Artificial Intelligence is often portrayed as a silver bullet in the race to combat climate change. From real-time deforestation monitoring to optimising energy grids and transforming agriculture, AI promises to make systems smarter, faster, and greener.

But behind the glossy headlines and tech conference keynotes lies a more complex reality: can AI itself be sustainable? Or are we simply accelerating environmental degradation under the guise of progress?

The Growth of AI and Its Hidden Cost

Let’s start with the scale of the issue… According to the International Energy Agency (IEA), global data centre electricity consumption is expected to double by 2026, driven largely by AI workloads. AI-related electricity demand is projected to jump from 460 TWh in 2022 to over 1000 TWh by 2026 - that’s more than Japan’s total electricity consumption today.

And training large AI models isn’t cheap in energy terms:

• A 2022 study from the University of Massachusetts Amherst found that training one large NLP model can emit over 626,000 pounds of CO₂ - equivalent to five round-trip flights between New York and San Francisco.
• OpenAI’s GPT-3 model was estimated to require 1,287 MWh of electricity to train, enough to power an average UK home for over 130 years.

This is before we even consider the energy used to run these models every day across millions of devices, or the supply chain emissions tied to manufacturing GPUs, chips, and data centre infrastructure.

The Case for AI as a Sustainability Enabler

Despite the concerns, AI can still be part of the solution - if used wisely. Some compelling examples:

• Google’s DeepMind helped reduce the energy used for cooling its data centres by up to 40%, simply by applying AI-driven optimisation.
• AI-powered precision agriculture is reducing fertiliser and pesticide use by up to 20%, according to a 2023 report by McKinsey.
• Smart logistics algorithms have cut fuel consumption in commercial fleets by 10–15%, according to the European Commission.

In theory, AI can help:

• Forecast demand and reduce energy waste.
• Improve public transport networks and reduce traffic.
• Monitor carbon emissions from factories in real time.
• Drive automation in renewable energy generation and distribution.

But the problem lies in how it’s deployed and who benefits.

Efficiency ≠ Sustainability: The Rebound Effect

The concept of the rebound effect (also known as Jevons Paradox) highlights a key challenge: as technologies become more efficient, they’re often used more, not less.

For example:

• The more efficient our cars became, the more we drove.
• The more powerful our smartphones became, the more devices we bought (many of us now own phones, tablets, laptops, smartwatches - all powered by AI in some form).

The same applies to AI. As models become faster and cheaper to run, adoption will scale rapidly, embedding AI into everything from fridges to city infrastructure. This can lead to net increases in energy use, materials demand, and emissions.

As Mike Berners-Lee points out in How Bad Are Bananas?, efficiency alone rarely delivers carbon savings unless paired with systemic regulation and behavioural change.

What Needs to Happen for AI to Be Truly Sustainable?

To ensure AI supports, rather than undermines, our sustainability goals, we need a conscious shift in how it’s developed, deployed, and governed.

1. Green AI by Design: Tech companies must prioritise energy-efficient algorithms, smaller training datasets, and low-impact deployment. Emerging practices like model quantisation and pruning can reduce compute power by up to 90%, according to Stanford’s AI Index 2024.

2. Clean Energy Infrastructure: Running AI on clean power should become the norm, not the exception. Hyperscalers like Microsoft and Google are investing in renewable-powered data centres, but this needs to become industry standard.

3. Transparent Reporting: Organisations should disclose the carbon cost of training and running AI models, just like they report financial metrics. Initiatives like the Green Software Foundation are pushing for better standards in this area.

4. Ethical & Equitable Deployment: Sustainability isn’t just environmental, it’s social. AI should be deployed to reduce inequality, not deepen it. That means resisting use cases that exploit workers or prioritise profit over people

AI holds immense promise in helping us transition to a low-carbon economy. But right now, it risks becoming both the arsonist and the firefighter, solving problems it also contributes to. To change that, we must move beyond the idea that efficiency alone is enough. We need transparency, ethical intent, and global standards that ensure AI is a force for good, not just a profitable one.

The question isn't whether AI can be sustainable. It's whether we're willing to do the work to make it so.