Category: Climate

Why Gains in Renewable Energy May Not Be Enough to Offset It

Like many people, I have thought about the climate impact of AI mainly in terms of data centres: their electricity use, cooling water, hardware and supporting infrastructure. These impacts matter, but a recent study by Alpine and colleagues (2026) helped me understand why they may not be the main thing to focus on.

The larger question is where AI is being used. AI can make an industry more efficient, lower its costs and expand what it can profitably produce. Its climate impact therefore differs greatly depending on whether it is applied to fossil fuels or renewable energy.

In renewables, AI can improve forecasting, maintenance, generation and grid integration, helping to avoid emissions. In fossil fuels, it can make exploration and extraction cheaper, bring previously uneconomic reserves into production and prolong the life of oil and gas fields. In other words, it can enable more fossil fuel to be extracted and burned.

The study models AI adoption in both sectors at comparable rates. Although its use in renewables helps avoid emissions, this benefit is outweighed by the additional emissions enabled when AI also makes fossil-fuel production more productive. Renewable productivity gains would need to be four to five times greater than fossil-fuel gains for the combined effect to break even (Alpine et al., 2026).

The most realistic response may not be to regulate AI itself. It may be to establish binding, progressively declining limits on fossil-fuel production. AI could still be used to detect methane leaks, prevent spills, improve safety and reduce operational emissions, but increased productivity could not simply lead to increased extraction.

What I take from the study is that AI’s climate impact depends not only on the resources needed to run it, but also on what we use it to make cheaper, faster and more profitable. Data centres remain part of the problem, but the more important question may be what we are asking AI to amplify.

Alpine, W., Geldner, N., Alpine, H., & Chepeliev, M. G. (2026). AI-driven productivity gains enable more CO₂ emissions than they avoid in a global energy–economy model. npj Climate Action, 5, Article 71.

https://doi.org/10.1038/s44168-026-00411-0

Both matter. They are not equally urgent.

There is growing concern about the environmental impact of AI data centres. It is an important concern, and one we should address. But it should not distract us from a far more urgent environmental challenge: the climate crisis.

The environmental impact of AI data centres is primarily an engineering problem. These facilities consume large amounts of electricity, generate significant heat, and in many cases require substantial quantities of water for cooling. Where freshwater is scarce, this can place additional pressure on local communities and ecosystems. New data centres should be powered by low-carbon electricity, designed for maximum efficiency, make greater use of reclaimed or non-potable water, and be located where they do not compete with local populations for limited freshwater resources. These are real environmental concerns, but they have identifiable engineering and planning solutions. There is every reason to believe AI infrastructure will become dramatically more efficient over time.

The climate crisis is different. It is not simply an engineering problem; it is a planetary systems problem. Its consequences are already escalating year after year: record-breaking heat waves, more destructive wildfires, longer droughts, heavier rainfall, stronger storms, rising sea levels, biodiversity loss, and growing pressures on food production, water supplies, and human health. Much of the damage already caused cannot be reversed on human timescales. Species lost to extinction will not return. Ice sheets and sea levels respond over centuries. Carbon dioxide remains in the atmosphere for generations. We cannot simply engineer our way back to the world we once had.

AI should be built responsibly, but its potential should also be recognized. It can advance science, improve healthcare, enrich culture, increase productivity, and help solve some of humanity’s most difficult problems—including the climate crisis itself through better energy systems, materials, forecasting, and optimization. The climate crisis, by contrast, is an extraordinarily difficult global challenge requiring coordinated action across energy, transportation, industry, agriculture, finance, and politics. It will take decades to address, and many of its consequences are already irreversible. In terms of environmental urgency, these are not comparable. If the environmental footprint of AI data centres is a 1, the climate crisis is closer to a 100. Both deserve attention. The climate crisis remains the defining environmental priority of our time.