The Irony of AI for Climate

The Irony of AI for Climate

AI is simultaneously:

Part of the problem:

  • Training runs consume megawatts
  • Inference at scale requires continuous power
  • Data centers use water, land, and grid capacity
  • Hardware manufacturing has its own environmental footprint

Part of the solution:

  • Climate modeling and prediction
  • Grid optimization and energy efficiency
  • Materials science for batteries and solar
  • Carbon capture optimization
  • Deforestation monitoring
  • Agricultural efficiency
  • Supply chain optimization

The same technology contributes to and helps address the climate crisis. Is the net positive or negative?

The Optimist Case

AI accelerates climate solutions faster than it adds to the problem:

  • Climate models that took months can run in hours
  • Materials discovery for batteries is dramatically accelerated
  • Grid optimization reduces energy waste at scale
  • Agricultural AI reduces land use and emissions
  • The knowledge generated has compounding returns

The energy consumed by AI is an investment that pays off in accelerated transition to sustainability.

The Pessimist Case

AI’s environmental benefit is overstated and its costs understated:

  • Climate models existed before AI; AI is incremental
  • Most AI use isn’t climate-related — it’s cat pictures and chatbots
  • Efficiency gains are offset by increased usage (rebound effect)
  • The infrastructure lock-in creates long-term energy demand
  • Benefits are speculative; costs are real and immediate

AI is a net negative dressed up in green marketing.

The Honest Answer

We don’t know. The accounting is hard:

  • Benefits are diffuse and hard to attribute (what would have happened without AI?)
  • Costs are real but not well-disclosed
  • Counterfactuals are impossible to verify
  • Different applications have different benefit-to-cost ratios
  • The same infrastructure serves both beneficial and frivolous uses

The irony resists resolution because the data doesn’t exist to resolve it.

The Accounting Problem

To answer “is AI net positive for climate?” we’d need:

  1. Total energy consumption of AI (training + inference + infrastructure)
  2. Carbon intensity of that energy by location
  3. Emissions enabled by AI use (not just direct consumption)
  4. Emissions avoided through AI-enabled efficiency
  5. Knowledge acceleration value (climate research faster)
  6. Counterfactuals (what would have happened without AI)

None of these is well-measured. Claims in either direction are poorly grounded.

Living with Irony

Given uncertainty, how should we reason?

Maximizing upside: Invest in AI for climate applications; accept the energy cost as worthwhile.

Minimizing downside: Constrain AI growth until net impact is clearer; prioritize low-impact applications.

Portfolio approach: Some AI use is clearly beneficial (climate research), some clearly harmful (pointless computation), most is ambiguous. Apply different standards to different uses.

Systemic focus: The problem isn’t AI; it’s the grid. Make the grid clean and AI’s direct impact becomes negligible. This deflects from AI-specific questions but may be most effective.

Open Questions

  • Is it possible to measure AI’s net climate impact?
  • Should climate benefit be a criterion for AI development priorities?
  • Is it hypocritical to use AI for climate work while ignoring its footprint?
  • Does focusing on AI’s impact distract from larger climate levers?

See Also