The Unproven Promise: Tech's AI Climate Claims Face Scrutiny

In 2023, Google made a striking assertion: artificial intelligence could slash global greenhouse gas emissions by 5 to 10 percent within seven years. That figure, equivalent to the annual emissions of the European Union, was widely circulated. Yet, according to a new report from energy researcher Ketan Joshi, the evidence behind this bold claim is surprisingly thin.

Joshi's investigation, supported by several environmental groups and released this week, traces the statistic's origin. It leads back to a Google-commissioned paper by consultancy BCG, which itself cited only the firm's "experience with clients" as a basis. Joshi describes this foundation as "flimsy," particularly given that the analysis predated the ChatGPT-driven explosion in energy-intensive AI infrastructure. Notably, Google's own 2023 sustainability report later acknowledged that its AI expansion was significantly increasing its corporate emissions, even as the company continued to promote the BCG numbers to policymakers.

Joshi’s report examined over 150 public claims that AI will be a net climate benefit. His analysis found only a quarter were backed by academic research; more than a third cited no public evidence at all. This gap highlights a critical tension: the rapid build-out of data centers, which is prolonging the life of coal plants and adding massive new gas capacity to power grids, is justified by promises of a future climate payoff.

Experts note a troubling pattern in these promises. "The narrative tries to sell us the idea that this is the only kind of AI we need, and the only future that's possible," says AI sustainability researcher Sasha Luccioni. She points out that many proven, less energy-intensive machine learning models already aid climate science, from grid optimization to species discovery. However, tech companies often conflate these with the large-scale generative AI, like ChatGPT, that drives their infrastructure spending.

David Rolnick, a computer science professor at McGill University, observes that many corporate claims rely on "hypothetical AI that does not exist now." He calls the amount of speculation "grotesque."

The core issue, Joshi argues, is a lack of transparency. Without clear data on how much energy specific AI models consume, the public and policymakers cannot assess the true cost. His report calls for detailed disclosure, suggesting that if tech companies believe in AI's climate potential, they should openly report its energy footprint. As the AI race accelerates, the demand for proof is growing louder.

Source: Wired science