Does Greater AI Productivity Increase Carbon Emissions? Beyond Data Centers

AI as a bidirectional productivity amplifier

AI is characterised as a "bidirectional productivity amplifier". On the green side, AI improves wind and solar forecasting, optimises grid dispatch, and raises equipment utilisation, thereby increasing the effective output of renewable infrastructure. On the fossil side, AI is applied to exploration, extraction, processing, transport and power generation, potentially lowering the cost of marginal oil and gas resources and making previously uneconomic reserves viable.

Global computable general equilibrium (CGE) model

The authors built a global CGE model that links industries, energy prices, capital, labour and consumer demand. The model uses a 2017 global economic database and is adjusted to reflect the 2024 power‑mix. It is a static comparative‑equilibrium framework intended to assess the direction and magnitude of AI‑driven productivity shocks rather than to produce year‑by‑year forecasts.

Sixty‑four scenarios are simulated, varying the intensity of AI applications from low to high across fossil‑fuel sectors, renewable‑energy sectors, grid operations, shipping and heavy industry.

Scenario results

When AI‑driven productivity gains are applied simultaneously to fossil‑fuel and clean‑energy sectors, the model predicts an increase in integrated annual CO₂ emissions of 0.47 × 10⁸ t to 1.8 × 10⁸ t, equivalent to roughly 1.2 %–4.8 % of global energy‑related emissions.

The dominant driver of the increase is the uplift in fossil‑energy productivity. Renewable‑energy efficiency gains offset only a portion of the rise. The model quantifies that a 1 % increase in fossil‑fuel productivity would require a 4 %–5 % increase in renewable productivity to neutralise the additional emissions.

Interpretation and model limitations

The static comparative‑equilibrium approach holds post‑2017 energy prices, policies, technology costs and industry structures constant, except for the updated power‑mix. Consequently, the results highlight a potential directional effect of AI on the energy system rather than precise future emission trajectories.

Policy implications

The authors outline five priority actions, including restricting AI tools in fossil‑fuel extraction, expanding systematic tracking of AI’s climate impact, and coupling support for green AI with measures that limit fossil‑fuel expansion.

Paper link: https://www.nature.com/articles/s44168-026-00411-0

Code example

来源：ScienceAI
本文
约2000字
，建议阅读
5
分钟
AI 不只是自己消耗能源，它还会改变能源产业本身，
而这可能才是更大的变量。