AI and climate change: solution or problem?
74% of Big Tech's AI climate claims are unproven

74% of Big Tech's claims that AI is good for the climate are unproven.
That's the finding of a February review of 154 claims by climate analyst Ketan Joshi, commissioned by groups including Beyond Fossil Fuels and Stand.earth (Beyond Fossil Fuels).
"AI will solve climate change" is a popular line. The evidence says AI and climate is a lot messier than that, and the honest answer sits somewhere awkward in the middle.
The consensus: AI will help solve climate change
The optimistic case isn't made up. The International Energy Agency's Energy and AI report says broad use of AI applications that already exist could cut energy-related emissions by around 5% in 2035.
Think grid balancing, smarter industrial processes, better forecasting. Mostly unglamorous stuff.
What the AI energy data actually says
Now the other column:
- Data centres used about 1.5% of the world's electricity in 2024, around 415 TWh (terawatt-hours), and the IEA expects that to more than double to roughly 945 TWh by 2030. That's slightly more than Japan uses today (IEA).
- Growth is already showing up. Data centre electricity use rose by around 17% in 2025 (IEA Global Energy Review), released this month. In the US, data centres made up about half of all growth in electricity use (Bloomberg Law).
- Company emissions are moving the wrong way. Microsoft says its total emissions are 23.4% above its 2020 baseline, citing AI and cloud expansion, while its energy use rose 168% (Microsoft).
- The benefit claims are thin. In Joshi's review, only 26% of claims cited published academic research and 36% cited no evidence at all. He found no case where consumer chatbots like ChatGPT, Gemini or Copilot delivered verifiable, substantial emissions cuts.
The trick he points to: blending old-school machine learning (which does help, e.g. forecasting) with generative AI (which is where the big energy growth is) so the benefits look bigger.
Where the "AI is a climate disaster" take goes wrong too
Honestly, the doom version oversimplifies as well.
→ Google says a median Gemini text prompt uses 0.24 Wh of energy and 0.26 mL of water (about five drops), and that energy per prompt fell 33x in a year (Google Cloud) → the IEA estimates data centre emissions rising from about 180 Mt of CO2 today to 300 Mt by 2035 in its base case, still small next to the whole energy system → but Google's numbers aren't independently verified, and tiny per-prompt costs times billions of prompts still add up
bad question: "is one AI prompt bad for the planet?" good question: "is total demand growing faster than clean power can keep up?"
The IEA's own line is the fairest summary: AI "is not a silver bullet", and rebound effects (efficiency making something cheaper, so people use more of it) could eat some of the benefits.
What uni won't teach you about AI and climate
Engineering and environmental degrees teach you to model systems. They rarely teach you to audit a corporate climate claim.
That's the same skill finance students need for AI stock hype, which we covered in AI investing: hype vs reality. Different industry, same move: ask for the evidence behind the headline.
The critical-thinking move: 5 questions for any AI climate claim
- Which AI? Traditional machine learning or a generative chatbot? They have very different footprints.
- Where's the evidence? A peer-reviewed study, a company blog, or nothing? Joshi found 36% cited nothing.
- Per unit or in total? A 33x efficiency gain means little if usage grows faster.
- Who measured it? Self-reported and unverified is a starting point, not a conclusion.
- What's the counterfactual? Would the emissions cut have happened anyway with a spreadsheet and good data?
What to do this month
- Read the IEA Energy and AI executive summary. It's free, short and the most quoted source in this debate.
- Pick one company climate claim about AI and trace it to its source. Note whether it's traditional ML or generative AI.
- Do the maths. Take Google's per-prompt figure and multiply it by a billion prompts a day. Compare it with a small town's daily electricity use.
- Watch for slop in this debate too. Viral stats about AI's water and energy use spread fast, often without a source, as we covered in our piece on AI slop. Check them.
More on the field in our energy and climate hub.
so TLDR: AI isn't saving the climate yet, and it isn't destroying it either. Judge every claim on evidence and totals, not vibes.
Which AI climate claim have you seen shared that you'd now want to check?
Sources
- Report exposes Big Tech's AI climate hoax: 74% of industry's claims about AI's climate benefits are unprovenBeyond Fossil Fuels · 17 FEB 2026
- Energy and AI: Executive summaryInternational Energy Agency · 10 APR 2025
- Global Energy Review 2026: Electricity demandInternational Energy Agency · 20 APR 2026
- Global 2025 Power Demand Rose as EV, Data Centers Grew, IEA SaysBloomberg Law · 20 APR 2026
- Measuring the environmental impact of AI inferenceGoogle Cloud · 21 AUG 2025
- Our 2025 Environmental Sustainability ReportMicrosoft · 29 MAY 2025
Get the next issue by email
theVANTAGEd is written by students to bring critical thinking to AI across real industries. One issue every two weeks.
