74% of the industry’s climate promises lack evidence. The most-cited figure traces back to a Google blog post. And no one can point to a single example where ChatGPT actually reduced emissions.
Microsoft increased its emissions by 23.4% since 2020. Google by 48% since 2019. Both companies have net-zero targets for 2030.
And both tell the same story: Yes, AI consumes more energy. But AI will solve the climate problem. The benefits will outweigh the costs.
A new report now shows: This narrative is built on almost nothing.
The Numbers
Climate and energy analyst Ketan Joshi, commissioned by a coalition of environmental organizations including Beyond Fossil Fuels, Friends of the Earth, and the Green Web Foundation, analyzed 154 specific claims that AI will benefit the climate.
The findings:
- 74% of claims lack verifiable evidence
- Only 26% cite peer-reviewed academic research
- 36% cite no evidence at all
- 29% reference corporate publications—which themselves mostly contain no primary evidence
And most importantly: The report found zero verified examples where ChatGPT, Gemini, or Copilot led to substantial, measurable emissions reductions.
Not one.
The Bait-and-Switch
The analysis reveals a systematic conflation that Joshi calls a “bait-and-switch.”
The climate benefits companies tout relate almost exclusively to traditional AI—predictive models, computer vision, statistical methods. These can genuinely help: grid optimization, weather forecasting, logistics planning.
But the AI currently driving data center expansion is generative AI—LLMs, image generators, chatbots. Of the 154 claims analyzed, only four related to generative AI.
Sasha Luccioni, AI and Climate Lead at Hugging Face, puts it plainly: “When we talk about AI that’s relatively bad for the planet, it’s mostly generative AI and large language models. When we talk about AI that’s ‘good’ for the planet, it’s often predictive models or old-school AI models.”
The tech industry is selling the benefits of one technology while investing massively in an entirely different one.
The Origin of the Industry’s Favorite Number
One claim appears repeatedly: AI could help mitigate 5-10% of global greenhouse gas emissions by 2030—equivalent to the entire annual emissions of the European Union.
Google was still using this figure in April 2025. It sounds impressive. Where does it come from?
The report traced it back. It originates from a 2023 BCG report—commissioned by Google. That BCG report cites a 2021 BCG blog post. And the blog post? References “client experience” as its source.
This isn’t science. It’s circular marketing.
The Reality of Emissions
While the promises remain vague, the actual numbers are concrete.
Google:
- Emissions +48% since 2019
- Data center energy consumption +27% between 2023 and 2024
- Location-based Scope 2 emissions from 5.8 million tons CO₂ (2020) to over 11.2 million (2024)
Microsoft:
- Emissions +23.4% since 2020 baseline
- Scope 3 emissions (supply chain) account for 97% of total footprint
- Electricity consumption nearly tripled: from 10.8 million MWh (2020) to 29.8 million MWh (2024)
The industry overall:
- Data centers currently consume ~1% of global electricity
- Goldman Sachs: US share rises from 3% (2022) to 8% (2030)
- IEA: Data centers will account for at least 20% of electricity demand growth in developed countries by 2030
A single ChatGPT query consumes roughly 10x as much electricity as a Google search, according to Goldman Sachs. Jesse Dodge from the Allen Institute for AI calculates: “One query to ChatGPT uses approximately as much electricity as could light one light bulb for about 20 minutes. With millions of people using it every day, that adds up to a really large amount of electricity.”
The Playbook
Ketan Joshi sees a familiar pattern in the tech industry’s communication strategy: “These tactics are greenwashing and draw directly from the fossil fuel playbook.”
The structure is identical:
- Downplay negative impacts: “Yes, there’s an increase, but we’re working on it.”
- Promise vague future benefits: “AI will solve the problem.”
- Externalize responsibility: “We need more clean energy in the grid.”
- Report selectively: Avoided emissions (speculative) are highlighted, enabled emissions (real) go unreported.
Google, for instance, claims its products “enabled others to reduce 26 million tons of CO₂ emissions.” More than half of that (15 million tons) is attributed to Google Earth Pro—based on interviews with corporate partners, without independent verification.
Meanwhile, the emissions generated when AI tools are used for oil exploration or logistics optimization for fracking companies are captured nowhere.
What This Means for Enterprises
The report was presented at the AI Impact Summit in Delhi—at a time when regulators in the EU and US are beginning to look more closely.
For companies deploying or planning to deploy AI, three implications emerge:
1. Scrutinize vendor climate claims
When an AI vendor claims their product saves emissions, ask for specific, verified figures. Not projections. Not “potential.” Not references to other corporate reports.
The probability that the claim is based on solid evidence is approximately 26%.
2. Understand the distinction between AI types
Predictive models for energy optimization, predictive maintenance, or logistics planning can have genuine climate benefits. These are also significantly less energy-intensive than generative AI.
The question isn’t “Am I using AI?” but “What type of AI am I using, and for what?”
3. Measure your own AI footprint
Most companies have no idea how much energy their AI usage consumes. Before writing sustainability reports that present AI as a solution, you should know what part of the problem you’re currently enlarging.
The Uncomfortable Truth
Microsoft Chief Sustainability Officer Melanie Nakagawa offered a remarkably honest assessment in February 2025: “In 2020, Microsoft leaders referred to our sustainability goals as a ‘moonshot.’ Nearly five years later, we have had to acknowledge that the moon has gotten further away.”
That’s the opposite of what the marketing department says.
The technology industry faces a dilemma it cannot solve without being honest: It wants to demonstrate ESG leadership while simultaneously winning the AI race. But the AI race requires data centers that devour energy. And being honest means admitting that the supposed climate benefits of generative AI don’t exist—or at least aren’t demonstrable.
The Joshi report forces an uncomfortable question: If 74% of claims are unsubstantiated and not a single verified example exists where ChatGPT or Copilot reduced emissions—why do we still believe the story?
Sources: Beyond Fossil Fuels Report (Ketan Joshi, Feb 2026), Microsoft Environmental Sustainability Report 2025, Google Environmental Report 2025, Goldman Sachs Research, International Energy Agency, NPR, The Register, Fast Company