According to new research, by improving efficiency in all aspects of power generation and in society generally, artificial intelligence (AI) has the potential to reduce by up to 54 billion tonnes the annual global greenhouse emissions – due to burning fossil fuels like coal and gas – that are driving climate change, 10% of the present figure. This is after allowing for the huge quantity of energy needed to run AI computer software, widely reported as a threat to the environment. How realistic is the report’s claim, particularly since the great majority of power is still generated using fossil fuels?
Processing the quantity and variety of data needed to run AI computer software like Chat-GPT requires far more electricity than that used, for example, for task-specific computer modelling, such as to solve particular medical or scientific problems. This is because AI software is multi-purpose, ie it is designed to answer any question on any subject. Also, ‘training’ AI programmes before they can be used uses large amounts of energy. Training Open AI’s GPT-3 required enough electricity to power 120 average US households for one year, according to Carbon Brief. Recent research claimed that compared to task-specific software, AI uses up to 33 times more energy. An AI data processing centre could consume as much energy as a small town.
In April 2025, the Financial Times reported that fossil-fuel companies are hoping that surging AI energy demand will ‘usher in a new golden era’ for gas production. In a world where fossil fuels still dominate and where AI use is exploding, it clearly has the potential to add an extra dangerous twist to the present climate emergency.
Realising this, some of the multi-billionaires funding AI are proposing to use nuclear power as a power source, and bring back into commission nuclear plants previously moth-balled. It has just been announced that Three Mile Island, the site in the USA of a near nuclear catastrophe in 1979, is to be brought back into operation.
Nuclear power generation does not produce greenhouse gases, leading to its supporters giving it a ‘green’ label, but as has been discussed in Global Warning columns in Socialism Today, it is not a safe alternative.
Despite the future potential environmental danger, at the moment AI accounts for only a very small proportion of global electricity demand, in 2022 between 0.15% and 0.3%, so there is no need to resort to nuclear as a panic measure. However, these figures can only be approximate since information on AI is still fragmentary and provisional in a very rapidly developing and changing situation. For instance, since data centre power consumption statistics include AI and conventional task-specific software, the 10% proportion accounted for by AI is an estimate.
AI proponents say that the technology has the potential to reduce greenhouse gas emissions. The research published by the Grantham Research Institute on Climate Change at the London School of Economics, claims that by 2035 annual global emissions could be reduced, after allowing for increasing AI energy demand, by between 3.2 to 5.4 billion tonnes of carbon dioxide equivalent, ie by 6%-10% approximately by using AI. ‘Equivalent’ here means including the effects of other drivers of temperature rise, chiefly methane, apart from the main polluter, carbon dioxide.
If true this would be a very useful contribution in the fight against global warming. The authors claim that AI can improve the efficiency of renewable energy systems by improving electricity grid management and increasing the load factor, ie energy efficiency, of wind and solar by up to 20%. Also, improving the adoption rate of alternative proteins by up to 50% (admittedly in the report’s ‘highly ambitious’ AI scenario). This could be very significant, because protein obtained, for example, from livestock raising, meat and dairy, accounts for over 20% of greenhouse gas emissions.
Other possibilities highlighted include better predicting of investment risks and returns, improving financial decision making, innovating technology discovery and resource efficiency, and ‘nudging’ and behavioural change. Finally, generating insights and predictions around complex climate policy scenarios
The report makes an important point that as extreme weather events become more frequent and serious, accurate forecasting of, and adaptation to, such events will be crucial. More accurate early warning of extreme weather will save lives and costs through mitigating damage to housing and infrastructure. AI can play a significant role in this. Although conventional weather forecasting based on applying physical laws has improved enormously with modern computer power, because of the complexity involved there are limits to this progress.
By combining science-based and data-based AI models, significant improvements can be possible. AI can analyse vast quantities of complex data from multiple sources in a fraction of the time needed for conventional, science-based, forecasting, and it is particularly good at predicting very short-term weather, called ‘now-casting’. It can also produce more accurate local forecasts, allowing for the effects of micro-climates. Long-term forecasting can also be improved, such as predicting the speed and track of tropical storms that conventional forecasting can miss. This of course will be welcome as it can save lives.
The report has been criticised for underestimating the growth of AI use, and therefore overestimating the potential environmental AI gains. A more fundamental criticism, however, is that the premise behind its analysis and conclusions is the continuation of a failed capitalist market system. Providing capitalists, their policy makers and politicians, with improved tools for decision-making or with AI tweaked varieties of institutional economics, assumes the ruling classes have any intention of seriously tackling climate change.
One of the report’s contributors is quoted as saying: “Our research shows that with the right collaboration – between governments, tech companies and energy providers – AI can be harnessed to accelerate climate action, not hinder it”. However, for decades, in actions if not always in words, governments and the corporations they represent, have shown they have no intention of seriously collaborating to tackle global warming, confirmed again by the outcome of the recent UN climate summit in Brazil – COP30. In a capitalist market economy, it is utopian to think that AI will lead to a net reduction in greenhouse gas emissions. On the contrary AI power demands will most probably drive emissions up, perhaps significantly, since all the signs are that AI use will escalate very rapidly.
AI has the potential to increase efficiency in power generation and in the wider economy increase productivity, including labour productivity. But if this is attempted, as it is beginning to be, in a profit-driven market system, the result will be greater environmental pollution and major dislocation of the economy, leading to mass unemployment, particularly for white-collar workers. AI is being used positively to improve weather forecasting or in medical research for example, but its significant environmental benefits will only be realised, and in general society, mass unemployment avoided, with a democratically planned socialist economy.
Pete Dickenson