AI is often discussed as an environmental cost, because of the energy used by data centers. But machine learning is also a tool for climate action. A widely cited 2019 paper, Tackling Climate Change with Machine Learning, written by a broad group of researchers and associated with the organization Climate Change AI, mapped dozens of areas where machine learning can help reduce emissions and adapt to a changing climate. Here are some of the most promising.

Weather and climate forecasting

AI weather models have advanced rapidly. Google DeepMind’s GraphCast, published in Science in 2023, produced 10-day global forecasts that outperformed the leading conventional system on most of the evaluated targets, while running in about a minute on a single machine. Other groups and weather agencies have developed similar models, and national forecasters increasingly use machine learning alongside physics-based models.

Better forecasts help with early warnings of extreme weather, and with planning renewable energy, which depends on wind and sunshine.

Electricity grids

As grids integrate more solar and wind power, balancing supply and demand becomes harder. Machine learning helps by:

  • forecasting renewable output and electricity demand;
  • optimizing battery storage and demand response;
  • detecting faults and planning maintenance;
  • improving the efficiency of data centers and buildings.

Materials and chemistry

Discovering new materials for batteries, solar cells, catalysts and carbon capture traditionally takes years of experiments. AI models can screen enormous numbers of candidate materials computationally and suggest promising ones to test in the lab. Google DeepMind’s GNoME project, for example, reported predicting the stability of hundreds of thousands of new crystal structures.

Monitoring emissions and ecosystems

Satellite imagery combined with machine learning can track deforestation, methane leaks, power plant activity, crop health and illegal fishing. Independent monitoring makes emissions more transparent and verifiable.

Industry, transport and buildings

  • optimizing industrial processes to use less energy;
  • improving logistics routes and vehicle efficiency;
  • smart building controls for heating and cooling.

Adaptation and resilience

  • flood and wildfire risk mapping;
  • agricultural advice for changing growing conditions;
  • planning infrastructure for heat and sea-level rise.

The caveats

  • AI’s own footprint: the energy and water used to train and run models must be weighed against the benefits. See how much energy AI uses.
  • Rebound effects: efficiency gains can increase consumption.
  • Dual use: the same optimization tools can make fossil fuel extraction more efficient.
  • Data gaps: many regions most affected by climate change have the least data.

Used deliberately, AI can be a meaningful climate tool — but it is a complement to policy and investment, not a substitute.

Balancing benefits and costs

AI’s climate benefits — better forecasts, optimized grids, efficient buildings, faster materials research — must be weighed against its own energy and water use. The balance depends on what AI is used for and how data centers are powered. Applications that cut emissions directly, run on clean energy and use efficient models deliver the clearest net benefit; uses that add demand without reducing emissions elsewhere increase the footprint.

Related: measuring AI’s carbon and water footprint and AI for social good.

Bottom line

AI can be a powerful climate tool when it is aimed at real emissions reductions and runs on clean, efficient infrastructure. Judge each application by its net effect.

What to watch next

  • Operational adoption of AI weather models by national meteorological services.
  • AI-discovered materials moving from prediction to lab validation and production.
  • Satellite-based emissions monitoring used in policy and enforcement.
  • Measurement standards for AI’s own environmental footprint.

Key takeaways

  • Machine learning can help reduce emissions and adapt to climate change across many sectors.
  • AI weather models such as GraphCast rival conventional forecasts at much lower computational cost.
  • AI helps balance grids with more renewables and accelerates materials discovery.
  • Satellite AI monitoring makes emissions and deforestation more transparent.
  • Benefits must be weighed against AI’s own energy use, rebound effects and dual use.

Frequently asked questions

Is AI good or bad for the climate?

Both effects exist: data centers consume energy, while AI applications can reduce emissions elsewhere. The net effect depends on how AI is used and powered.

Are AI weather forecasts used in practice?

Increasingly. Weather agencies and companies run machine learning models alongside traditional physics-based forecasts.

How can I learn more about AI for climate?

Organizations such as Climate Change AI publish research summaries, courses and community resources.