The truth behind AI’s climate claims 

Artificial intelligence is hailed as an effective tool in the fight against climate change. But a new study shows that AI is currently doing more to boost oil and gas production than to cut emissions, argues Peter Freudenstein of AlgorithmWatch CH.

A new study suggests that AI use in the energy sector could result in up to 1.8 gigatonnes of additional net CO2 emissions per year, almost 3% of global annual CO2 emissions. Until now, evidence of the link between AI use and fossil fuel production has been mainly anecdotal: Amazon, Microsoft and Google, for example, each have their own website for their oil and gas customers. State-owned oil companies have no qualms about advertising how AI boosts their productionExternal link. The link between the use of AI in the energy sector and CO2 emissions has now been quantified for the first time in a journal published by Nature PortfolioExternal link. The study finds that while AI can help cut emissions by supporting the expansion of renewable energy, the additional emissions from increased oil and gas production outweigh these savings by far. 

It also exposes big tech firms’ marketing pitch selling AI is a miracle solution and a climate saviour as a bluff. The debate over AI’s climate impact has largely focused on data centres, but that’s only part of the picture. Just as important is what AI is used for and how it affects the climate.  

Indeed, the physical footprint of AI infrastructure in the form of ever-larger data centres is growing rapidly and needs to be managed. In Texas, for example, Amazon plans to build a data centre powered by 35 gas turbines for which the company has received an environmental permit allowing it to emit 33 million tonnes of CO2 per year. For comparison, Switzerland’s annual CO2 emissions amount to around 40 million tonnes. Local opposition to such centres is growing and has already had some effect. In the US, just ahead of this autumn’s midterm elections, politicians from both ends of the political spectrum are vying to take the strongest stance against data centres in their neighbourhoods. Switzerland has also seen its first protests against planned data centres including in Beringen in the canton of Schaffhausen. Calls for temporary moratoriums or at least for some time to consider whether and why a new data centre should be built are growing louder. This debate is extremely important.  

Just as important is the question of whether and how AI’s indirect impact is further exacerbating the climate crisis. This is where the concept of “enabled emissionsExternal link“ comes in. The term was coined by two former employees of Microsoft’s sustainability unit. It describes how technologies such as AI make the production of additional fossil fuels more efficient than it would be without AI. This has long been a reality.  

The fossil fuel industry was quick to recognise AI’s potential. A promotional video by the state-owned Norwegian oil company EquinorExternal link explains how AI helped discover 27 new oil and gas deposits last year, enabling the additional extraction of up to four billion barrels of oil. The CEO of Saudi Aramco told BloombergExternal link that AI has allowed the company to maintain extraction costs at $3 (CHF2.50) per barrel. These real-world examples make it clear: AI is helping the fossil fuel industry thrive.  

The tech industry is only too happy to meet this demand. AI companies may claim – and maybe even ensure – that their data centres are built more sustainably. Yet, their AI services can also reduce extraction costs of fossil fuels, drive up oil production and ultimately lead to significantly higher CO2 emissions overall. This aspect is not adequately accounted for in existing ESG (Environment, Social and Governance) standards or in climate and AI regulations.  

How do we tackle this? The usual response from politicians and industry is to pin their hopes on innovation and self-regulation. Even the Swiss government has left AI’s sustainability out of its plans for AI regulationExternal link, relying instead on “sector-specific codes of conduct”External link and non-binding measures. If self-regulation worked, companies and organisations would not be quietly watering down their own climate targets one after anotherExternal link.   

Apart from stricter approval procedures for data centres, tech firms need to be more transparent about their cooperation with the oil and gas industry. This means being honest about how their products impact the climate. Stricter legal disclosure and liability requirements for AI services used in oil and gas production could help ensure greater accountability. To put this into practice, global carbon accounting standards such as the GHG Protocol, the world’s leading standard, need to incorporate emissions caused by AI. This would also require investors to examine the CO2 impact of individual products when considering potential investments in the tech sector. 

Above all, we need to rethink how we use AI. Specialised, resource-efficient AI models can certainly help protect the climate, and we want to see more of these. But as long as AI is used to pump CO2External link into the atmosphere, its promise of sustainability remains little more than hot air. For tech companies that portray themselves as climate saviours, greater transparency is the bare minimum. More than that, a genuine commitment to climate protection would mean ending existing partnerships with the oil and gas industry. But at a time of increasingly frequent heat waves, and despite the scientific consensus that the transition away from fossil fuels must be accelerated, the trend currently seems to be towards “AI for oil and gas” rather than “AI for climate”. 

Edited by Veronica De Vore 

Adapted from German by Billi Bierling/ds