many main tech firms have net-zero targets, and a few have made modest gestures towards utilizing AI for the good thing about the planet: in 2017 Microsoft launched a $50-million initiative referred to as “AI for Earth” that guarantees to help researchers and environmental teams with grants for Microsoft AI instruments.
Now that they seem prone to fall in need of these net-zero targets, given their present degree of emissions and growing energy wants, a number of of those corporations have begun an effort to resurrect the US nuclear business as a carbon-neutral supply of vitality for AI. In October 2024 Google signed a contract with Kairos Energy, a “next-generation” nuclear firm that has promised to construct its first small modular reactor by 2030; Microsoft has signed a twenty-year energy buy settlement with Constellation Vitality, which can reopen the positioning of the US’s worst nuclear catastrophe; Amazon is investing straight within the Maryland-based firm X-energy, which can also be promising to ship small modular reactors. The White Home has been a powerful supporter of nuclear-powered AI. This Could, President Trump signed 4 govt orders aimed toward accelerating the development of nuclear energy within the US.
The issue with these guarantees, in accordance with M. V. Ramana, writer of Nuclear is Not the Answer, is that the reactor designs utilized by Kairos and X-energy draw on fashions from the mid-twentieth century which have confronted failures, unplanned shutdowns, and elementary technological hurdles when deployed up to now. Nuclear energy solely generated 9.1 p.c of world electrical energy in 2023, down by practically half from its peak at 17.5 p.c in 1996. America has seen solely three nuclear reactors come on-line since 1996. Critics reminiscent of Ramana level to this historical past to recommend that nuclear energy is each gradual to assemble and costly, significantly when in comparison with photo voltaic and wind energy. Trump’s govt orders point out that he believes a fast scale-up of the nuclear business will solely be achieved by expediting and deregulating processes within the Nuclear Regulatory Fee and decreasing the company’s independence. This underlines a 3rd concern concerning the business: the hazard of a catastrophic system failure.
If these guarantees of a brand new nuclear age do turn into little greater than greenwashing, and if the Trump administration’s efforts to gradual the buildout of renewable vitality succeed even partly, then AI will probably find yourself getting a major share of its energy from extra fossil fuels. Certainly, the guarantees of future AI progress are already spurring plans for the development of extra fossil-fuel crops—even within the absence of exhausting proof that essentially the most extravagant of the forecasts will come true. A new report commissioned by the Southern Environmental Legislation Middle (SELC) means that it is going to be virtually not possible to construct the variety of knowledge facilities anticipated by vitality demand projections, as a result of doing so would require greater than 90 p.c of the worldwide chip provide over the subsequent 5 years. That is hardly shocking: demand tends to be overstated as a result of firms put in requests to construct extra initiatives than they intend to finish, understanding many received’t be permitted. And but these forecasts, significantly from the Southeastern US, “are driving a dramatic and unnecessary overbuild of infrastructure,” as SELC Senior Lawyer Megan Gibson has argued; that overbuild, in flip, “threatens to lock in fossil fuels, hike energy bills, and crowd out more reliable, cost-effective clean energy.” Daniel Brookshire, an analyst with the group, studies that monopoly utilities within the South are planning to quickly construct 43,000 megawatts of methane fuel crops inside the subsequent fifteen years—the equal, he notes, of forty nuclear energy crops.
It’s not straightforward to know how a lot energy AI actually consumes. Estimates of the variety of knowledge facilities worldwide run from 9,000 to 11,000, however many further ones are at present being constructed. The Worldwide Vitality Company projected final yr that knowledge facilities’ annual electrical energy consumption might double between 2022 and 2026, to 1,000 terawatt-hours–roughly equal to that of the whole lot of Japan. However that estimate contains not simply AI however all knowledge heart actions, from storing folks’s emails to serving up Netflix movies and “mining” Bitcoins.
One option to get a tough sense of the impact of the AI growth is to have a look at the historical past of knowledge heart vitality consumption in recent times. Within the US, the quantity of electrical energy going to knowledge facilities remained fairly flat from 2005 to 2017, regardless of the expansion of cloud-based on-line companies reminiscent of Fb and Netflix, in addition to the cryptocurrency market in these years. (Cryptocurrency consumes a substantial amount of vitality: crypto mining represents between 0.6 and a couple of.3 p.c of all US electrical energy demand, the Vitality Info Administration reported final yr. However miners can work when vitality costs are low, which works a way towards diminishing their influence on the grid.) In 2017 energy-intensive {hardware} designed for AI started to be put in in knowledge facilities, and by 2023 their electrical energy consumption had doubled. The newest studies present that knowledge facilities devour 4.4 p.c of all of the electrical energy within the US. Latest projections from the Lawrence Berkeley Nationwide Laboratory estimate that greater than half of the electrical energy going to knowledge facilities shall be used for AI by 2028, at which level AI would possibly use as a lot electrical energy per yr as 22 p.c of all US households. However such projections are subtle guesswork.
Exploring what precisely synthetic intelligence is and the way it works sheds some mild on why it’s exhausting to trace how a lot vitality it makes use of to do its computational work. Earlier than you ask an AI mannequin to draft a analysis paper (an more and more frequent phenomenon that has led some to conclude that the school essay is lifeless), write you a recipe, or generate a cool video, it must be educated in a knowledge heart. These hangar-like amenities, owned and operated by Huge Tech firms but in addition by much less well-known builders like Tract, Swap, EdgeCore, Novva, Vantage, and PowerHouse, are full of servers containing particular chips referred to as graphics processing models (GPUs) and the processors (CPUs) that serve them data. Because the journalists James O’Donnell and Casey Crownhart describe in a main current report for the MIT Expertise Overview, one AI mannequin could also be housed on over a dozen GPUs, and a big knowledge heart is prone to include over 10,000 wired-together chips, in addition to the followers and pumped water that siphon off all the warmth generated by these processors.
To say that an AI mannequin should be “trained” primarily signifies that it should digest an enormous quantity of knowledge and carry out an enormous variety of apply computations. That is an costly, energy-intensive course of: coaching OpenAI’s GPT-4, in accordance with O’Donnell and Crownhart, value over $100 million and required fifty gigawatt-hours of electrical energy, which is how a lot San Francisco makes use of in three days. Solely after months of such coaching can folks “inference” the AI fashions to get solutions to their prompts, which requires a special amount of vitality, partly because the course of is repeated many occasions—for instance, in 100 million self-driving vehicles. O’Donnell and Crownhart notice that 80 to 90 p.c of whole AI computing energy is now estimated to go to inference, moderately than coaching.
As O’Donnell and Crownhart level out, the electrical energy used (and carbon emissions generated) by particular person AI queries varies significantly based mostly on the character of the question, the sort and measurement of the AI mannequin that solutions it, the time of day it’s made, and the vitality mixture of the electrical grid that powers the information heart to which it’s despatched. O’Donnell and Crownhart labored with a crew on the College of Michigan to generate examples derived from Meta’s Llama, a preferred open-source AI mannequin. They calculated that small fashions use a lot much less vitality than massive ones, which have a tendency to offer higher solutions however should run on extra chips: the typical textual content question on a small mannequin consumed solely the quantity of vitality it takes to run a microwave for one-tenth of a second, whereas a bigger mannequin required the equal of eight seconds. It additionally takes extra vitality for a mannequin to reply to a posh immediate—for instance, to put in writing a artistic story—than to a easy one. Queries to generate photographs, which use completely different sorts of fashions, moderately surprisingly use much less vitality than queries to massive text-generating fashions. However making a video makes use of way more: Sasha Luccioni, an AI and local weather researcher, calculated that producing a five-second video utilizing CogVideoX, an open-source mannequin, used the identical quantity of vitality as working a microwave for over an hour. If thousands and thousands of individuals begin making their very own movies, AI’s urge for food for vitality will grow to be actually gargantuan.
But as helpful as these calculations are, we can’t make sure that they precisely gauge the vitality consumption of many frequent AI fashions. It is because most Huge Tech firms use closed-source AI—that means that the code behind them isn’t obtainable to the general public—and don’t launch enough details about their vitality and useful resource use. The Trump administration’s refusal to mandate reporting of AI’s environmental influence signifies that we now have scant thought about what AI is basically doing to the planet. As Luccioni places it, “We should stop trying to reverse-engineer numbers based on hearsay and put more pressure on these companies to actually share the real ones.” Luccioni has created the AI Vitality Rating, which charges AI fashions on their vitality effectivity, however Huge Tech firms should decide in. So far, removed from sufficient have executed so.
Providing an vital glimpse behind this wall of company secrecy, final month Google launched a report that quantifies how a lot vitality its Gemini AI makes use of with each immediate. Though the amount per median textual content question (0.24 watt-hours, sufficient to run a microwave for a second) could appear small, multiply that quantity by what might nicely be a mammoth variety of queries made to Gemini day by day and a worrying image of AI’s vitality and water consumption begins to emerge. The report additionally bears out, as O’Donnell and Crownhart’s analysis urged, that advanced prompts use way more vitality. It’s additionally value noting that the Google report solely discusses textual content prompts, leaving out the a lot increased vitality prices of video era. Past that, Google nonetheless hasn’t revealed what number of Gemini queries it processes day by day, that means that it’s nonetheless not possible to calculate the AI’s whole vitality footprint.
One more reason that it’s exhausting to precisely assess vitality projections for AI is that Huge Tech is speeding to combine it into just about each facet of on a regular basis life. A few of these AI functions are user-facing, like Google’s lately introduced “AI-powered personal health coach,” or Spotify’s uncanny potential to foretell what tune you need to hearken to subsequent. Others are hidden from shoppers—together with, for instance, the efforts of bank card firms to identify fraud and the work of know-how companies like Siemens to foretell gear failures earlier than they occur. Researchers are at present constructing autonomous AI “agents” that may carry out duties for us with far much less supervision than the present fashions require, fixing multistep issues based mostly on a single immediate. These brokers could spend hours researching and writing advanced studies for subscribing customers, utilizing monumental portions of vitality within the course of. This actuality isn’t far off: OpenAI is reportedly planning to supply the companies of autonomous AI brokers for $20,000 per 30 days. All of this makes the know-how’s future environmental influence virtually not possible to foretell. As O’Donnell and Crownhart put it, “every researcher we spoke to said that we cannot understand the energy demands of this future by simply extrapolating from the energy used in AI queries today.”
AI’s environmental toll seems nonetheless better when one considers the know-how’s local weather implications past the vitality consumption of its quickly proliferating knowledge facilities. The political economist Benedetta Brevini, in her 2021 e book Is AI Good for the Planet?, notes that corporations are promoting AI companies not simply to companies that need to assess reams of knowledge, reminiscent of funding banks and hedge funds, but in addition to fossil gas companies, promoting the brand new know-how’s potential to optimize the tempo and productiveness of extraction.1 Each Amazon and Microsoft provide AI instruments to fossil gas firms to assist them establish new oil and fuel reserves. This helps the fossil gas business with out even the excuse lengthy used to guard coal: that it creates jobs and advantages employees. The US’s crude output has been breaking data for the previous few years, however jobs within the oil and fuel sector have grow to be more and more scarce—having fallen to 380,000 final yr from a excessive of 600,000 in 2014—as AI takes over figuring out optimum drilling areas, monitoring pipelines, and predicting the rise and fall of demand.
Additional environmental harm, as Brevini and the South African sociologist Michael Kwet each notice, comes from the manufacturing of the gadgets on which AI runs. Regardless of the conveniently intangible associations of “cloud” computing, the GPUs that host AI fashions include a skinny layer of semiconductor, often silicon, onto which elements made of assorted metals are layered. Typical metals employed embrace copper, aluminum, cobalt, and tungsten, all of which include an environmental value. Lots of of tons of ore must be dug as much as yield a single ton of useable metallic–a course of that’s doubly polluting, since extraction not solely poisons the air and water but in addition requires massive portions of vitality. Many of those metals are, furthermore, mined in areas rife with human rights abuses and armed conflicts. In keeping with an evaluation of its 2023 Battle Minerals Report, for instance, Amazon couldn’t rule out the likelihood that a few of its suppliers had sourced minerals from 9 of ten African nations the place human rights–violating militias finance themselves via mining.
The advantages of the know-how are as inconsistently distributed as their prices. Brevini notes that whereas consultancies emphasize AI’s “global benefits,” these advantages are concentrated within the US and Asia. In keeping with a 2024 report by the World Mental Property Group, between 2014 and 2023 China filed 70 p.c of patents associated to generative AI, adopted by the US at 11 p.c, Korea at 8 p.c, and Japan at 6 p.c.
Whereas European nations are amongst these lagging behind on analysis and patents, the implications are particularly critical for World South nations: most of the latter are locked out of AI features by their lack of infrastructure, Brevini argues, whereas their economies stand to be the worst impacted by AI’s uptake. At the same time as “rich tech giants in the Global North monopolize the means of computation and knowledge,” as Kwet places it in his current e book Digital Degrowth, “the poor countries perform the menial labor, like digging in the dirt for metal, picking coffee beans, labeling data to train artificial intelligence models, or cleansing social media networks of disturbing content.”2
There are usually two proposals from the left for mitigating the abuses of Huge Tech: antitrust authorized measures that may supposedly reestablish “fair” and “competitive” capitalism by curbing the business’s concentrated energy and wealth; and human rights measures that may set up “ethical AI” by rectifying algorithmic bias and surveillance. Neither of those, as Kwet argues, would do a lot to curb AI’s vitality prices. Breaking apart Huge Tech companies utilizing antitrust laws, as an illustration, would do little greater than exchange a couple of megacompanies with myriad smaller rivals, all of which might nonetheless be vying to develop as a lot as potential.
Rather than such proposals, Kwet advocates a worldwide effort to dismantle Huge Tech’s present near-monopoly on the technique of computation. Web customers in each the World North and South ought to, he argues, arrange to boycott software program and platforms created by companies like Amazon, Meta, and Microsoft, and as a substitute use free applications, whose licenses make sure that anybody can entry and modify the software program itself. Prized freed from the ceaseless progress crucial that drives Huge Tech, such “People’s Tech” could possibly be used to assist society determine extra democratically and rationally the best way to use environmental sources–and even the best way to wind down pointless manufacturing and consumption. Kwet argues that Individuals’s Tech must be an integral factor in a broader “Digital Tech Deal” that aligns with a sweeping slate of environmental reforms, such because the Crimson Deal proposed by the Indigenous group The Crimson Nation. However that is, he admits, a longer-term and much more bold undertaking.
Chip Somodevilla/Getty Photographs
Donald Trump addressing a summit on “winning the AI race” cohosted by the Hill & Valley Discussion board and the enterprise capital podcast All-In, Washington, D.C., July 23, 2025
It’s simply as exhausting to think about difficult the fabric foundations of Huge Tech’s hegemony. As Kwet acknowledges, producing {hardware} like semiconductors, a few of which at the moment are solely an atom thick, requires important concentrations of technical data and capital. However that ought to not dissuade us, he argues, from working to socialize provide chains in addition to digital intelligence and knowledge as part of what he calls “People’s Tech for People’s Power.” He factors to examples just like the Open Knowledge motion, which advocates for privateness, safety, transparency, and democratic decision-making over the gathering, storage, and utilization of knowledge.
It might prove that the present AI frenzy is merely a bubble, as Emily Bender and Alex Hanna argue of their current e book The AI Con. However given the Trump administration’s unbridled backing of the tech business, it should take important political organizing to pop it. There are some vital reforms that we will struggle for whereas we additionally try to construct the Individuals’s Tech that Kwet requires. We should first problem the rhetoric about AI’s limitless want for extra vitality. AI might be made much more environment friendly if knowledge facilities are required to be even considerably versatile about once they devour energy, dialing again their use through the comparatively uncommon durations when regional energy grids expertise peak stress (assume warmth waves or polar vortexes). A current report from researchers at Duke College means that the US grid would have already got sufficient capability to energy many new knowledge facilities with minimal enlargement if tech firms had been required to undertake these protocols. AI fashions might equally be placed on versatile coaching schedules in order that they use vitality at optimum moments.
Proof means that none of those reforms will occur with out authorities regulation. Activists within the US should revive and struggle for the passage of the AI Environmental Impacts Act that did not go in 2024, which might have made the know-how’s vitality results extra clear. However main tech firms must also be made to pay for expansions within the infrastructure needed to provide the elevated quantities of electrical energy they are saying they want. As M.V. Ramana argues, when tech corporations pledge to help a nuclear renaissance within the US, their promised contributions are paltry compared to the precise value of constructing the fleet of energy crops they suggest.
Making Huge Tech pay for the facility it consumes would possibly encourage tech executives to align themselves towards the present Republican effort to defund the renewable vitality buildout that started with the Biden administration’s Inflation Discount Act. Including utility-scale photo voltaic and wind energy to the grid is, in spite of everything, far cheaper and sooner not merely than the nuclear boondoggle but in addition than including new fossil fuel energy crops. States must be requiring a speedy and thoroughly deliberate buildout of such renewable capability. The Construct Public Renewables Act, handed in New York by the Public Energy NY marketing campaign in 2023, can serve for instance.
Such reforms are all of the extra needed provided that, as a current report by Eliza Martin and Ari Peskoe of Harvard Legislation Faculty paperwork, the nation’s for-profit utilities stand to spice up their earnings by attracting influential tech companies. Utilities reminiscent of Duke Vitality proudly publicize their offers to generate clear electrical energy (together with small modular nuclear reactors) for corporations reminiscent of Amazon, Microsoft, and Google; in addition they repeatedly signal different contracts with knowledge facilities to which they supply electrical energy. The vitality firms guarantee their state’s public utility commissions “that the deals for Big Tech isolate data center energy costs from other ratepayers’ bills and won’t increase consumers’ power prices,” Martin and Peskoe write. “But verifying this claim is all but impossible,” they observe, not least as a result of “regulators frequently approve special contracts in short and conclusory orders,” and utility firms have a transparent monetary incentive to supply tech corporations discounted vitality charges and lift charges elsewhere to make up the distinction. In different contexts, utilities have reportedly executed simply that: current litigation towards Duke, one of many nation’s largest utilities, revealed that the corporate supplied town of Fayetteville, North Carolina a $325 million low cost; the district court docket famous that an inside Duke doc “disclosed a plan to shift the cost of the discount…back to its wholesale [and] retail customers in years to come.”
The nation’s century-old regulatory regime relies on the concept that the general public advantages from new infrastructure like energy crops and transmission traces. State regulators due to this fact virtually at all times approve utilities’ requests for brand new infrastructure, the price of which might be handed on to prospects—whether or not that infrastructure is being constructed to accommodate the wants of a rising city inhabitants or a single new company consumer. Martin and Peskoe argue that this method dangers lumping unusual shoppers along with institutional shoppers of energy, together with knowledge facilities: “The very same rate structures that have socialized the costs of reliable power delivery are now forcing the public to pay for infrastructure designed to supply a handful of exceedingly wealthy corporations,” Martin and Peskoe write. We’re, in essence, all being requested to shoulder the vitality and environmental burden of the AI growth.
However, like different environmental burdens, the toll of the AI growth doesn’t fall equally. It cleaves to already deep traces of race and sophistication injustice. Elon Musk’s xAI knowledge heart in Memphis, for instance, was constructed close to a predominantly Black neighborhood that environmental justice advocates say has been plagued for many years by air pollution from close by industrial crops. The thirty-five fossil fuel generators that energy the supercomputer that runs the corporate’s AI chatbot, Grok, emit 1,200 to 2,000 tons of nitrogen oxides a yr. In keeping with analysis by the SELC, that is much more air pollution than the gas-fired energy plant emits throughout the road from the xAI facility, or than the oil refinery generates down the street. The county leads the state in emergency division visits for bronchial asthma. At first, Politico reported in Could, none of xAI’s generators had been geared up with air pollution controls often required by the federal authorities; the corporate solely obtained a allow from the Shelby County Health Division in July, after sustained neighborhood criticism. The SELC needed to accomplice with a conservation group of volunteer pilots to do a flyover simply to learn the way many gas-fired generators xAI was utilizing.
Communities within the South are beginning to struggle again. Final month the SELC alleged that residents nonetheless lacked “critical information about financial and environmental impacts” from a “hyperscale data center complex” that public officers and builders hoped to construct in Bessemer, Alabama. In keeping with the SELC, the unique plan for the middle, often called “Project Marvel,” referred to as for the development of eighteen server farms, every the scale of a Walmart Supercenter, on a seven-hundred acre campus that may devour no less than two million gallons of water a day and require 1,200 megawatts of energy to function—9 p.c of Alabama Energy’s whole capability, per some estimates. At a crowded public assembly this August, the Bessemer Metropolis Council voted to ship the information heart undertaking again to the planning and zoning board. One month later, the board voted in favor of the proposed improvement, which now heads to town council. Whether or not or not the event goes via, the episode urged that ever extra persons are coming round to the hurt that AI and its attendant infrastructure does to our environments and communities. “God’s most important rule is to love your neighbor as yourself,” an area resident named Mary Rosenbloom mentioned to the SELC. “But data centers are anything but good neighbors.”
