Connecting business research with policy, practice and public debate Connecting business research with policy, practice and public debate Does the potential of AI to…
Connecting business research with policy, practice and public debate Connecting business research with policy, practice and public debate Does the potential of AI to tackle climate change outweigh its environmental footprint? During the 2026 United Nations Climate Change Conference (COP31), we asked five academics and experts for their views on technology and the climate. Machine learningcan help tackleclimate change, for example by forecasting wind and solar output or detecting methane leaks. But industry uses such applications toadvertise AI’s green credentials, while also embarking on a colossal data centre expansion. This expansion is drivenlargely bygenerative AI, particularly large language models, which show little potential to reduce emissions. This has enormous environmental and social impacts. Data centres consume considerableelectricity and water,compete withhousing deliveryand disproportionately harm marginalisedcommunities. Industry estimates suggest they use6 per cent of electricityin the United Kingdom and United States.Global energy demandis set to double by 2030, fuelling newgas-firedpower plantsthat underminecorporate net-zero targetsand threatennational ones. The AI boom also risks eroding the social and political conditions needed to address climate change.Trillions of dollarsin AI infrastructure spending creates incentives to embed generative AI everywhere, from work and education topublicservicesandpersonalrelationships, whether or not it improves our lives. Underlying this expansion is a vision oflimitless growthand automation, with technology supposedly overcoming ecological limits. Tech leaders promise AI will“eradicate poverty”and usher in“massive prosperity”. In reality, AI is built onastonishing theftand fuelsmisinformation,cyberattacksandmass surveillancewhile threatening cognitive deskilling and posingpotentiallycatastrophic risks. If humans won’t, as Bill Gates claims, be needed“for most things”, mass unemployment and unprecedented concentration of wealth and power could follow under existing ownership structures. At the ideological extreme, whoseinfluencehasgrown, sustainability is viewed as anenemy of progressand freedom is declaredincompatible with democracy. None of this bodes well for collective climate action. A new front has thus opened in the struggle for a liveable Earth. But this is also bringing new opportunities. With theAIbacklashrisingand the Democrats’blue tsunami in the US midterms, recall Milton Friedman’s observation that only a crisis – actual or perceived – produces real change. If, and when, the AI bubble bursts, we must resist austerity and corporate bailouts and seize the opportunity todemocratise AI, including throughpublic ownership, and put it to work for a good and sustainable life for all. Whether artificial intelligence is a threat or a solution to climate change ultimately depends on the application. People tend to forget that “AI” is an umbrella term covering a wide range of technologies and approaches, from basic machine learning models to the most advanced large language models. Even “generative AI” is still an umbrella term encompassing anything from text to image and video generation. In particular, this category of AI applications has been driving up demand for data centre infrastructure in recent years. This technology thrives on a “bigger is better” dynamic, where larger models built with more data and computing resources can perform better, thereby driving adoption and encouraging the development of even bigger models. Efficiency improvements may reduce resource use per task, but this does not necessarily mean less resource use overall. This dynamic is fundamentally at odds with environmental sustainability, which requires reducing resource consumption rather than continually expanding it. Between 2022 and 2025, spanning the generative AI boom triggered by ChatGPT’s launch, Google and Microsoft each more than doubled their total electricity consumption. Where this additional demand is met by fossil fuels, it leads tomore carbon emissions. Power generation and data centre cooling also contribute to water consumption. Tech companies repeatedly argue that the future benefits of AI will outweigh these costs. But the applications that bring those benefits are not necessarily large language models. A relatively simple and low-cost machine learning model may deliver greater environmental benefits than these massive models ever could. Benefits from one application cannot simply justify the environmental costs of another. We also need to consider whether the same results could be achieved without AI. Showing that an application has environmental benefits is not enough: those benefits need to outweigh its costs and be compared with less resource-intensive alternatives. At the moment, tech companies are not providing the granular data required to make these assessments at an application level. Without this transparency, we cannot properly assess the pros and cons of individual applications, let alone conclude that the environmental benefits of the current AI boom will ultimately outweigh its costs. Is AI a threat or a solution to climate change? Both, and which one it becomes depends on us. The threat is real. The International Energy Agency forecasts thatdata-centre electricity demandis set to more than double by 2030, to around 945 TWh, roughly Japan’s consumption today. More broadly, our econometric work on the 23 countries leading AI, including global south economies, shows that more AI spending means more energy use, more carbon dioxide emissions and, at first, a lower renewable share. The good news is that the relationship is not linear. We find a“green AI Kuznets curve”: above 600, output per tonne of CO2and the renewable share rise. AI can offset its emissions by greening industries, cities and itself, and could find new ways to fight climate change and preserve biodiversity. The question is who gets there? Today, only Singapore and the United States. The global south supplies much of the minerals, water and clean energy AI needs, but for most of it, even $200 per person is unaffordable, and its position in value chains is limited. Yet the threshold is not fixed. WhatAI is used formatters as much as how much is spent. The right composition can lower the threshold and shift the curve. Ahead of COP31, I’d call for “Green AI Valleys” in the global south, built on four I’s: Industrial AI: steer AI toward grids, agriculture, mining, manufacturing and climate, moving up the minerals/energy/compute value chain rather than just hosting data centres. Integration: regional co-ordination among willing countries and companies, with a negotiated division of tasks, can turn competition into complementary hubs and corridors. Inclusion: invest in skills, local startups and finance, so value and jobs powered by AI infrastructure stay at home. Nature’s Intelligence: use biodiversity as a bank of ideas, from nature-inspired cooling to circular heat reuse and seed-shaped wind turbines, moving AI from resource-intensive to symbiotic. Governed this way, AI’s energy hunger becomes a green window of opportunity and a key climate solution. Left alone, the threats will dominate. Artificial intelligence (AI) applications could drive key climate solutions able to save energy, reduce greenhouse gas (GHG) emissions and accelerate clean tech innovation. At the same time, AI is energy intensive and, if its expansion is left unmanaged, it could heighten demand for electricity, resulting in raised electricity prices, and increase pressure on water systems. It could also lead to higher GHG emissions if renewable energy capacity does not expand fast enough to meet the additional electricity demand, resulting in new demand being met by fossil fuels instead. In the energy sector, AI is already contributing to emissions reductions by i