A scathing new report released on February 17, 2026, has cast a shadow over the technology industry’s narrative that artificial intelligence (AI) will be a primary savior in the fight against climate change. The study, commissioned by a coalition of non-profits including Beyond Fossil Fuels and Climate Action Against Disinformation, analyzed 154 public statements from major tech firms and international bodies. It found that a staggering 74% of claims regarding AI’s climate benefits are unproven, with 36% citing no evidence at all and only 26% supported by peer-reviewed academic research.
The report, authored by energy analyst Ketan Joshi and presented at the AI Impact Summit in Delhi, accuses industry giants of using "diversionary" tactics to mask the environmental toll of the generative AI boom. According to Joshi, tech companies are systematically conflating "traditional" machine learning—which has been used for years in weather forecasting and grid optimization—with the new, energy-intensive generative AI models like Google’s Gemini and Microsoft’s Copilot. The analysis did not find a single instance where these popular generative tools led to a "material, verifiable, and substantial" reduction in global emissions.
The timing of this report is critical as the U.S. leads a record global surge in gas-fired power, largely driven by the insatiable electricity demands of AI data centers. While the International Energy Agency (IEA) previously suggested AI could reduce global emissions by up to 5% by 2035, the new findings suggest such projections are often based on anecdotal evidence or corporate blog posts rather than rigorous science. For instance, a widely cited claim that AI could mitigate 5-10% of global greenhouse gas emissions by 2030 was traced back to a 2021 consulting report based on "experience with clients" rather than empirical data.
The core of the controversy lies in the fundamental difference between predictive and generative models. Sasha Luccioni, AI and climate lead at Hugging Face, noted that while "old-school" AI can indeed optimize energy use, the large language models (LLMs) driving current market valuations are inherently "bad for the planet" due to their massive carbon and water footprints. Data centers, which currently consume roughly 1% of global electricity, are projected to see their share of U.S. electricity demand more than double to 8.6% by 2035. In some regions, this growth is already forcing utilities to delay the retirement of coal plants or build new natural gas infrastructure, directly contradicting the climate goals championed by U.S. President Trump’s administration and global tech leaders.
From a financial perspective, this "greenwashing" represents a significant risk for ESG-focused investors. If the climate benefits of AI are indeed marginal, the massive capital expenditures (CapEx) being poured into data centers may eventually face regulatory headwinds or carbon taxes that are not currently priced into the market. Google has defended its position, stating its emissions estimates are based on a "robust substantiation process," but the lack of transparency regarding the specific energy consumption of complex tasks—such as video generation or deep research—remains a point of contention for analysts.
Looking forward, the tension between AI expansion and climate commitments is likely to reach a breaking point. As data centers begin to account for a projected 20% of electricity demand growth in developed nations through 2030, the "carbon debt" created by training and running these models will become harder to ignore. Analysts predict that unless the industry shifts toward radical transparency and moves away from speculative climate claims, it may face a wave of litigation similar to that seen in the fossil fuel sector. The current trajectory suggests that AI’s primary climate impact in the near term will not be a reduction in emissions, but rather a significant stress test for global power grids and the credibility of corporate sustainability pledges.
Explore more exclusive insights at nextfin.ai.
