Mitigating AI’s climate costs

The escalating climate crisis and the rapid rise of artificial intelligence are set to reshape our world, transforming our political systems, economies and daily lives. What is often overlooked, however, is the myriad ways climate change and AI overlap and influence one another.

Much of the current debate sidesteps the complexity of this relationship. While techno-optimists tout AI as a panacea for the world’s problems, techno-skeptics highlight its environmental costs, warning that emerging technologies could end up exacerbating the crisis they aim to solve. In fact, AI has the potential to help us achieve critical climate targets and well-crafted policies can and must mitigate its environmental costs. Against this backdrop, French President Emmanuel Macron’s AI Action Summit in Paris next week offers a unique opportunity to chart a more sustainable course.

Silicon Valley technologists’ wildly optimistic predictions about AI’s potential role in tackling the climate crisis underscore the need for a nuanced approach. In September, for example, OpenAI CEO Sam Altman published an essay in which he envisioned a future where “nearly limitless intelligence and abundant energy” enable “astounding triumphs,” such as “fixing the climate,” “establishing a space colony” and unlocking the mysteries of physics.

A month later, former Google CEO Eric Schmidt echoed this sentiment, acknowledging AI’s climate impact but stating, “I’d rather bet on AI solving the problem than constraining it.” And tech giants like Google and Microsoft often highlight their use of AI to promote sustainability. Yet their carbon dioxide emissions have surged by 48 percent and 29 percent since 2019 and 2020, respectively, largely owing to the growing energy demand of massive data centers.

To be sure, the advances in data processing, mathematics and computation fueling the rise of AI could accelerate scientific research and enable us to tackle urgent global challenges. For example, specialized models like DeepMind’s AlphaFold have revolutionized our understanding of protein structures, with far-reaching implications for biological sciences.

But these advances, the result of years of interdisciplinary collaboration between AI researchers and scientists, are a far cry from an omniscient artificial general intelligence, or AGI, capable of instantly solving complex scientific and technological problems.

In fact, not only are we nowhere near a superintelligent AGI, but no one can predict when (or if) it will even become a possibility. Even if the most optimistic AI proponents are correct and an AGI emerges within the next five years, leading to breakthroughs like stable nuclear fusion or increased solar-cell efficiency, humanity would still have to contend with the messy economic and political realities of the clean-energy transition. After all, many of the technologies required to achieve net-zero emissions already exist. While they could be more efficient or affordable, the barriers to their deployment and scaling reflect the conflicting political, economic, financial and societal interests that shape the global geopolitical landscape.

Altman and other techno-optimists’ assurances that AI will solve the climate crisis gloss over these realities and relying on such promises is risky, given data centers’ massive carbon footprint. Already, data centers account for 2 percent to 4 percent of total electricity consumption in the US, the EU and China, and more than 20 percent in Ireland. The major offenders — Amazon, Apple, Microsoft, Google and Meta — obscure the full climate impact of their electricity consumption through creative accounting tricks like renewable energy certificates.

To be sure, the models recently released by the Chinese company DeepSeek have caused a stir precisely because they appear to be much more energy-efficient than their US counterparts. However, researchers have shown that increased efficiency, and therefore lower costs, may lead to greater demand for new AI functionalities — and higher overall energy consumption.

We should ensure that energy and resource-hungry data centers do not push the planet past environmental tipping points.

Sherif Elsayed-Ali

Instead of asking whether AI can help us achieve climate goals, we should ensure that energy and resource-hungry data centers do not push the planet past environmental tipping points before these technologies can deliver on their promises. To that end, governments must increase transparency and incentivize emissions cuts.

There are two immediate steps that policymakers can take to mitigate AI’s environmental impact. First, they should require the largest owners and operators of data centers to disclose location-based emissions data, reflecting actual electricity consumption. Such a requirement could eventually be expanded to all cloud-based services. Since AI and cloud services are typically billed based on resource usage and time, existing systems could be adapted to provide customers with detailed emissions reports.

Second, policymakers should adopt the International Monetary Fund’s recommendation to introduce a targeted tax on data center emissions. They should also consider requiring all new data centers to include renewable energy infrastructure, ensuring that their electricity demands do not place additional strain on power grids.

While these measures will not entirely offset data centers’ climate impact, they represent an important first step toward holding the tech industry accountable for its greenhouse gas emissions. By devising effective policies, we can cut through Silicon Valley’s hype and transform AI into a net positive development for the environment. Only with a stable climate can humanity afford the luxury of pursuing AI pioneers’ grand ambitions.

• Sherif Elsayed-Ali, Executive Director of the Future of Technology Institute, is a co-founder and former CEO of Carbon Re. Copyright: Project Syndicate