Meta tags:
Headings (most frequently used words):
and, environmental, use, impact, of, ai, carbon, footprint, energy, policy, individual, level, contents, water, usage, waste, mining, social, justice, climate, solutions, regulation, see, also, notes, references, system, decisions, strategies, by, companies, conflict, on, the, for, research, united, states, european, union, efficiency, state, france, germany, italy,
Text of the page (most frequently used words):
and (156), the (131), water (68), #environmental (58), energy (54), for (52), data (51), 2026 (42), use (39), that (35), retrieved (34), 2024 (30), 2025 (28), edit (26), waste (24), doi (24), also (24), with (21), april (21), artificial (21), systems (21), electricity (21), impact (20), about (20), footprint (20), related (19), study (19), intelligence (18), climate (18), report (17), centres (17), have (17), from (16), july (16), efficiency (16), infrastructure (15), consumption (15), states (15), hardware (15), learning (14), emissions (14), impacts (14), carbon (14), can (13), could (13), policy (12), may (12), model (12), research (12), are (12), reported (12), used (12), per (12), change (11), center (11), effects (11), united (11), global (11), 2023 (11), has (11), litres (11), text (10), models (10), training (10), google (10), system (10), centre (10), been (10), between (10), this (9), human (9), applications (9), neural (9), network (9), prompt (9), sustainable (9), sustainability (9), estimated (9), cooling (9), its (9), billion (9), under (8), image (8), large (8), generation (8), green (8), power (8), european (8), servers (8), 2030 (8), international (8), agency (8), reporting (8), demand (8), while (8), article (8), wikipedia (7), generative (7), centers (7), machine (7), pollution (7), strategy (7), federal (7), 1038 (7), nature (7), weather (7), acm (7), national (7), such (7), which (7), year (7), growth (7), local (7), using (6), page (6), was (6), state (6), computer (6), alex (6), based (6), companies (6), mitigation (6), pdf (6), digital (6), management (6), estimates (6), their (6), usage (6), withdrawal (6), than (6), processing (6), level (6), toggle (5), environment (5), technology (5), social (5), regulation (5), david (5), world (5), deep (5), source (5), assessment (5), ecological (5), cycle (5), issues (5), manufacturing (5), france (5), forecasting (5), over (5), economy (5), strategies (5), 1145 (5), review (5), scientific (5), lower (5), deployment (5), performance (5), including (5), million (5), found (5), not (5), specific (5), depend (5), example (5), one (5), supply (5), other (5), main (5), total (5), accounting (5), tasks (5), contents (4), search (4), you (4), act (4), government (4), john (4), reasoning (4), gpt (4), general (4), engineering (4), loss (4), urban (4), recycling (4), production (4), land (4), gas (4), mining (4), life (4), commission (4), italy (4), germany (4), november (4), issn (4), hdl (4), 025 (4), gov (4), pmid (4), pmc (4), chen (4), medium (4), associated (4), facility (4), development (4), 2021 (4), isbn (4), collection (4), 1016 (4), more (4), electronic (4), grid (4), february (4), chips (4), reports (4), arxiv (4), transparency (4), much (4), scale (4), includes (4), some (4), published (4), require (4), legislation (4), would (4), direct (4), standards (4), because (4), results (4), concerns (4), vary (4), stressed (4), authors (4), difficult (4), upstream (4), framework (4), increased (4), individual (4), hide (4), move (4), sidebar (4), view (3), organization (3), multiple (3), url (3), short (3), description (3), wikidata (3), renewable (3), architecture (3), economic (3), daniel (3), andrew (3), knowledge (3), inference (3), self (3), openai (3), list (3), language (3), diffusion (3), automated (3), open (3), improvement (3), gradient (3), history (3), conservation (3), degradation (3), resource (3), depletion (3), greenhouse (3), agriculture (3), marine (3), paper (3), boundaries (3), 2020 (3), net (3), communications (3), how (3), memphis (3), naacp (3), xai (3), decisions (3), around (3), june (3), 2022 (3), october (3), vries (3), gao (3), health (3), semiconductor (3), december (3), shehabi (3), arman (3), 979 (3), 4007 (3), footprints (3), what (3), berkeley (3), laboratory (3), operational (3), effectiveness (3), wue (3), metric (3), patterson (3), will (3), ben (3), paris (3), 2008 (3), 3732 (3), writing (3), proceedings (3), cloud (3), congress (3), computational (3), annual (3), references (3), include (3), efficient (3), introduced (3), stated (3), information (3), union (3), governments (3), benefits (3), extent (3), traffic (3), reduce (3), studies (3), solutions (3), stress (3), facilities (3), half (3), regions (3), materials (3), but (3), depends (3), scenarios (3), higher (3), chain (3), computing (3), high (3), expected (3), remain (3), typical (3), average (3), when (3), consume (3), intensity (3), 000 (3), median (3), design (3), methods (3), operators (3), measures (3), changes (3), small (3), pue (3), log (3), account (3), assumptions (3), benchmark (3), roughly (3), tools (3), subsection (3), add (2), languages (2), table (2), developers (2), legal (2), safety (2), contact (2), privacy (2), additional (2), terms (2), site (2), last (2), categories (2), pages (2), cs1 (2), maint (2), status (2), index (2), category (2), workplace (2), visual (2), video (2), race (2), ethics (2), war (2), autoencoder (2), rnn (2), recurrent (2), unit (2), vision (2), transformer (2), turing (2), architectures (2), ian (2), quoc (2), fei (2), james (2), joseph (2), control (2), action (2), driving (2), ibm (2), project (2), recognition (2), speech (2), synthesis (2), protocol (2), laws (2), agent (2), context (2), playing (2), symbolic (2), post (2), attention (2), method (2), descent (2), planning (2), software (2), projects (2), forest (2), transport (2), ecology (2), roof (2), advocacy (2), meat (2), community (2), alternative (2), surface (2), freshwater (2), deforestation (2), decline (2), populations (2), biodiversity (2), transportation (2), reservoirs (2), fishing (2), personal (2), fashion (2), oil (2), industry (2), bitcoin (2), causes (2), watch (2), francesco (2), zero (2), challenge (2), president (2), trump (2), light (2), robin (2), ioannis (2), gonzalez (2), extreme (2), events (2), zhang (2), range (2), come (2), musk (2), company (2), groups (2), unctad (2), journal (2), circular (2), application (2), household (2), huang (2), 106498 (2), oise (2), godfrey (2), konyeha (2), susan (2), august (2), 108872 (2), resconrec (2), resources (2), generate (2), tons (2), ssrn (2), value (2), sarah (2), reuters (2), siddik (2), abu (2), shumba (2), pengfei (2), ren (2), shaolei (2), dataset (2), african (2), 101430 (2), lei (2), nuoa (2), masanet (2), eric (2), 108310 (2), key (2), lawrence (2), making (2), less (2), 2011 (2), generating (2), 1009674 (2), compute (2), dean (2), jeff (2), gomes (2), iea (2), link (2), cite (2), where (2), s41598 (2), programming (2), 024 (2), here (2), luccioni (2), sasha (2), strubell (2), emma (2), cost (2), conference (2), fairness (2), accountability (2), epoch (2), does (2), chatgpt (2), science (2), clean (2), often (2), defined (2), sources (2), notes (2), see (2), documents (2), cubic (2), current (2), targets (2), least (2), first (2), costs (2), issue (2), california (2), standardise (2), measurement (2), institute (2), there (2), rebound (2), gains (2), conflict (2), whether (2), proposed (2), optimisation (2), uses (2), intended (2), strong (2), evaluation (2), air (2), burdens (2), existing (2), fifth (2), came (2), moderately (2), highly (2), nearly (2), were (2), fully (2), partially (2), powered (2), plants (2), located (2), watersheds (2), commonly (2), online (2), after (2), justice (2), specialised (2), chains (2), minerals (2), extraction (2), these (2), increase (2), components (2), expansion (2), suggested (2), help (2), tonnes (2), examined (2), projected (2), rapid (2), contribute (2), rising (2), identified (2), public (2), due (2), sensitive (2), location (2), materially (2), affect (2), said (2), length (2), 200 (2), respectively (2), reach (2), released (2), they (2), 700 (2), included (2), depending (2), gemini (2), assistant (2), serving (2), estimate (2), rise (2), consumed (2), suggests (2), across (2), all (2), amount (2), workload (2), substantially (2), estimating (2), equipment (2), comparisons (2), party (2), manufacture (2), selected (2), later (2), researchers (2), units (2), university (2), stages (2), substantial (2), twh (2), relatively (2), share (2), primarily (2), only (2), overhead (2), supporting (2), according (2), times (2), measured (2), parameters (2), response (2), joules (2), seconds (2), different (2), factor (2), charge (2), smartphone (2), watt (2), hours (2), appearance (2), upload (2), file (2), links (2), read (2), create (2), donate (2), menu (2), topic, mobile, cookie, statement, statistics, code, conduct, contacts, disclaimers, available, apply, agree, registered, trademark, non, profit, wikimedia, foundation, inc, creative, commons, attribution, sharealike, license, rendered, parsoid, edited, september, utc, hidden, auto, scaled, images, dmy, dates, empty, articles, https, org, php, title, environmental_impact_of_ai, oldid, 1374551616, portals, warfare, military, art, games, marketing, chatbot, psychosis, healthcare, fiction, education, engine, optimization, explainable, competition, arms, anthropomorphism, winter, slop, literacy, effect, bubble, boom, opposition, propaganda, virtual, politician, precautionary, principle, nationalism, elections, takeover, alignment, cold, political, graph, gnn, adversarial, gan, variational, vae, mamba, highway, residual, convolutional, cnn, multilayer, perceptron, mlp, echo, gated, gru, long, term, memory, lstm, vit, differentiable, françois, chollet, kokotajlo, jan, leike, mustafa, suleyman, schulman, aidan, gomez, noam, shazeer, ashish, vaswani, andrej, karpathy, silver, demis, hassabis, goodfellow, oriol, vinyals, ilya, sutskever, krizhevsky, goodnight, graves, stephen, grossberg, lotfi, zadeh, yoshua, bengio, yann, lecun, jürgen, schmidhuber, hopfield, geoffrey, hinton, paul, werbos, seppo, linnainmaa, seymour, papert, weizenbaum, bernard, widrow, frank, rosenblatt, oliver, selfridge, herbert, simon, cliff, shaw, allen, newell, nathaniel, rochester, mccarthy, marvin, minsky, takeo, kanade, kunihiko, fukushima, shun, ichi, amari, claude, shannon, christopher, manning, von, neumann, walter, pitts, warren, sturgis, mcculloch, alan, people, yago, dbpedia, conceptnet, bases, opencog, lida, clarion, soar, cognitive, rule, semantic, reasoners, procedural, logic, programs, engines, expert, deductive, classifiers, robot, autogpt, selection, muzero, car, five, alphazero, alphago, decisional, watsonx, watson, debater, oasis, genie, udio, suno, riffusion, music, veo, seedance, sora, kling, hailuo, runway, gen, dream, stable, recraft, midjourney, imagen, ideogram, flux, firefly, dall, aurora, alphafold, facial, whisper, elevenlabs, ocr, hwr, wavenet, alexnet, audio, implementations, physical, agent2agent, hypothetical, superintelligence, asi, agi, weak, lethal, autonomous, weapons, humanity, exam, companion, intelligent, nmt, game, theorem, proving, actor, critic, algorithm, situated, approach, neuro, sovereign, blended, vibe, coding, word, embedding, hallucination, recursive, reflection, supervised, rlhf, llm, uncanny, valley, rag, adversary, autoregression, latent, imitation, sarsa, reinforcement, augmentation, datasets, regularization, weight, initialization, gating, rectifier, sigmoid, softmax, activation, batchnorm, normalization, convolution, backpropagation, conjugate, quasi, newton, sgd, clustering, overfitting, double, bias, variance, tradeoff, regression, functions, hyperparameter, parameter, representation, constraint, satisfaction, concepts, lists, proprietary, institutions, algorithms, glossary, timeline, country, minimization, forestry, reuse, reforestation, rainwater, harvesting, organic, farming, banking, industrial, building, restoration, decoupling, cycling, cultured, resilience, cleaner, blue, birth, bioswale, fuel, vehicle, propulsion, scarcity, tropical, cyclones, technofossilization, phosphorus, ozone, ocean, acidification, belts, damage, nitrogen, holocene, extinction, habitat, destruction, dieback, externality, externalities, cars, erosion, coral, reefs, insecurity, livestock, breadbasket, failure, crisis, ecocide, desertification, defaunation, runaway, insect, amphibian, threats, sprawl, urbanization, shipping, roads, rail, aviation, tourism, quarrying, particulates, overgrazing, overexploitation, groundwater, overconsumption, overfishing, down, food, web, pharmaceuticals, care, pesticides, paint, nanotechnology, plastics, concrete, cleaning, agents, industrialisation, explosives, crime, genetic, petroleum, shale, nuclear, fracking, coal, biodiesel, biofuels, corporate, behavior, palm, cocoa, irrigation, cannabis, cultivation, animal, planetary, anthropocene, german, xiao, tianqi, nerini, fuso, matthews, damon, tavoni, massimo, fengqi, 1553, 2398, 9629, 11311, 1309509, s41893, 01681, 1541, pathways, usa, scarboro, morgan, miller, kim, multistate, push, www, plan, camps, valls, gustau, fernández, torres, miguel, ángel, cohrs, kai, hendrik, höhl, adrian, castelletti, andrea, pacal, aytac, claire, martinuzzi, papoutsis, prapas, pérez, aracil, jorge, weigel, katja, calabuig, maria, reichstein, markus, rabel, martin, 1919, 39994190, 11850610, s41467, 56573, modeling, understanding, kaifeng, xie, lingxi, hengheng, xin, xiaotao, tian, accurate, networks, 538, s41586, 023, 06185, 533, 619, wittenberg, ariel, politico, breathe, draws, backlash, appeal, permit, south, unpermitted, methane, turbines, press, group, notify, elon, intent, sue, trade, calls, equitable, mitigate, haomiao, jin, zian, krishnamoorthy, sujatha, 978, 4503, 8980, 3469968, 3469973, shreyas, madhav, rajaraman, raghav, harini, kiliroor, cinu, 1053, 34726090, 9109239, 0734, 242x, 1177, 0734242x211052846, 1047, enhance, potential, solution, weee, segregation, india, jie, shoujun, balamurugan, tamizharasi, eiar, 2731, 0809, 1007, s44163, 00376, discover, perfectson, mdpi, 3390, engproc2024067066, 1871, a98e54d1, c817, 45db, a1f6, 8a40365311df, 229, recalibrating, iqbal, saima, american, millions, rakesh, 1556, 5068, 2139, 6731365, accelerated, decay, terminal, hyper, capital, intensive, forum, big, tech, labrecque, desert, storm, slake, insatiable, thirst, bakar, marston, landon, 064017, 10919, 109747, 1088, 1748, 9326, abfba1, letters, noah, tshekiso, opelo, fanti, giulia, 453, 460, 1484, 3715335, 3735483, 41583976, 12827721, patter, patterns, mean, jun, 219, workloads, determinants, smith, hubbard, newkirk, bakkar, holecek, billie, koomey, jonathan, sartor, dale, lbnl, 2001637, 242, 243, yang, jianyi, islam, mohammad, 3724499, thirsty, macknick, jordan, newmark, heath, garvin, hallett, march, factors, technologies, nrel, 6a20, 50900, osti, 2172, azevedo, 2018, hölzle, urs, liang, munguia, lluis, miquel, rothchild, texier, maud, plateau, then, shrink, 1109, 3148714, brandt, kate, keyword, our, alšauskas, oskaras, outlook, news, bhattarai, abha, lerman, rachel, msci, washington, charts, show, heading, investments, woo, nolan, 39182, 41214116, 12603251, 24658, comparative, tomlinson, bill, black, rebecca, donald, torrance, 54271, illustrating, humans, mit, did, math, story, haven, heard, jernite, yacine, hungry, watts, 0450, 3630106, 3658542, 2311, 16863, josh, lambert, katherine, 518, 540, 2596, 3805689, 3812401, 2605, 05416, cradle, library, 118th, elsworth, cooper, keguo, schneider, sedivy, robert, goodman, savannah, townsend, ranganathan, parthasarathy, vahdat, amin, manyika, measuring, delivering, 2508, 15734, wang, peng, ling, tzachor, asaf, wei, qiang, challenges, 823, s43588, 00712, 818, mytton, 10044, 86841, s41545, 021, 00101, npj, ganesh, ananya, mccallum, 2019, considerations, nlp, 3650, 18653, p19, 1355, 1906, 02243, 3645, 57th, meeting, association, linguistics, executive, summary, minus, discharge, evaporated, otherwise, removed, immediate, refers, taken, ground, either, temporarily, permanently, decarbonisation, computers, priorities, discuss, areas, goals, sections, into, discussed, relation, transition, maintain, member, directive, obligations, database, collects, relevant, delegated, sets, out, indicators, scheme, approximately, 731, 125, meters, annually, trajectories, unlikely, meet, without, looked, address, infrastructural, considering, put, being, pass, requires, bear, new, pushing, requiring, addressed, looking, focus, utah, ohio, proposals, senate, convene, consortium, establish, voluntary, split, regulations, minimal, saying, website, achieve, dominance, whoever, largest, ecosystem, set, reap, broad, security, ongoing, debate, balance, possible, scaling, discussion, offset, produces, remains, question, outcomes, choices, decarbonised, oriented, recommend, signal, timing, adjustments, stop, intersections, peer, reviewed, frameworks, pangu, relative, leading, numerical, prediction, event, modelling, despite, earth, observation, raised, permitting, communities, hosting, civil, rights, challenged, permits, connected, area, arguing, fall, disproportionately, historically, overburdened, neighbourhoods, dispute, subject, regulatory, distributed, evenly, region, intersect, availability, adopted, symbol, resistance, protests, campaigns, involving, critical, whose, significant, unique, grows, complex, metals, manufactured, increases, raw, raises, distributional, linked, tracking, end, represent, along, instead, 131, 224, thousand, turnover, accelerators, growing, primary, contributor, contain, hazardous, like, swapped, frequently, completely, replace, every, three, years, competitive, consumer, electronics, possibility, storing, chromium, mercury, lead, occurs, relies, quantities, ultrapure, fabrication, significance, citing, verisk, maplecroft, naturefinance, mid, sized, day, experience, currently, planned, phoenix, suggest, email, 120, words, llama3, 70b, liters, ten, draw, 312, 764, same, colleagues, metres, 600, 2027, onsite, offsite, responses, 500, deployed, varied, microsoft, 560, two, thirds, quarter, remainder, microchip, indirect, 800, unspecified, however, variation, server, role, argued, observe, summarised, emphasised, distinction, divided, itself, measure, distinguish, directly, indirectly, through, inconsistent, mix, announced, investment, low, procurement, programmes, comparability, third, verification, disclosures, firms, jurisdictions, stewardship, affects, attributed, part, embodied, construction, early, analyses, frontier, historical, although, trends, depended, decrease, time, assuming, adoption, best, practices, picking, locations, proportion, optimized, moving, off, premise, strongly, utilisation, rates, lifecycle, performed, contributes, 415, 945, driver, alongside, services, overall, pronounced, geographically, concentrated, publish, metrics, case, sourcing, averages, own, ratio, indicates, normalized, equals, 100, fueled, 2020s, labour, produced, mixed, output, quality, 130, 2900, writers, illustrators, counterexample, finding, evaluated, programmers, various, smaller, 114, 03167, larger, 405, 861, corresponds, needed, run, tenth, second, eight, microwave, oven, llama, meta, michigan, measurements, environments, noting, produce, watching, nine, television, improvements, reduced, within, request, widely, presented, differences, task, types, test, conditions, simple, classification, 002, 007, prompts, summarisation, each, averaged, equivalent, query, compared, minute, almost, arises, storage, networking, conversion, focuses, rather, aspects, creation, decommissioning, fine, tuning, received, greater, regulators, increasingly, considered, requirements, broader, idle, capacity, computation, material, integrated, circuits, tensor, die, blackwell, processors, gpus, graphics, nvidia, gb200, free, encyclopedia, item, printable, version, download, print, export, switch, legacy, parser, get, shortened, permanent, actions, english, talk, русский, português, 한국어, bahasa, indonesia, հայերեն, hausa, français, español, čeština, العربية, top, special, recent, portal, learn, random, navigation, jump, content,
Text of the page (random words):
ance 7 2 2 germany 7 2 3 italy 8 see also 9 notes 10 references toggle the table of contents environmental impact of ai 10 languages العربية čeština español français hausa հայերեն bahasa indonesia 한국어 português русский edit links article talk english read edit view history tools tools move to sidebar hide actions read edit view history general what links here related changes upload file permanent link page information cite this page get shortened url switch to legacy parser print export download as pdf printable version in other projects wikidata item appearance move to sidebar hide from wikipedia the free encyclopedia nvidia gb200 die with blackwell processors an example of graphics processing units gpus used for ai a google tensor processing unit an example of application specific integrated circuits used for ai the environmental impact of the design training deployment and use of artificial intelligence includes the greenhouse gas emissions from generating electricity for data centres and computing hardware operational and upstream water use and material impacts from hardware manufacturing mining and electronic waste 1 2 3 4 estimating ai s environmental effects can be difficult because results depend on how impacts are measured including whether accounting includes only model computation or also data centre overhead idle capacity hardware manufacture and local electricity supply 5 as these issues have received greater attention governments and regulators have increasingly considered data centre reporting requirements energy efficiency standards and broader transparency measures for ai related resource use 6 7 carbon footprint and energy use edit ai related energy use arises at multiple stages including model training fine tuning inference storage networking and supporting infrastructure such as cooling and power conversion 2 existing research primarily focuses on the cost of model training and deployment rather than other aspects of the ai life cycle such as dataset creation and decommissioning 8 individual level edit according to research institute epoch ai energy consumption per typical chatgpt query 0 3 watt hours is small compared to the average u s household consumption per minute almost 20 watt hours 9 published estimates of energy use per ai request vary widely across models tasks and measurement methods 5 a benchmark study presented at the 2024 acm conference on fairness accountability and transparency found substantial differences between task types with lower energy use for some text tasks and much higher energy use for image generation in the study s test conditions 10 in that benchmark simple classification tasks consumed about 0 002 0 007 wh per prompt on average about 9 of a smartphone charge for 1 000 prompts while text generation and text summarisation each used about 0 05 wh per prompt image generation averaged 2 91 wh per prompt and the least efficient image model in the study used 11 49 wh per image roughly equivalent to half a smartphone charge 10 first party measurements in production environments have also been published a 2025 google study on gemini assistant serving reported median per prompt energy emissions and water use estimates under the authors accounting framework while noting that different system boundaries can produce substantially different results 5 the study reported a median text prompt estimate of about 0 24 wh which is roughly as much energy as watching nine seconds of television the study also stated that software and infrastructure improvements reduced energy use by a factor of 33 and carbon emissions by a factor of 44 for a typical prompt over one year within the authors framework 5 researchers at the university of michigan measured the energy consumption of various meta llama 3 1 models released in 2024 and found that smaller language models 8 billion parameters use about 114 joules 0 03167 wh per response while larger models 405 billion parameters require up to 6 700 joules 1 861 wh per response this corresponds to the energy needed to run a microwave oven for roughly one tenth of a second and eight seconds respectively 11 comparisons between ai systems and human labour for specific tasks have produced mixed results and remain sensitive to assumptions about output quality workload and system boundaries a 2024 study in scientific reports reported 130 to 2900 times lower estimated carbon emissions for selected ai systems than for human writers and illustrators under its assumptions 12 a later scientific reports paper reported a counterexample for programming tasks under its assumptions finding 5 to 19 times higher estimated emissions for the evaluated ai system than for human programmers on the benchmark used in that study 13 system level edit energy use and efficiency edit fueled by growth in artificial intelligence data centres demand for power increased in the 2020s 14 according to the international energy agency data centres are expected to account for a relatively small share of global electricity demand growth by 2030 15 efficiency improvement of ai related computer chips 2008 2023 index of energy intensity of ai computer chips normalized so 2008 efficiency equals 100 log scale 16 ai electricity intensity depends not only on model architecture but also on hardware and facility efficiency data centre operators commonly report power usage effectiveness pue which measures the ratio of total facility energy to it equipment energy a lower pue indicates less overhead energy for cooling and other supporting infrastructure 5 operators may also publish metrics and case studies on hardware efficiency cooling systems and power sourcing in its 2024 environmental report google stated that its 2023 total greenhouse gas emissions increased 13 year over year primarily because of increased data centre energy consumption and supply chain emissions while also reporting lower pue than industry averages for its own facilities 17 the international energy agency has also reported that data centres remain a relatively small share of global electricity use overall but that their local effects can be much more pronounced because demand is geographically concentrated 1 carbon footprint edit at system level ai contributes to rising electricity demand in data centres and related infrastructure the international energy agency estimated that data centres used about 415 twh of electricity in 2024 or around 1 5 of global electricity consumption and projected that data centre electricity use could rise to about 945 twh by 2030 with ai identified as the main driver of that growth alongside other digital services 1 the carbon footprint of ai systems depends strongly on electricity sources hardware efficiency utilisation rates and what stages are included in the accounting training large models can require substantial electricity while total lifecycle impacts also depend on deployment scale and the amount of inference performed after training 2 18 5 early analyses of frontier model development reported rapid historical growth in training compute for selected systems although later trends have depended on changes in model design hardware and efficiency gains 19 an article by university of california berkeley and google researchers suggests that the carbon footprint of training machine learning models will decrease over time assuming the adoption of best practices such as picking data center locations with high proportion of clean energy using ml optimized processing units and moving computing off premise 18 accounting methods that include upstream or embodied impacts such as hardware manufacture and facilities construction can materially affect estimates of ai related emissions 18 4 decisions and strategies by individual companies edit large technology companies have reported that the expansion of ai and cloud infrastructure affects their sustainability targets electricity demand and resource use google for example attributed part of its emissions growth in 2023 to increased data centre energy consumption and supply chain emissions in its 2024 environmental report 17 cloud and ai companies have also announced measures intended to reduce environmental impacts including investment in more efficient hardware low carbon electricity procurement alternative cooling systems and water stewardship programmes the extent comparability and third party verification of such disclosures vary between firms and jurisdictions 17 6 water usage edit data centres can use water directly for cooling and indirectly through the water used in electricity generation depending on the local energy mix 3 public reporting on data centre water use has often been inconsistent making comparisons between operators and regions difficult 3 to standardise operational reporting the green grid proposed the metric water usage effectiveness wue defined as annual site water use divided by it equipment energy use 20 wue does not by itself measure local water stress source sustainability or all upstream water impacts 20 3 studies of ai water use also distinguish between water withdrawal and water consumption a 22 research on ai specific water use has argued that the water footprint of ai systems can be difficult to observe and may vary substantially by location cooling design and electricity source a 2025 communications of the acm article summarised methods for estimating ai water footprints and emphasised the distinction between water withdrawal and water consumption 22 a 2025 report by the international energy agency said that global water consumption by data centres was around 560 billion litres in 2023 expected to rise to 12 000 billion litres in 2030 of the water consumed in 2023 about two thirds was associated with energy generation and one quarter with cooling with the remainder related to semiconductor and microchip manufacturing 23 a lawrence berkeley national laboratory report from 2024 suggests an indirect water footprint of 800 billion litres in 2023 from electricity generation with direct water consumption of 66 billion litres across all data centres in the united states with an unspecified amount due to ai 24 however there is high variation in water use based on workload with server efficiency playing a large role 25 based on estimates of water intensity and power draw a 2025 article by de vries gao suggested the total water footprint of ai systems could reach between 312 5 and 764 6 billion litres that year 26 the same year li and colleagues estimated that global ai water withdrawal could reach 4 2 6 6 billion cubic metres 4 200 6 600 billion litres in 2027 under the scenarios examined in their article 22 using gpt 3 released by openai in 2020 as an example they estimated that training the model in microsoft s us data centres could consume about 700 000 litres of onsite water and about 5 4 million litres in total when offsite electricity related water use was included they also estimated that 10 50 medium length gpt 3 responses could consume about 500 ml of water depending on when and where the model was deployed 22 published prompt level estimates have also varied by system and accounting framework the 2025 google study on gemini assistant serving reported a median text prompt estimate of about 0 26 ml under its framework 5 an article by shumba et al about water efficiency in african data centers suggest when writing a medium length email of 120 200 words llama3 70b and gpt 4 could consume about 0 12 liters and 2 6 litres of water respectively while writing a ten page report could use 0 6 and 53 litres 27 location can materially affect the significance of data centre water use research on u s data centres found that one fifth of servers direct water footprint came from moderately to highly water stressed watersheds while nearly half of servers were fully or partially powered by plants located in water stressed regions 28 a 2025 reuters report citing data from verisk maplecroft and naturefinance said that an average mid sized data centre uses about 1 4 million litres of water per day for cooling and that phoenix would experience a 32 increase in annual water stress if currently planned data centres come online 29 water use also occurs upstream in semiconductor fabrication which relies on large quantities of ultrapure water 30 e waste edit main article electronic waste ai systems depend on specialised computing hardware and rapid turnover in servers and accelerators may contribute to rising e waste 4 the world health organization identified e waste as a growing environmental and public health issue 31 the primary contributor to e waste from ai is the high performance hardware used in data centers which can contain hazardous materials like lead mercury and chromium 4 ai data center components are also swapped frequently with companies expected to completely replace their hardware every three years to remain competitive 32 recycling ai hardware can also be more difficult than typical consumer electronics due to the possibility of storing sensitive data 33 a 2024 study in nature computational science estimated that generative ai could add between 1 2 and 5 million tonnes of e waste by 2030 under the scenarios examined by the authors 4 in the study s higher end scenarios this would represent up to 12 of projected global e waste by 2030 4 the authors also reported that circular economy strategies along the generative ai value chain could reduce ai related e waste generation by 16 86 4 a study from 2026 instead estimated that by 2030 ai servers would generate 131 0 to 224 8 thousand tonnes of e waste per year 34 deep learning and ai systems have also been suggested to help with e waste tracking collection and management 35 36 37 38 39 mining edit ai hardware depends on complex supply chains for metals minerals and manufactured components unctad has reported that the expansion of digital infrastructure increases demand for raw materials and raises environmental and distributional concerns linked to extraction processing and manufacturing 40 specialised chips used in ai systems can depend on supply chains involving critical minerals and other materials whose extraction and processing may have significant environmental and social effects these impacts are not unique to ai but may increase as demand for ai related hardware grows 40 social impact and environmental justice edit the commonly adopted symbol of ai resistance used in protests and online campaigns after concerns of environmental impact the environmental effects of ai related infrastructure are not distributed evenly research on data centres in the united states has found that their environmental footprints vary by region and may intersect with local electricity systems water availability and existing environmental burdens 28 in that study one fifth of servers direct water footprint came from moderately to highly water stressed watersheds while nearly half of servers were fully...
|
