The Double-Edged Algorithm: How Artificial Intelligence Is Both Fighting and Fuelling the Climate Crisis - Lethal Heating Editor BDA

Is the world's most powerful technology an instrument of decarbonisation
or an accelerant of the crisis it promises to solve?

Key Points

Global data centre electricity consumption reached 415 TWh in 2024, roughly 1.5% of world electricity use, and is projected to nearly double to 945 TWh by 2030, driven primarily by AI workloads. 1 While AI demonstrably improves renewable energy forecasting and grid balancing, tech companies' climate disclosures rely heavily on market-based accounting that can obscure rising location-based emissions. 2 A typical large data centre consumes 300,000 gallons of water per day, and more than 160 new AI facilities have been built in water-stressed regions of the United States since 2022. 3 AI is enabling fossil fuel companies to boost extraction yields by up to 15%, generating so-called "enabled emissions" that are largely absent from corporate climate disclosures. 4 The environmental costs of AI hardware, including cobalt and rare earth mining, fall disproportionately on communities in the Global South, while e-waste from accelerated chip cycles is growing five times faster than documented recycling. 5 Independent analysts warn that tech sector climate pledges lack credibility without regulatory enforcement, standardised carbon labelling for digital services, and mandatory Scope 3 disclosure covering enabled downstream emissions. 6

A Machine That Runs on the World

In a windowless building on the edge of a drought-afflicted county in Arizona, tens of thousands of processors hum without pause, handling everything from product recommendations and legal contract drafts to the modelling of atmospheric systems that may determine the habitability of entire coastlines.

The building is a data centre, and it is, in the language of its operators, merely "infrastructure".

That word — infrastructure — does a great deal of work.

It smooths the edges off a structure that consumes as much electricity as a small city, draws hundreds of thousands of gallons of water from a strained aquifer each day, and sits at the centre of one of the defining tensions of the climate era: the question of whether artificial intelligence, and the digital economy that sustains it, is helping humanity escape catastrophic warming, or quietly making the problem worse.

There is no clean answer.

The evidence, assembled from a growing body of peer-reviewed research, international energy reports, corporate disclosures, and the testimony of engineers, climate scientists, and community advocates, suggests that the relationship between information technology and climate change is neither the salvation the industry promises nor the simple villain its critics describe.

It is something messier, and more consequential.

The Scale of the Problem

Global data centres consumed approximately 415 terawatt-hours of electricity in 2024, representing around 1.5 per cent of the world's total electricity use, according to a landmark report by the International Energy Agency (IEA).¹

That figure, significant on its own, is growing at roughly 12 per cent a year, and the IEA projects it will reach approximately 945 TWh by 2030, an amount roughly equivalent to the total annual electricity consumption of Japan today.¹

AI is the primary driver of that growth.

In 2024, AI-specific servers accounted for an estimated 15 per cent of total data centre energy demand, a share projected to reach 35 to 50 per cent by 2030 as generative AI workloads proliferate.²

A single AI-focused hyperscale data centre, the IEA notes, consumes as much electricity as 100,000 households; the largest now under construction will require twenty times that.

To put those numbers in economic terms: global investment in data centres nearly doubled between 2022 and 2024, reaching half a trillion US dollars in a single year.¹

The pace of this build-out is outrunning the decarbonisation of the grids that power it.

The IEA estimates that natural gas and coal together will meet more than 40 per cent of the additional electricity demanded by data centres through to 2030.³

In the United States, natural gas currently supplies more than 40 per cent of data centre electricity, and it remains the largest source of additional supply as demand surges over the coming five years.³

In China, where the majority of data centres are located in the coal-heavy east, coal accounts for close to 70 per cent of the current data centre electricity mix.³

Training, Inference, and the Carbon Ledger

The carbon costs of AI are distributed unevenly across its life cycle, and the full accounting is rarely made visible in corporate disclosures.

Training a large model like GPT-4 has been estimated to consume approximately 50 gigawatt-hours of electricity, an energy-intensive but one-off event.²

Inference — the continuous process of responding to user queries — accounts for an estimated 80 to 90 per cent of total AI computing work, and therefore forms the dominant share of ongoing energy consumption.²

A single text prompt to a large language model such as Llama 3.1 at 405 billion parameters consumes roughly 6,700 joules; a short AI-generated video can require more than three million joules.²

Multiplied across billions of daily interactions, those figures begin to accumulate in ways that are difficult to track and that most technology companies have shown little enthusiasm for disclosing with precision.

A 2025 study published in a peer-reviewed journal estimated that AI systems may carry a carbon footprint equivalent to that of New York City, with associated water consumption potentially rivalling the world's annual consumption of bottled water.⁷

The lifecycle footprint extends further still, encompassing chip fabrication, the energy-intensive mining of rare earths and critical minerals used in GPU production, and the eventual disposal of hardware that is rapidly made obsolete by the pace of AI development.

According to the United Nations Global E-Waste Monitor, the world produced a record 62 million tonnes of electronic waste in 2022, up 82 per cent from 2010, and is on track to generate 82 million tonnes by 2030, growing five times faster than documented recycling.⁵

Promises in the Ledger: What AI Does for the Climate

The case for AI as a climate instrument begins with energy itself.

Grid operators face an increasingly complex challenge: integrating intermittent renewable sources, particularly wind and solar, into electricity systems designed for predictable, dispatchable generation.

AI-powered weather forecasting has produced measurable improvements in this domain.

Google DeepMind's WeatherNext platform, developed in collaboration with Google Research, can produce accurate forecasts up to 15 days in advance in minutes, compared to the hours previously required by computational weather models.⁸

Working with the UK's National Grid Electricity System Operator and Open Climate Fix, a non-profit research group, DeepMind's solar nowcasting models have helped reduce large forecast errors by 10 per cent over a 24-to-48-hour horizon, a result that allows grid operators to reduce their reliance on carbon-emitting backup generation held in reserve.⁹

Machine learning systems developed for wind forecasting can project power output up to 36 hours in advance, enabling operators to manage renewable variability with greater confidence and integrate clean energy more aggressively.¹⁰

National Grid's own control room has recorded a 33 per cent improvement in the accuracy of solar generation forecasts over recent years, attributable in part to machine-learning applications.⁹

In climate science itself, AI is accelerating the processing of vast observational datasets, contributing to improvements in atmospheric modelling and extreme weather prediction.

These are real gains, and their defenders argue that the carbon return on investment — the emissions avoided per unit of AI energy consumed — more than justifies the energy use in such applications.

Ben Gaiarin, a Technical Program Manager at Google DeepMind, has described this framing explicitly: a narrower cone of forecast uncertainty allows grid operators to make more aggressive use of clean energy, reducing greenhouse gas emissions and improving system stability in ways that can offset the cost of running the AI itself.

Whether that calculus holds across the full range of AI applications is a harder question, and one the industry has been slow to address rigorously.

The Transparency Deficit

Corporate sustainability reports from the major technology companies contain a peculiar structure: they present declining emissions, often achieved through market-based accounting, against a backdrop of sharply rising actual energy consumption and location-based emissions.

Under market-based accounting, companies can claim reduced emissions by purchasing renewable energy certificates or entering into power purchase agreements with clean energy providers, regardless of whether the electricity physically delivered to their facilities is clean.¹¹

The result, documented by the NewClimate Institute's 2025 Corporate Climate Responsibility Monitor, is a widening divergence between the story companies tell and the physical reality on the grid.⁶

Microsoft's location-based Scope 2 emissions, which reflect the actual carbon intensity of the electricity it consumes, more than doubled between 2020 and 2024, rising from 4.3 million metric tonnes of carbon dioxide to nearly 10 million, even as the company's market-based figures showed declining emissions over the same period.¹¹

Google's own environmental report acknowledged that a sharp increase in emissions reflected "the challenge of reducing emissions while compute intensity increases".¹²

A broader assessment by the International Telecommunication Union and the World Benchmarking Alliance, tracking 200 leading digital companies, found that Scope 3 emissions, those generated across the full value chain, including suppliers and downstream product use, are on average six times larger than Scope 1 and 2 emissions combined, yet only a fraction of companies fully disclose them.¹³

Critically absent from most corporate disclosures are what researchers call "enabled emissions": the greenhouse gases released when AI tools are deployed by fossil fuel companies to optimise extraction, pipeline management and reserve discovery.

A report by Global Witness found, on conservative estimates, that AI, the Internet of Things and cloud computing are enabling the fossil fuel industry to boost extraction yields by up to 15 per cent.⁴

Microsoft, Palantir, Amazon Web Services and other technology providers have each maintained commercial relationships with oil and gas operators, supplying tools that are used at every stage of fossil fuel production, from subsurface seismic analysis to predictive maintenance and demand forecasting.⁴

These arrangements are rarely reflected in any company's published climate commitments.

Water, Land, and the Communities Left Holding the Bill

A typical data centre consumes approximately 300,000 gallons of water per day, equivalent to the daily needs of around 1,000 households.¹⁴

Large hyperscale facilities can require up to five million gallons daily, comparable to a town of 50,000 residents.

More than 160 new AI data centres have been built across the United States in the past three years in regions already classified as water-stressed.¹⁵

In Texas, data centres are projected to consume 399 billion gallons of water annually by 2030, according to research by the Houston Advanced Research Center and the University of Houston, an amount that would draw down a reservoir the size of Lake Mead by more than five metres in a single year.¹⁶

In Spain, Amazon sought to increase its water consumption permit at three existing Aragon data centres by 48 per cent in December 2024, noting that climate change would increase cooling demand, at the same time as the region was applying to the European Union for drought relief.¹⁷

In Chile, Google paused a planned $200 million data centre after an environmental court partially reversed its permit, citing the project's reliance on the Santiago Aquifer during a drought that had persisted for fifteen years.¹⁷

Community groups in Spain have formed under the name Tu Nube Seca Mi Río — Your Cloud Is Drying My River — and are calling for a moratorium on new data centre construction in the country.

These are not marginal concerns.

An MSCI analysis of 13,558 data centre assets worldwide found that Alphabet, Amazon, Microsoft and Meta have each faced community opposition and regulatory scrutiny over water and power use, and that exposure to water scarcity is projected to worsen for many of these facilities through to 2050.¹⁸

The Governance Vacuum

The global infrastructure underpinning AI is controlled by a small number of corporations headquartered primarily in the United States and China.

That concentration of computational power is not a neutral fact.

It means that the infrastructure decisions that shape AI's climate impact, where data centres are built, how they are powered, what workloads they prioritise, are made by private actors largely beyond the reach of the multilateral climate governance mechanisms that govern other high-emission sectors.

Calls for mandatory carbon disclosure for AI systems have grown louder among researchers and regulators.

The US Securities and Exchange Commission adopted new climate disclosure rules in March 2024 requiring publicly listed companies to report emissions deemed financially material to investors, but these rules face ongoing legal challenge and do not capture the full Scope 3 picture.¹⁹

The European Commission is developing an energy efficiency labelling package for data centres, including water use metrics, with publication expected in early 2026.²⁰

China remains the only country to have incorporated water use effectiveness standards directly into its data centre building code, according to the IEA.²¹

Voluntary corporate pledges — net zero by 2030, water positive by 2030, 100 per cent renewable energy — are, as the NewClimate Institute has documented, increasingly difficult to reconcile with the trajectory of actual emissions, raising serious questions about their credibility in the absence of binding standards and independent verification.⁶

The Equity Dimension

The global AI economy extracts its materials from one part of the world, operates its infrastructure in another, and concentrates its benefits in a third.

Cobalt, required for the batteries and processors that run AI systems, is mined predominantly in the Democratic Republic of the Congo, often under conditions that have been widely described by human rights researchers as hazardous and exploitative.²²

In Baotou, Mongolia, the refining of rare earth minerals critical to chip production has created a toxic waste lake spanning more than five kilometres, contaminating surrounding ecosystems and exposing local communities to significant harm.²²

The IEA's 2025 Energy and AI report provided the first detailed estimates of the sector's critical mineral requirements, noting that the rapid expansion of AI data centres is adding to pressure on supply chains for copper, aluminium, silicon, gallium and rare earth elements already strained by the clean energy transition.²³

AI-driven climate adaptation tools, from sophisticated flood modelling to precision agriculture platforms, are most accessible to governments and corporations in high-income countries with significant computational capacity.

The frontline communities most exposed to climate impacts, predominantly in the Global South, are least likely to hold the infrastructure or institutional capacity required to deploy these tools at scale.

The question of whether digital climate tools can be designed to serve those communities, rather than merely extracting resources from them, remains largely unanswered by the industry.

The Net Balance: An Unresolved Equation

The IEA has been measured about the climate implications of data centre growth, noting that data centres will account for approximately 1 per cent of global CO₂ emissions by 2030 under its central scenario — significant, but modest relative to heavy industry, transport and buildings.²⁴

The IEA also acknowledges that if AI enables broader emissions reductions across the energy, agriculture, and industrial sectors, those savings could offset its direct footprint.

But that conditional framing depends on factors that are not guaranteed: that AI tools are deployed in emissions-reducing applications rather than efficiency-enhancing ones in high-emission industries; that clean energy scales fast enough to keep pace with AI's electricity demand; and that the rebound effects of efficiency gains, where doing something cheaply leads to doing much more of it, do not erode the savings.

The history of digital efficiency is not encouraging on the last point.

Streaming services, cloud computing and electronic commerce each promised to reduce physical consumption and travel; each also stimulated new forms of demand that expanded the overall energy footprint of the sector.

Researcher Alex de Vries of VU Amsterdam has argued that the IEA's projections may understate AI's direct energy impact, and that the sector's growing reliance on fossil fuels to bridge the gap between clean energy supply and AI demand represents "a serious risk for our ability to achieve our climate goals."²⁵

Conclusion: The Double-Edged Tool

Information technology, including AI, holds genuine promise as a force for climate mitigation.

The evidence for this is real: more accurate renewable energy forecasting, improved grid management, accelerated materials science, and better climate modelling are all credible contributions to the decarbonisation effort.

But those contributions exist alongside a growing physical footprint that is powered in significant part by fossil fuels, draws water from communities that cannot afford to lose it, externalises its material costs onto the Global South, and is deployed by some of its most powerful operators in the direct service of the fossil fuel industry.

The net balance of that ledger is not yet settled.

What is clear is that it cannot be settled by the industry alone.

The questions at stake — how AI's energy growth is governed, whether its emissions are disclosed honestly, whether its benefits are distributed equitably — are questions of public policy and democratic accountability, not corporate strategy.

Without mandatory disclosure of the full lifecycle emissions of AI systems, including Scope 3 and enabled emissions, without regulatory standards for data centre siting and water use, and without binding requirements that large-scale AI deployment be matched by verifiable clean energy, the technology sector's climate promises amount to a ledger in which the costs are borne by those least able to carry them, and the benefits flow to those least likely to be asked to account for them.

The machine keeps running.

The question is whether it will help fix the world it is helping to heat, or simply make the heating more efficient.

References

1. International Energy Agency. (2025). Energy and AI: Executive Summary. IEA.
2. AIM Multiple Research. (2025). AI Energy Consumption Statistics. AIM Multiple.
3. International Energy Agency. (2025). Energy and AI: Energy Supply for AI. IEA.
4. Global Witness. (2024). Enabled Emissions: How AI Helps to Supercharge Oil and Gas Production. Global Witness.
5. CEPR/VoxEU. (2024). An Eco-Political Economy of AI to Understand the Complexities of Its Environmental Costs. Centre for Economic Policy Research.
6. NewClimate Institute. (2025). Corporate Climate Responsibility Monitor 2025: Tech Sector. NewClimate Institute.
7. de Vries, A., & Bieser, J. (2025). The Carbon and Water Footprints of Data Centers and What This Could Mean for Artificial Intelligence. Patterns, Elsevier.
8. Google Cloud. (2025). Transforming Energy Operations with AI-Powered Weather Forecasting. Google Cloud Blog.
9. National Energy System Operator. (2023). Former DeepMind Expert's AI Tool Could Help Boost National Grid ESO's Solar Forecasts. NESO.
10. AI Time Journal. (2024). 7 Groundbreaking AI Trends Reshaping the Renewable Energy Landscape in 2024. AI Time Journal.
11. Policy Review / Internet Policy Review. (2025). Not Greenwashing, But Still… A Closer Look at Big Tech's 2025 Sustainability Reports. Internet Policy Review.
12. International Bar Association. (2024). Sustainability: Big Tech's AI Push Putting Climate Targets at Risk. IBA Global Insight.
13. International Telecommunication Union. (2025). Greening Digital Companies 2025: Monitoring Emissions and Climate Commitments. ITU.
14. Brookings Institution. (2025). AI, Data Centers, and Water. Brookings.
15. Environmental Law Institute. (2025). AI's Cooling Problem: How Data Centers Are Transforming Water Use. ELI.
16. Lincoln Institute of Land Policy. (2025). Data Drain: The Land and Water Impacts of the AI Boom. Lincoln Institute.
17. EthicalGEO. (2025). The Cloud is Drying Our Rivers: Water Usage of AI Data Centres. EthicalGEO.
18. MSCI. (2025). When AI Meets Water Scarcity: Data Centres in a Thirsty World. MSCI.
19. CEPR/VoxEU. (2024). An Eco-Political Economy of AI: Environmental Costs and Regulatory Responses. Centre for Economic Policy Research.
20. European Commission. (2025). In Focus: Data Centres — an Energy-Hungry Challenge. Directorate-General for Energy.
21. S&P Global. (2025). Beneath the Surface: Water Stress in Data Centers. S&P Global Sustainable1.
22. Human Rights Research. (2025). The Human and Environmental Impact of Artificial Intelligence. Human Rights Research.
23. Carbon Credits. (2025). How AI and Clean Energy Are Competing for Critical Minerals. Carbon Credits.
24. Carbon Brief. (2025). AI: Five Charts That Put Data-Centre Energy Use — and Emissions — into Context. Carbon Brief.
25. Scientific American. (2025). AI Will Drive Doubling of Data Center Energy Demand by 2030. Scientific American.

Back to top