Artificial intelligence (AI) has a growing branding problem. As with Bitcoin, it has become increasingly clear that the technology is an energy hog. Creating AI models, generating answers and building the data centres consumes copious amounts of electricity – much of which ultimately comes from fossil fuels.
To put this in context, responding to a ChatGPT query requires close to 10 times as much electricity as a Google search, according to a recent estimate by Goldman Sachs. Meanwhile, generating a single AI model can gobble up more electricity in a year than 120 US homes and put out more carbon emissions than 110 cars, according to a calculation by Bloomberg.
Most normal companies won’t see anything close to the emissions increases experienced by the likes of Microsoft and Google
The issue hit the headlines recently after Microsoft revealed that its greenhouse gas (GHG) emissions had climbed almost 30% last year, driven higher by the greedy energy demands of developing AI. Notably, the overwhelming majority of the increase came from so-called Scope 3 emissions, arising from the company’s supply chain, including external data centres used to power AI. Microsoft’s Scope 1 and 2 emissions – from Microsoft’s own operations and the electricity it buys – actually fell by around 6%.
Other AI pioneers are suffering a similar embarrassment. Google’s GHG emissions last year were almost 50% higher than in 2019.
Ripple effect
This raises several key issues for companies well beyond the tech sector. As AI tools become important for business operations – in everything from streamlining processes to improving customer service – will more firms end up reporting a surge in emissions?
The answer is far from straightforward. Unlike Bitcoin, AI can also be part of the solution. It can also be used to help companies pin down emissions arising from their supply chains – a complex and time-consuming task that only about 25% companies manage, according to a 2022 study by the Carbon Disclosure Project.
For a start, outside a few tech giants, most normal companies won’t see anything close to the emissions increases experienced by the likes of Microsoft and Google – which have been pioneering these technologies. And although the electric consumption from AI is rising fast, it is still coming from a relatively low base. Data centres currently account for less than 2% of global energy use. Goldman Sachs expects this to reach 4% by the end of the decade – a headwind for emissions, to be sure, but not a ‘make-or-break’ issue for climate change.
Generative AI is being harnessed in a wide range of ways to reduce emissions.
And Microsoft co-founder Bill Gates has stressed that top tech companies are committed to minimising the headwind. ‘There are lots of ways to make the AI computations more efficient, so they are not using as much energy,’ he told Bloomberg television. Companies like Microsoft, he added, were willing to pay a premium for green energy, and that AI demand would push green technology forward. Asked if AI would end up paying for itself in terms of emissions, he answered ‘absolutely’.
Reducing emissions
This optimism may seem self-serving. But it also not without justification. Even at the early stages of its development, generative AI is being harnessed in a wide range of ways to reduce emissions. One major opportunity is in improving energy efficiency in commercial buildings. German’s Aedifion, based in Köln, is using AI to help companies monitor every aspect of energy consumption in real time and identify potential savings, driving down emissions by up to 40. AI is also being deployed to monitor other forms of ecological destruction – including deforestation – making it easier for authorities to clamp down on illegal practices. Other use cases include improving agricultural efficiency and cutting down waste in industrial processes.
It is already realistic to expect AI to significantly improve our understanding of ecological problems
Gates also makes the case that more powerful second-level effects could be on the way, if AI ends up pioneering advances in material science, such as devising ways to make solar panels without silicon or grow food more effectively in labs.
However, even if such moonshot projects never come to fruition, it is already realistic to expect AI to significantly improve our understanding of ecological problems – which is the first step towards addressing them. Scientists are using AI to track Antarctic icebergs 10,000 times faster than humans could.
Companies should also be able to achieve a far more accurate vision of the impact their operations have – including through their supply chains. As mentioned above, calculating Scope 3 emissions is typically both onerous and often imprecise, especially for businesses that source goods and services from a wide range of suppliers, explaining why most small companies can’t manage it.
Getting it right
Yet failure to get this right can cause companies to massively misjudge their own ecological footprint. A recent report by the Carbon Disclosure Project and Boston Consulting Group found that Scope 3 emissions are typically 26 times larger than the emissions arising directly from the operations of a business – Scope 1 and 2. This can also range hugely from industry to industry, with supply chain emissions from the retail sector around 92 times their operation total. IBM has argued that AI can provide the answer and has been exploring ways to streamline Scope 3 calculations, shifting through mountains of data to help businesses pin down where emissions are being generated.
It should make it easier for investors and other stakeholders to hold executives accountable
As ever, knowledge should be power. Once companies can accurately identify the quantity and source of emissions, they should be better able to cut their footprint. Such technology should also make it easier for investors and other stakeholders to hold executives accountable.
The bottom line is that whether AI turns out to be an eco-villain or saviour, or more likely something in between, remains to be seen. The outcome will depend on the willingness of tech companies to ensure that the huge power demands of AI are met increasingly from renewable energy. It will also hinge on whether companies are willing to deploy their AI spend not just on improving their bottom lines, but also on reducing their ecological footprint.
