World Environment Day 2026, today, 5 June, focuses on climate change, highlighting the urgency of sustained climate action. The clock is ticking towards the 2030 targets of the Paris Agreement on climate change but are we really winning the climate war and moving fast enough? In this moment of panic, some voices in Silicon Valley are offering a deus ex machina (a god from a machine): artificial intelligence (AI).
We live in an era of spectacular technological irony. We are told that machine learning is the missing puzzle piece that will improve our fractured energy grids, discover next-generation super-batteries and carbon-negative concrete and rewrite the rules of photosynthesis through precision agriculture. However, the reality is different.
Global data centres are projected to consume upwards of 1 050 terawatt-hours of electricity this year alone, a footprint rivalling the total energy consumption of industrialised nations like Japan. The very tool being counted on to decarbonise our civilisation is fast becoming one of the most power-hungry infrastructure networks on Earth. Yet dismissing AI as a climate villain is to miss one of the most consequential opportunities of the decade.
The question is not whether AI should be integrated into climate mitigation. It should. But integrating it must not be a Faustian bargain. Are we serious enough to deploy it wisely and with honest accounting? Otherwise, AI will simply automate our inefficiencies while consuming enough energy to melt the polar ice caps and burn down the planet we are trying to save. This is the great AI-climate paradox.
Let us be fair: AI’s potential for a warming planet is immense. Climate change is a problem of chaotic systems, massive datasets, fragmented teams and delayed feedback loops — exactly the kind of multivariable mess machine learning thrives on. AI is finally transitioning from experimental laboratory toys to operational workhorses. Machine learning models are enhancing the dispatch of renewable energy on grids that were never designed for the intermittency of solar and wind.
In South Africa, grid instability is a lived reality for millions. Predictive AI systems that can forecast demand spikes and smooth the integration of renewables are an infrastructure imperative, not a luxury. Germany’s Energiewende and India’s National Solar Mission share similar stories. For operators ditching fossil fuels, grid management has become a nightmare because the sun does not always shine and the wind does not always blow.
Advanced neural networks (AI models inspired by the structure of the human brain that learn from data) are changing the game by combining weather forecasts, satellite imagery and grid telemetry (real-time data sent from energy systems to operators to monitor the grid) to predict renewable output and demand with surgical precision.
Beyond the grid, generative AI models are simulating millions of molecular combinations in days and accelerating the discovery of sustainable alternatives for heavy industries. AI-powered precision farming is reducing the overuse of fertilisers. Route optimisation algorithms are quietly shaving millions of tonnes of COٖ2 from global freight. AI is also accelerating the modelling of weather and climate systems at resolutions that were computationally impossible just a few years ago.
These are not theoretical benefits. They are measurable, scalable and under way. Solely looking at the examples, AI’s capacity for large-scale optimisation is an indispensable weapon in our mitigation arsenal.
But there is a caveat: the Jevons paradox. In the 19th century, English economist and logician William Stanley Jevons observed that more efficient steam engines led to more coal consumption, not less. Efficiency lowered the cost of use, which exploded demand. We are close to repeating this cycle with AI because progress has a way of flattering us into complacency.
The race towards artificial general intelligence involves clusters of tens of thousands of graphics processing units running continuously for months, guzzling gigawatt-hours of electricity. In tech hubs like Frankfurt and Dublin, data centres are projected to consume staggering portions of electricity capacity.
By 2027, AI servers could consume as much electricity as Argentina or Sweden. Is using a carbon-heavy grid to power an AI that tells us how to reduce carbon solving a problem? Or are we simply moving emissions from the factory floor to the server rack? Just like bailing out a sinking ship with a bucket that has a hole the size of a cannonball, an algorithm reducing flaring by 10% but increasing hydrocarbon production volume by 15% is not a climate solution; it is greenwashing with a PhD in calculus.
What does genuine integration of AI into climate mitigation look like? Effective integration requires the following immediately:
Transparency: AI developers must disclose the energy consumption and carbon intensity of their systems with the same rigour expected from listed companies disclosing their Scope 1 (direct emissions) and Scope 2 (indirect electricity-related emissions) emissions. A smart thermostat saves 10 megawatt-hours (MWh) of electricity but developing the neural network running it consumes 100MWh. This has to stop.
Coordination: Stakeholders must work in genuinely integrated teams, not the present “AI for climate” silos. Models being built to predict flood risk in Lagos, Nigeria or deforestation in the Congo Basin in Central Africa are only as useful as the governance structure acting on their outputs.
Accountability: Closing the loop between AI-generated climate intelligence and measurable emissions reduction is also crucial. It is no use having an AI system that can identify inefficiencies in a municipal waste grid when the political will to fix them is lacking. Algorithms cannot be a substitute for governance.
Ultimately, the integration of AI into climate mitigation is a mirror reflecting our own psychology. We should be excited about the prospects of AI. But excitement is not a strategy. Neither is simply hoping for the best. Otherwise, we will arrive in 2050 — the year we are supposed to reach net zero — with perfectly optimised, hyper-efficient, AI-managed ruins, having built the smartest possible engine for our own extinction. The algorithm is ready. But are we?
Dr Blessing Afolayan is a postdoctoral research fellow at the School for Climate Studies and the National Institute for Theoretical and Computational Sciences at Stellenbosch University.
The very tool being counted on to decarbonise our civilisation is fast becoming one of the most power-hungry infrastructure networks on Earth. Yet dismissing AI as a climate villain is to miss one of the most consequential opportunities of the decade