MIT Technology Review published an article by Mary Nichols and Jeremy Tamanini titled "The climate case for AI has not been made," timed to coincide with the UN climate conference in Antalya. The authors reviewed major companies in the energy, meat and dairy, and transport sectors and concluded that the specific AI strategies on which 2035 emissions reduction forecasts are based have only been implemented by a few. Working cases exist, but the climate case for AI at the industry level has not yet been proven.
What happened
The article was written by Mary Nichols (UCLA School of Law) and Jeremy Tamanini (Dual Citizen): they conducted an AI-assisted review of major players in the energy, meat and dairy, and transport sectors to check how many companies have deployed the solutions on which AI climate impact forecasts rely. The occasion was the UN climate conference in Antalya, where the host country called on nations to ensure that AI is designed, operated, and deployed in the interest of the climate. There were few confirmed production cases: in energy, only one major operator, UK National Grid ESO together with Open Climate Fix, has deployed working AI forecasting that halved the error in solar generation forecasts. Didi Chuxing reduced emissions per trip by 16% through AI routing, and CATL accelerated battery material screening by approximately 100 times. In alternative proteins, NotCo and Kraft Heinz use the Giuseppe platform, while venture funding for cultivated meat has dropped from $989M.
Context
The article's conclusion should be read against the specific forecasts it verifies: Google allows for a 5–10% reduction in global emissions by 2030, while the LSE Grantham and Systemiq assessment promises 3.2–5.4 billion tons of CO2 per year by 2035. Both figures assume the mass spread of AI strategies, not isolated pilots, so the key flaw in the forecasts is extrapolating pilot results to entire industries when the gap between demonstration and industrial scale has not actually been bridged. Meanwhile, the carbon footprint of data centers themselves is growing, and AI is being used by oil companies to increase fossil fuel extraction, so the net climate balance of the technology is currently unclear.
Why this matters for the industry
For the industry, the material raises the bar for evidence: claims about climate impact without deployment metrics become vulnerable to criticism, and the benchmark is set by verified production cases with measurable results. The gap between forecasts and implementation reveals a map of unoccupied niches: the market needs vertical tools integrated into existing industry operational processes that provide a verifiable metric — forecast accuracy, screening time, emissions per unit of service. Against the backdrop of the Antalya conference, a shift in regulator and investor demands from promises to deployment metrics and reporting with confirmed rollouts is expected, and in energy, attempts to copy the Open Climate Fix pattern are likely. Potential public pressure on Microsoft and Google due to lagging behind climate goals is currently a predictive interpretation, not an established fact.
Why this matters for users
For the reader, this is a sober filter: "AI will save the climate" remains more marketing than fact for now, and separating real mechanisms for emissions reduction from declarations should be done by a simple criterion — is there a deployed solution with a measurable impact metric. Where AI is actually working, the tasks are narrow and verifiable: generation forecasting, transport routing, material screening, so a reasonable reaction to a pitch or presentation is to ask for "deployed metrics" rather than forecasts. Another practical benchmark is to remember the growing carbon footprint of data centers and evaluate the net climate effect of the technology, not slogans.
What is still unknown / limitations
The authors' methodology is not described: the material does not show which models and criteria were used in the AI review, how companies were selected, or how the absence of deployments was verified — this is a serious limitation for a negative result. The CATL case is presented without a model, dataset, or comparison conditions, so its reproducibility requires separate verification from primary sources. The lack of mass deployment today does not in itself prove either failure or future effect: it is the same extrapolation for which the article criticizes other industry forecasts. Assessments of future actions by regulators and investors remain scenario assumptions, not confirmed facts.
Sources
Author
Look at AI, editorial team