AI’s climate benefits have yet to be proven

At the UN climate conference in Antalya, Turkey called on countries to ensure that AI is developed, powered and used in line with climate goals. But a review of public information about companies in the energy, meat and dairy, and transport industries found that only a few use AI specifically to cut emissions, and fewer still have verified results. A 2024 study estimated that AI could reduce global emissions by 3.2–5.4 billion tonnes of CO₂ a year by 2035, even after accounting for AI’s own energy use. For now, those figures remain a projection rather than an achieved outcome.

AI’s climate benefits remain a promise

From projections to evidence

In 2024, OpenAI CEO Sam Altman said AI would bring “astonishing breakthroughs,” including helping to “solve climate change.” A year earlier, Google’s head of sustainability and her co-author wrote in an opinion piece that AI could cut global emissions by 5–10% by 2030.

Last year, a peer-reviewed study led by researchers at the Grantham Research Institute at the London School of Economics and the company Systemiq estimated that AI could reduce annual emissions by 3.2–5.4 billion tonnes of CO₂ by 2035. The high end of that estimate is roughly equal to the US energy sector’s annual emissions. Most of the potential impact lies in energy, meat and dairy production, and transport.

The authors of the current analysis checked whether the largest companies in these industries are deploying AI in the ways underlying those estimates. They reviewed public information about companies using AI, and a second author checked the results. Many companies had announced early AI projects, but few reported using the specific emissions-cutting methods included in the study. Fewer still reported a measurable climate impact.

Energy: one verified example

In energy, the estimate of AI’s potential rests on the most concrete mechanism. More accurate forecasts of electricity supply and demand could increase by up to 20% the amount of renewable energy from existing solar and wind farms that reaches the grid. The study estimates this could prevent 1.8 billion tonnes of emissions a year.

But among the grid operators examined in China, the US and Europe, the analysis found only one known active program of this kind. The UK’s National Grid ESO, the national electricity system operator, worked with Open Climate Fix to halve the error in solar power generation forecasts. The error is the gap between the amount of sunlight forecast and the amount that actually reaches solar panels. According to the analysis, the project avoided hundreds of thousands of tonnes of CO₂ emissions a year.

Public materials from other grid operators in the EU, the US and China contained almost no examples of AI deployment matching the study’s scenario.

Alternative proteins: development and weak demand

To achieve the projected emissions reductions in meat and dairy, AI would need to help narrow the price and taste gap between familiar products and alternatives, such as plant-based or lab-grown burgers and milk substitutes.

One of the most prominent projects is a partnership between NotCo, an AI and food technology company, and Kraft Heinz. They formed a joint venture and use the Giuseppe platform, which analyzes molecular data from plants to identify combinations that could replicate the taste, texture and smell of animal products.

The analysis found that few other alternative protein producers have AI so deeply embedded in product development. The industry also faces more fundamental challenges: consumers are slow to switch to these products, and funding is shrinking. Venture capital investment in cultivated meat fell from a peak of $989 million in 2021 to about $83 million last year.

Transport: results in batteries and rides

The transport industry has examples of AI use that come closer to the study’s scenarios. They relate primarily to battery production and ride-sharing platforms.

CATL, which holds about 40% of the global electric vehicle battery market, launched an AI system trained on more than 100,000 battery designs and 600 terabytes of test data. The company says the system cut the time needed to screen promising materials by a factor of 100.

Independent researchers also found that ride-sharing platform Didi Chuxing used AI-optimized routing to reduce detours, cutting average emissions per trip by 16%.

Fossil fuels and AI’s own costs

AI’s climate impact depends not only on how it is used to expand clean energy. Oil and gas companies are also interested in these tools, including as a way to increase fossil fuel production. Technology companies that have previously announced climate goals should explain how their technologies could help both clean energy and the established hydrocarbon industry.

Deploying AI is only a first step. It remains unclear how much the technology will accelerate the improvements and development these industries need. It does not shorten the usual timelines for certification, permitting, construction, packaging, marketing and sales of new products, facilities or services.

The analysis was carried out using Anthropic’s Claude Code, and its findings were checked against public documents and media reports. Some companies and utilities may already be using AI in projects that fit the scenarios in the original study, but have not yet disclosed this information.

Wider and more targeted AI adoption, along with greater efficiency in the three industries, could bring the 2035 projections closer to reality. That will require sustained, focused effort to ensure AI’s environmental benefits outweigh the harm it causes.

Meanwhile, the rapid spread of AI models and construction of data centers have sharply increased energy consumption and emissions. Microsoft, Google and other major technology companies are consequently falling well behind their own climate goals. Under the International Energy Agency’s baseline forecast, emissions associated with the world’s data centers, which serve both AI workloads and other computing, could exceed 300 million tonnes a year by 2035. That makes it doubtful the industry can bring its climate emissions to net zero.

Industry leaders should deploy AI where it can cut emissions most. Corporate customers should choose models with the smallest environmental footprint and developers that are doing more to reduce their own climate impact. Governments and regions need to expand incentives or requirements that encourage companies to act more responsibly.

Mary Nichols is a distinguished advisor at the University of California, Los Angeles School of Law and a former chair of the California Air Resources Board. Jeremy Tamanini is the founder of Dual Citizen, a sustainability research and consulting firm.

Dataist’s take: The examples show that AI can already help with specific tasks in energy and transport, but they do not yet confirm the scale of emissions reductions projected in the study. At the same time, rising data-center energy use and the possibility of using AI to expand fossil fuel production complicate the technology’s climate balance.