The environmental cost of AI
MANY people have seen enough of artificial intelligence by now to appreciate its potential benefits to humanity, as well as its risks. Not so many may be aware of what a big guzzler of natural resources it is. As the Financial Times reported this week, Big Tech companies are using ever more water to cool their energy-intensive data centres. The thirst for power and water of the server farms needed to run generative AI is even more intense. In Davos last month, Sam Altman, the OpenAI CEO, warned that future AI would consume so much electricity it would require an energy breakthrough — say, nuclear fusion — to power it.
There is an irony here, given tech companies’ image as shiny, clean replacements for old smokestack industries — and the pledges of Microsoft and others to be good climate citizens. The International Energy Agency says data centres, cryptocurrencies and AI accounted for almost 2 per cent of global power demand in 2022 — and this could double by 2026 to nearly match the electricity consumption of Japan. Ireland, a favoured location for server farms, is now limiting new data centre connections to the power grid; others are exploring whether to do the same.
AI has a stronger case than crypto — already well-known as a prodigious energy user — that its potential for social good can justify the power consumed. But in return for access to the grids, generation and scarce water resources they need, tech companies should accelerate the use of AI to assist the green transition. The technology has been touted as allowing more sophisticated weather forecasting — helping to balance power demand — and emissions tracking. It might help to find clever ways of cutting energy use in farming, manufacturing, supply chains and office buildings.
Google reduced energy use for cooling by 40 per cent at one of its data centres by applying machine learning from DeepMind, its AI arm, to forecasting variables such as demand and weather. It is also using software to seek out parts of the world with excess solar and wind power to site data centre operations.
The tech companies behind many of the world’s largest data complexes — Microsoft, Google and Amazon — have already been investing in solar and wind generation, and re-engineering data centres to conserve water. But, especially as generative AI takes off, they will have to do more to ensure they can harness power in a climate-neutral way. That means putting money, too, into energy innovation. OpenAI’s Altman has invested in Helion, a US fusion power start-up, and Microsoft has signed up to be a future Helion customer.
The tech world also needs to be much more open about its use of power and water and efforts to curb it; researchers complain data is hard to come by. Pressure is already growing on business, including tech companies, to disclose environmental impacts beyond carbon emissions. The Taskforce on Nature-related Financial Disclosures has been developing reporting systems that would track companies’ impacts on nature, including water usage.
Such voluntary frameworks have been criticised for being slow to have an effect. But legislative efforts are expanding, too. A recent Democratic bill in the US would create standards for assessing the specific environmental impact of AI, though this may not pass. The EU’s Corporate Sustainability Reporting Directive now requires data centre operators to disclose their environmental impact. As with other areas of regulation, it is in tech companies’ interests to take their own initiatives, rather than wait to be forced. AI could have a useful role in tackling climate change. But it needs to ensure it is seen as part of the solution — and not part of the problem.