One of the breeziest promises made about artificial intelligence is that it will enable us to tackle the world’s biggest challenges such as climate change. AI can help run smarter electricity grids, design more efficient electric vehicles and track plastic pollution in our oceans. But the data centres that host the latest AI models consume shocking amounts of energy and water. Is AI more of a problem than a solution when it comes to the climate emergency?
The ways in which technology is tugging in opposite directions is highlighted by the experience of Microsoft, which in 2020 made one of the boldest environmental commitments in corporate history. By 2030, the technology company promised, it would be carbon negative and by 2050 it would have offset all the emissions it had generated since its birth in 1975. But Microsoft still has a long way to go. Last month, it reported that its emissions had risen 29 per cent since 2020 as it continued to invest massively in data infrastructure.
This week, the company announced it was investing $3.2bn over the next two years to expand its cloud computing infrastructure in Sweden. In total, Microsoft intends to spend more than $50bn this year on data centres in what one analyst has called “the largest infrastructure buildout that humanity has ever seen”. The data consultancy Gartner forecasts that worldwide spending on data centres will rise 10 per cent this year to $260bn.
Further out, Microsoft and OpenAI, the AI start-up it is heavily backing, are also planning to invest as much as $100bn on building a US-based supercomputer and data centre, according to The Information.
The supersized data infrastructure needed to run energy-hungry generative AI models is already causing environmental damage. As Microsoft noted in its latest sustainability report, the infrastructure and electricity required to power the latest technologies create “new challenges for meeting sustainability commitments across the tech sector”. Although the company remained upbeat that it would hit its long-term goals of becoming carbon negative, water positive and zero waste, it acknowledged it was not yet on track in reducing indirect emissions and replenishing more water than is used in its data centres.
Researchers are working hard to make AI models do more with less energy, which clearly has big financial as well as environmental appeal. As part of its green commitment, Microsoft is also spending heavily on renewable energy. Last month, the corporate giant committed to bringing 10.5 gigawatts of renewable energy online in the US and Europe in partnership with Brookfield Asset Management. The additional capacity, equivalent to powering 1.8mn homes, will cost an estimated $10bn.
The company is making some wilder bets on small nuclear modular reactors to produce carbon-free energy. It has also agreed to buy electricity generated by nuclear fusion from Helion Energy by 2028 too (assuming it can be produced by then). Helion is a lavishly funded start-up backed by Sam Altman, OpenAI’s chief executive.
The big AI companies argue that, in the global scheme of things, the energy demands of the digital economy remain comparatively small. According to the International Energy Agency, data centres account for less than 1.5 per cent of worldwide electricity use. But what is unnerving is how fast they are growing. The IEA forecasts that by 2026, data centres might be consuming 1,000 terawatt hours of electricity, more than twice the level in 2022 and about the same usage as Japan.
Given that the environmental downsides of AI are so glaring, it is all the more imperative to capture the upsides. One of the most intriguing fields is weather prediction, which can help us adapt to climate change. In November, researchers at Google DeepMind unveiled GraphCast, an AI model that is more accurate (and more energy efficient) at making 10-day weather forecasts than conventional methods.
As more countries consider geoengineering responses to climate change, such as cloud seeding and carbon capture, AI researchers are also focusing on modelling the possible impacts. “The national security aspects of weather are a huge topic right now,” the chief executive of one AI company tells me. “People assume that geoengineering is going to happen and they want to know what the effects are going to be.”
For the moment, the environmental costs of AI are real while the benefits remain more nebulous. It’s time for the industry to deliver more on its ambitious promises.