Opinion: the climate case for AI has not yet been made

This opinion piece, published in MIT Technology Review, is by Mary Nichols, a distinguished counsel at UCLA School of Law and former chair of the California Air Resources Board, and Jeremy Tamanini, founder of the sustainability research and advisory firm Dual Citizen. It argues that AI leaders' climate promises are not yet backed by evidence. The piece comes out during the UN climate conference in Antalya, Turkey, where the host country has called on nations to commit to designing, powering and deploying AI in ways that support climate action.

The authors recall some highly optimistic claims from the industry. In 2024, OpenAI CEO Sam Altman spoke of AI's "astounding triumphs," including "fixing the climate." In 2023, Google's head of sustainability and a co-author wrote in an op-ed that AI could cut global emissions by 5% to 10% by 2030.

For an academic counterpoint, they cite a peer-reviewed study published last year, led by researchers at the London School of Economics' Grantham Research Institute and the firm Systemiq. It estimates that AI could help cut between 3.2 and 5.4 billion tonnes of CO2 equivalent per year by 2035, even after accounting for AI's own energy footprint. The high end is comparable to annual US energy-related emissions. Those gains would be concentrated mainly in three sectors: electricity, meat and dairy, and vehicles.

The authors ask whether the largest companies in those sectors are adopting AI in the ways those projections require. To answer this, they conducted an AI-assisted review of public information from major players in each sector, carried out and checked by the second author. The analysis was performed by Anthropic's Claude Code, and its findings were checked against public documents and press reports. They conclude that few companies have announced they are pursuing the specific emissions-reduction pathways the projections rely on, and even fewer claim to have made significant climate gains through them. The authors acknowledge that some companies may be working on applications that are not yet public.

The study's numbers are most concrete for the power sector. There, AI-driven supply and demand forecasting could increase by up to 20% the renewable electricity that reaches the grid from existing solar and wind installations, avoiding about 1.8 billion tonnes of emissions a year. Yet of the grid operators reviewed in China, the US and Europe, only one had a significant AI program of this kind up and running: the collaboration between National Grid ESO (UK) and Open Climate Fix. It halved solar forecasting error and avoids hundreds of thousands of tonnes of CO2 a year. The other operators analyzed had little to show on the public record.

According to the authors, the meat and dairy sector offers the starkest contrast between technological maturity and market reality. To deliver the projected reductions, AI would need to close the taste and cost gap between traditional proteins and alternatives, whether cultivated or plant-based. One of the clearest efforts is by NotCo, which has formed a joint venture with Kraft Heinz using its Giuseppe platform. The platform analyzes molecular plant data to find combinations that replicate the taste, texture and smell of animal products. But few other alternative-protein companies use AI so extensively, and the sector faces deeper problems: slow consumer adoption and shrinking funding. Venture capital investment in cultivated meat fell from a peak of $989 million in 2021 to about $83 million last year.

Transportation offers more encouraging examples, particularly among battery makers and ride-hailing platforms. CATL, which controls around 40% of the global EV battery market, launched an AI system trained on more than 100,000 battery design cases and 600 terabytes of test data. The company says it cuts the time needed to screen candidate materials by two orders of magnitude. Meanwhile, independent researchers found that Didi Chuxing cut emissions per trip by 16% on average using AI-optimized routes that minimize detours.

The authors flag another problem: oil and gas companies are also interested in these tools, partly for their potential to increase fossil fuel production. Tech companies that have made sustainability commitments in the past should explain how their tools could be used to boost traditional dirty energy as well as clean energy. They add that adopting AI is only the first step. It is unclear how far AI can speed up the projected improvements, since it barely shortens the usual processes of certification, permitting, construction, packaging, marketing and sales.

Finally, the piece stresses that the rapid rollout of AI models and the construction of data centers have sent energy use and emissions soaring. As a result, major tech companies such as Microsoft and Google have fallen far short of their corporate climate targets. Under the International Energy Agency's base scenario, emissions from data centers, which can host both AI and non-AI workloads, could exceed 300 million tonnes a year by 2035. The authors say it would be surprising if the industry even managed to offset its own climate pollution.

They conclude that better, more targeted uses of AI in these three sectors could bring the 2035 projections closer to reality. But that will not happen without deliberate, sustained efforts to ensure AI delivers more environmental benefit than harm. They urge industry leaders to deploy the technology where it cuts emissions the most. They call on corporate customers to choose models with lower environmental impact and to work with the developers doing the most to shrink their footprint. And they ask states and regions to encourage or require more responsible behavior from all of these companies.