Big Tech Says Generative AI Will Save the Planet. It Doesn’t Offer Much Proof

Big Tech climate claims lack evidence

Energy researcher Ketan Joshi released a report Monday finding that the majority of corporate claims regarding artificial intelligence’s climate benefits lack scientific backing. The analysis examined 150 assertions from technology companies and energy associations regarding AI’s potential as a "net climate benefit."

The report finds that only 25 percent of these claims cited academic research. More than one-third of the examined statements provided no public evidence at all to support the idea that AI will reduce global emissions. Environmental organizations supported the research to quantify the gap between industry rhetoric and climate reality.

Joshi specifically targeted a widely circulated statistic from Google claiming AI could reduce global greenhouse gas emissions by 5 to 10 percent by 2030. Google’s chief sustainability officer coauthored an op-ed featuring this number, which press outlets and academic papers later repeated. Joshi’s investigation traced the figure back to a 2021 analysis by the Boston Consulting Group (BCG).

The BCG report relied on "experience with clients" rather than peer-reviewed data to estimate these massive reductions. BCG published this analysis a full year before the launch of ChatGPT triggered the current race for energy-intensive infrastructure. Joshi describes the evidentiary basis for this cornerstone industry claim as "flimsy."

Google emissions rise despite AI promises

Google recently admitted in its 2023 sustainability report that AI development is significantly increasing its corporate carbon footprint. Despite this internal data, the company continues to use the BCG 5 to 10 percent estimate in policy recommendations. The company included the figure in a memo sent to European policymakers last year.

Google spokesperson Mara Harris stated the company stands by its methodology and claims it is grounded in the "best available science." Harris provided a link to Google’s general methodology for calculating product emissions but did not explain how those standards apply to the BCG figures. BCG has not responded to inquiries regarding the data source.

The energy demands of the AI build-out are already altering the US power grid. In the world’s largest data center market, the surge in electricity demand is forcing coal plants to remain operational. Utilities are currently planning to add hundreds of gigawatts of new gas power to meet the requirements of the tech sector.

• 100 gigawatts of new power capacity in the US is currently earmarked specifically for data centers.
• Google’s corporate emissions have increased as it scales its AI infrastructure.
• The 5 to 10 percent reduction claim targets an amount of carbon equivalent to the entire annual output of the European Union.

Executives bet on hypothetical technology

Tech leaders continue to position AI as a necessary tool for environmental survival. At New York City’s Climate Week, the Bezos Earth Fund hosted sessions framing AI as an "environmental force for good." Former Google CEO Eric Schmidt recently argued that the world will likely miss its climate goals, making AI development more urgent.

Schmidt stated he would rather "bet on AI solving the problem" than constrain the technology’s growth. OpenAI CEO Sam Altman has also made public promises that AI will eventually "fix" the climate. These assertions often ignore the immediate environmental cost of training large language models (LLMs).

Jon Koomey, an energy and technology researcher, warns that self-interested claims from tech giants lack rigor. He notes that while some claims might eventually prove true, the public should not take them at face value. The industry currently relies on speculative future benefits to justify massive energy consumption today.

The report highlights a lack of transparency regarding the energy cost of specific AI tasks. Most companies only provide "back-of-the-napkin" estimates for data center power usage. Google only began releasing energy estimates for individual AI prompts last year, while many competitors disclose no environmental data at all.

Companies conflate different AI types

The tech industry frequently confuses traditional machine learning with modern generative AI when discussing climate benefits. Machine learning has assisted scientists for decades in fields like flood prediction and species tracking. These older models require significantly less power than the generative models currently driving the data center boom.

David Rolnick, chair of Climate Change AI, calls the speculation surrounding generative AI "grotesque." He notes that deep learning already helps increase grid efficiency and cut emissions in various sectors. However, these successes involve specialized models rather than the consumer-facing chatbots like Gemini or Claude.

Rolnick argues there is a mismatch between the technology Big Tech is building and the technology that actually helps the planet. Companies often use a flood-detection algorithm as a marketing example to justify the energy use of a massive LLM. These two technologies are fundamentally different in their architecture and energy requirements.

Researcher Sasha Luccioni argues the "bigger-is-better" narrative serves companies with the deepest pockets. These firms have spent decades collecting data and now claim that mammoth models are the only path forward. Luccioni’s research suggests that smaller, more efficient models often perform just as well for specific climate applications.

Experts demand granular energy disclosure

A separate study by Luccioni and Yacine Jernite of Hugging Face compared the costs of training various AI models. They found that massive proprietary models are not the only option for powerful AI solutions. Smaller models can be deployed at a fraction of the cost to both the planet and the companies using them.

The current lack of data makes it impossible for the public to verify if AI is helping or hurting the climate. Most tech firms do not separate the energy used for generative AI from their general data center operations. This lack of granularity hides the specific environmental impact of the newest, most power-hungry tools.

Joshi argues that the solution to this skepticism is simple: mandatory disclosure. If tech companies believe AI is a net positive, they should release specific data on their energy growth. Transparency would allow researchers to see exactly how many terawatt-hours are going toward generative AI versus other services.

The report concludes that without rigorous proof, AI climate claims remain marketing tools rather than scientific facts. The current trajectory suggests that AI is delaying the transition away from fossil fuels by increasing base-load power demand. Until companies prove otherwise with hard data, the environmental cost of AI remains its most measurable feature.