Is AI sustainable? A SOAS academic on climate, finance and the future of work

As AI expands rapidly, Dr Yannis Dafermos asks a critical question: Is artificial intelligence sustainable, or is it deepening environmental, economic and social challenges?

The rapid development and expansion of artificial intelligence have raised important questions about its impact on our world, the environment, the economy, and the workforce. As AI relies on vast physical infrastructures, complex financial systems and political choices, it shapes how the technology evolves – and who it serves. 

With the launch of the new MSc AI and Sustainability at SOAS, I consider the key question: ‘Is AI sustainable?' That is, is it exacerbating the environmental crisis, are we in an AI bubble and will traditional labour forces cease to exist?  

AI and the environment: Will the expansion of AI help to alleviate or only exacerbate the environmental crisis?

From an environmental point of view, the rapidly expanding AI infrastructure is exacerbating the ecological crisis.

The expanded AI physical infrastructure requires an increasing amount of electricity that, in several places, is mostly generated by fossil fuels, as well as large amounts of water that are necessary for cooling data centres.

The backlash against the green transition is making things worse: it delays the decarbonisation of the electricity generation that the data centres rely on to operate. This is very alarming.

The techno-optimist view that AI can solve the environmental crisis is highly problematic: it acts as an excuse to delay the systemic transformations that are essential.

However, there are expectations that AI systems will allow us to gradually attenuate our environmental footprint. AI technologies can, for example, improve energy efficiency in buildings, optimise heating systems, manage smart grids more efficiently, reduce waste in supply chains and enable precision agriculture that reduces water and fertiliser use. 

Will these AI-driven improvements in the use of natural resources be sufficient to offset the increasing environmental footprint of the AI physical infrastructure? This is highly uncertain. But irrespective of how optimistic or not we can be about the improvements in ecological efficiency that can be generated by AI technologies, the techno-optimist view that AI can solve the environmental crisis is highly problematic: it acts as an excuse to delay the systemic transformations that are essential for developing socio-economic systems that are consistent with the limits of a finite planet.

AI and economics: Are we repeating the economic mistakes of the dot-com bubble with AI?

There are clear similarities between the current AI boom and the dot-com bubble of the late 1990s: highly euphoric expectations about the transformative potential of a new technology; stock price over-valuations without profit materialisation; and too much investment compared to demand. 

But there are also significant differences: the technology leaders now are mega players with high profits mobilised to fund a significant part of the AI investment; and the financial system is now different, with shadow banking activities, such as private credit and securitisation, playing a key role in financing investment.

Financial investors are to some extent repeating the mistakes of the dotcom bubble by being over-confident about the profits that the investments in the new technology will generate. But will the AI investment lead to a financial bust? This largely depends on whether AI technologies will be adopted widely in the economy and lead to productivity gains in line with expectations. This would allow those who invest in the AI infrastructure to materialise sufficient profits consistent with optimistic asset price valuations. If this doesn’t happen, stock market prices will go down and there will be financial losses in a similar way to the dot-com bubble. 

On top of it, debt repayment problems can arise. As AI investment expands, there is an increasing reliance on debt and shadow banking. This does not refer so much to the companies that are at the core of the AI investment boom, such as Alphabet, Amazon, Meta, Microsoft and NVIDIA. It has mostly to do with other companies in the broader AI ecosystem, such as data centre developers and AI startups, as well as utilities that have been expanding electricity generation to meet the growing energy demand. These companies are more financially fragile and any failures to achieve anticipated profits can lead to defaults.      

AI and labour: Do you agree with the vision of an AI-driven society where traditional labour will cease to exist?   

Any vision that is centred around the role of a specific technology, putting aside political, social, ethical and environmental issues, is deeply problematic. Rather than pursuing abstract, techno-centric futures, our priority should be to develop societal visions that put social justice, well-being and environmental justice at their core. We should then explore how AI and other technologies can be developed in a way that can serve these visions.

Any vision that is centred around the role of a specific technology, putting aside political, social, ethical and environmental issues, is deeply problematic.

If we uncritically embrace the idea of an “AI-driven society”, we risk concentrating unprecedented political and economic power in the hands of the mega-corporations that claim they can deliver it. Then, instead of regulating AI to ensure that it serves societal and environmental purposes, we will give the space to these companies to expand their influence and entrench their infrastructure.

The large-scale robotics and AI systems required for such a vision would likely depend on vast amounts of energy, raw materials, and water - raising serious concerns about environmental sustainability. At the same time, these systems could reproduce or amplify existing biases and forms of exploitation embedded in data, labour supply chains, and colonial legacies. Rather than enhancing human flourishing, they might exacerbate social injustice, economic precarity, and ecological degradation.

Our vision should not be to organise societies around technology, but to shape technology around democratically-defined social and environmental priorities.