Big Tech's AI Climate Claims Lack Evidence, Study Finds

Big Tech companies are making sweeping claims about artificial intelligence's potential to combat climate change, but a new analysis reveals these promises often lack scientific backing. Google alone claims AI could reduce global greenhouse gas emissions by 5-10% by 2030—equivalent to eliminating the entire European Union's emissions—yet this figure traces back to flimsy consulting reports rather than rigorous academic research.

Energy researcher Ketan Joshi investigated the origins of these claims and found troubling gaps in evidence. His new report, supported by environmental organizations, examined over 150 specific claims about AI's climate benefits made by tech companies and industry groups. The findings are stark: only 25% cited academic research, while more than one-third provided no public evidence whatsoever.

The Evidence Gap

Google's widely-cited 5-10% emissions reduction claim originates from a BCG consulting paper that relied on the firm's "experience with clients" rather than peer-reviewed research. Despite this weak foundation, the number has been repeated across media outlets, academic papers, and policy documents—including Google's own recommendations to European policymakers.

The timing is particularly concerning. This analysis was published before ChatGPT's launch, which triggered the current energy-intensive AI infrastructure buildout. Google has since quietly admitted that its AI expansion is significantly driving up corporate emissions, yet continues promoting the unsubstantiated 5-10% figure.

The Wrong Type of AI

A critical issue lies in what type of AI companies are actually discussing. Most climate benefits come from traditional machine learning applications that have been used in scientific disciplines for decades—tools for grid optimization, species discovery, and flood detection that require minimal energy.

However, the current AI boom centers on energy-intensive generative models like ChatGPT, Claude, and Google Gemini. These consumer-facing chatbots require massive data centers and power consumption but offer unclear climate advantages over more efficient alternatives.

Energy Reality Check

The infrastructure buildout for generative AI is driving significant energy demands:

• US data centers are keeping coal plants operational longer than planned
• Hundreds of gigawatts of new gas power are planned for the grid
• Nearly 100 gigawatts of that power is earmarked solely for data centers

Tech executives continue defending this energy investment by promising future climate benefits, but concrete examples of how large-scale generative AI outperforms less energy-intensive models remain scarce.

The Path Forward

Experts emphasize the need for transparency. Companies should disclose specific energy usage for different AI applications and provide evidence-based assessments rather than relying on marketing claims.

"If tech companies are worried that people are overstating the climate impacts of generative AI, then there should be nothing stopping them from providing detailed energy breakdowns," Joshi notes.

The solution isn't abandoning AI for climate applications—traditional machine learning already delivers proven benefits. Instead, the focus should be on deploying the right AI tools for specific problems rather than assuming bigger, more energy-intensive models are automatically better.

🔗 Read the full investigation from WIRED