Artificial intelligence is making climate change worse, not better, according to new research from MIT that challenges the tech industry's optimistic narrative about AI's climate potential. A comprehensive framework developed by MIT Sloan researchers shows that while AI's direct emissions remain relatively small, the technology's indirect effects through economic growth could significantly increase global warming unless governments implement strict carbon pricing and emissions policies.

Why Doesn't Making AI More Efficient Help?

The intuition seems straightforward: if AI chips and data centers become more energy-efficient, they should produce fewer emissions. But MIT Sloan professor John Sterman and researcher Jennifer Turliuk discovered a counterintuitive problem called the "rebound effect." When technology becomes cheaper to run, demand for it increases, eroding the efficiency gains. Their analysis found that even a fourfold drop in AI's energy intensity yields negligible warming reduction without new climate policy.

Sterman uses electric vehicles as an illustration. "When I ask students and executives what they'd do with that extra money from lower fuel costs, they often say 'buy more stuff' and 'take my family to Disney World,'" he explained. "But the emissions from that trip can outweigh the reduction from your EV." The same logic applies to AI: as the technology becomes cheaper, companies and consumers use it more, canceling out efficiency improvements.

What's the Real Climate Risk from AI's Growth?

The bigger threat comes from what researchers call the "indirect rebound effect." AI is widely expected to boost productivity and accelerate economic growth globally. But faster economic growth means more energy consumption and more emissions across the entire economy. According to the MIT framework embedded in En-ROADS, a climate simulator developed by Climate Interactive and MIT Sloan, if AI eventually boosts gross world product by 25% (well below more bullish forecasts), expected warming in 2100 climbs from 3.3°C to 3.6°C. That additional 0.3°C warming would sharply worsen sea level rise, wildfire threats, extreme weather, crop losses, and other climate-related damage.

"Indirect rebound is the big issue. Individuals, corporations, and governments all want faster growth in their income, sales, and economy, and AI may help achieve those goals. But unless we implement policies that rapidly lower emissions economy-wide, any economic boost from AI will generate a lot more emissions, the climate will get notably worse, and that feeds back to harm the economy," said John Sterman, co-faculty director of the MIT Sloan Sustainability Initiative.

John Sterman, Co-Faculty Director, MIT Sloan Sustainability Initiative

AI's direct contribution to warming is estimated at roughly 0.1°C by 2100, which sounds manageable. But that calculation assumes renewables continue decarbonizing electricity production and efficiency improvements continue. The real problem emerges when you layer in economic growth driven by AI productivity gains.

How Are Companies Hiding AI's True Climate Impact?

Current greenhouse gas accounting practices allow corporations to report progress that the atmosphere never actually sees. One major tech firm recently claimed its new AI data center would run on 100% carbon-free power drawn from an existing nuclear plant. But here's the catch: the nuclear plant's output didn't grow, so every megawatt diverted to the data center requires a megawatt more electricity for households and businesses from elsewhere on the grid, where coal and gas supply nearly 60% of generation and almost all of the CO2 emissions.

"The atmosphere does not care about clever accounting," Sterman noted. "It responds only to the laws of physics." This accounting gap means many corporations are no longer on track for their net-zero targets, with some explicitly stating that AI expansion is responsible for missing their climate goals.

Steps to Address AI's Climate Problem

• Carbon Pricing: Implement a meaningful price on carbon across the economy to boost efficiency, incentivize green power, and focus more AI research on climate solutions rather than consumer engagement.
• Mandatory Disclosure: Require large corporations to disclose AI's environmental impacts using physics-based accounting that reflects actual emissions, not accounting tricks that hide grid-wide effects.
• Policy-Driven Efficiency: Promote energy efficiency standards, stimulate renewable energy deployment, and prevent AI greenwashing through regulatory oversight and transparent reporting requirements.

Can AI Still Help Solve Climate Change?

Researchers acknowledge that AI could eventually accelerate climate solutions through innovations in renewable energy, battery storage, and carbon capture. But two major obstacles stand in the way: focus and timing. Most AI today is not aimed at climate solutions; instead, much is deployed to keep people engaged with social media and buy more products. Meanwhile, AI raises emissions now, while any AI-enabled climate breakthroughs arrive after long delays for lab-to-pilot development, capital raising, and global scaling.

"AI-enabled innovations might eventually offset the additional warming from AI's earlier emissions," Sterman said. "But Greenland and Antarctica aren't going to magically re-freeze, and the forests burned, crops lost, businesses and homes destroyed along the way won't reappear".

The good news is that both problems are fixable, largely through policies needed to cut emissions economy-wide. Sterman argues that acting now is in business's interest because transparency can prevent costly AI regulation and moratoria. Additionally, economy-wide climate policies raise the benefits of AI by reducing the fires, flooding, and supply chain disruptions that hurt business and slow growth. "If we enact strong policies to cut emissions throughout the economy, it's still possible to stay below 2°C," Sterman concluded. "Harder, but still possible".

Sterman

The MIT framework is now available to the public through En-ROADS, allowing anyone to explore AI's climate impacts and test different policy scenarios. Rather than accepting tech industry promises at face value, policymakers and business leaders can use evidence-based tools to understand exactly how AI will affect the climate under different economic and regulatory conditions.