A peer-reviewed study published this week in Nature Climate Change delivers an uncomfortable truth that the AI industry has largely avoided confronting: artificial intelligence may be accelerating climate change rather than mitigating it. The research, which modeled economy-wide interactions between AI-driven productivity gains and competing energy pathways, found that under parallel adoption conditions, AI increased net global annual CO₂ emissions by 0.47–1.8 gigatonnes—equivalent to 1.2–4.8% of total global energy-related emissions in 2024.

The Core Finding: Enabled Emissions Dwarf Avoided Emissions

The study employed a computable general equilibrium (CGE) model to quantify AI's bidirectional role as both an enabler and reducer of emissions. When applied to fossil fuel extraction, refining, and power generation, AI-driven productivity gains produced "enabled" emissions that the researchers estimate at 0.6–2.4 Gt CO₂ annually—3.3–13x greater than the IEA's 2025 datacenter footprint estimates (0.18 Gt). Meanwhile, AI applications in renewables optimization produced "avoided" emissions that partially offset but failed to overcome this effect. The critical asymmetry emerges from upstream dynamics: small productivity improvements in fossil fuel extraction disproportionately impact the net calculation compared to gains in renewable generation. The researchers found that for every 1% marginal productivity gain in fossil fuels, renewables would need to achieve 4–5% gains just to break even on emissions—a ratio they describe as a "structural asymmetry" rooted in how AI reinforces existing energy economics rather than displacing them.

Why Carbon Pricing Doesn't Fix It

One might expect that higher carbon prices could correct this market failure. The study tested this assumption and found some relief but no salvation: under an $80/tCO₂ price, enabled emissions (1.0 Gt) still exceeded avoided emissions (0.3 Gt) by roughly 3x. Even at $308/tCO₂—far above current global averages—the gap narrowed to only 1.5x with a net increase of 0.1 Gt CO₂ annually. Carbon pricing "narrows but does not eliminate" the asymmetry, according to the paper.

The Rebound and Induction Effects Nobody Talks About

Beyond direct operational footprint analysis—which industry reports and ESG frameworks tend to emphasize—the researchers identified two indirect mechanisms that amplify AI's climate impact at a systemic level. Rebound effects occur when efficiency improvements stimulate additional consumption; induction effects arise when new technological options reshape production economics by altering capital, labor, and resource productivity across sectors. These second-order dynamics "may substantially exceed" the comparatively well-studied datacenter emissions, yet remain "largely unquantified" in existing frameworks.

The Governance Gap

The study identifies what it calls a "governance gap": policy, disclosure, and assessment frameworks built on narrow framings centered on operational footprints and avoided-emissions potential cannot address emissions pathways that their underlying analyses do not recognize. Prominent AI and climate governance frameworks currently leave these indirect system-level emissions-increasing effects outside their substantive scope—a structural blind spot with potentially massive implications for net-zero commitments.

Key Takeaways

  • AI-driven productivity gains increase net global CO₂ by 0.47–1.8 Gt annually under parallel adoption scenarios
  • Renewables must outperform fossil fuel productivity by 4-5x just to achieve emissions breakeven
  • Enabled emissions from fossil applications exceed IEA datacenter estimates by up to 13x
  • Carbon pricing at realistic levels cannot reverse the asymmetry—only extreme prices come close
  • Economy-wide rebound and induction effects likely dwarf AI's direct computational footprint

The Bottom Line

The AI industry's self-image as a green technology ally doesn't survive contact with rigorous economic modeling. Until governance frameworks account for economy-wide enabled emissions rather than just operational footprints, companies claiming climate benefits are likely counting savings from one bucket while ignoring costs drained from another—and that asymmetry is baked into the structure of how AI adoption actually works in practice.