Version 2 of this summary, as approved on . It weighed 60 gains, 56 problems and 20 Traces. This is the current version. See the topic as it stands today
Company reports show Microsoft, Amazon and Google emissions rising from about 101m to 119m tonnes of CO2e in the year to March 2026, which the reports link to datacentre construction for AI and cloud. TRV-2026-0296 Most evidence that AI helps the climate comes from benchmarks and simulations, and modelling studies suggest that AI infrastructure adds emissions before any savings catch up. TRV-2026-2616TRV-2026-2777TRV-2026-2626
A systematic review and meta-analysis of experimental and modeling studies shows AI-driven irrigation cut water use by 30% to 50% and raised crop yields by 20% to 30%. TRV-2026-2738 A meta-analysis found reinforcement-learning building controls cut energy use by an average 22.3% against conventional systems, though the authors say most studies were simulations or single-season tests. TRV-2026-2777 In a five-month test against 123 Shandong weather stations, Pangu-Weather started from ECMWF data cut 2 m temperature error by about 13.2% against ECMWF's own operational forecasts. TRV-2026-2492 On the other side, a review of 146 studies found AI's expansion in energy systems also raises electricity and infrastructure demand. TRV-2026-1905
A UNESCO and University College London test of 15 popular AI models found that eight times out of 10, a model switched to a wrong climate answer at the first sign of mild user disagreement. TRV-2026-2968 In a test by the developer's own staff, an AI carbon-footprint modeler had a 33% median error against 269 product declarations. TRV-2026-2490 Researchers found emission-classification models scored 0.91 accuracy on a random country-year split but only 0.637 to 0.676 on countries they had not seen. TRV-2026-1743 Over more than four years of real-time forecasts, random forests built on HRRR rainfall forecasts did not match the operational day-one excessive rainfall outlook. TRV-2026-2655
The 99.95% accuracy reported for spotting apple leaf diseases comes from image datasets, not from farms. TRV-2026-0262 In a simulation, a machine learning controller cut data-center cooling energy by 28.1% on average, but real-world sensor noise and actuator delays are not fully captured. TRV-2026-2626 A review of 103 studies found that AI's water use, Scope 3 value-chain impacts and end-of-life hardware are thinly covered. TRV-2026-2630 The typhoon track result rests on 14 storms from the 2024 Northwest Pacific season, and intensity errors persisted across multiple cases. TRV-2026-2618
Across 67 countries, greater AI development was associated with significantly lower ecological footprints and carbon emissions, and a study of 51 countries links greater AI use to lower physical climate risks. TRV-2026-0377TRV-2026-1231 A study of 1,090 data centers in 162 Chinese cities found a rise in an AI index was associated with extra operational carbon emissions, and a separate China study found AI robotics did not help lower CO2. TRV-2026-2878TRV-2026-2379 Among Chinese listed firms, real AI investment was associated with lower pollution intensity, while AI talk beyond actual investment was associated with higher intensity. TRV-2026-2116
Company reports, modelling studies and scenario projections point to rising electricity, carbon and water use, but most figures are modelled rather than metered. TRV-2026-2498TRV-2026-2616TRV-2026-0296TRV-2026-0474TRV-2026-2599TRV-2026-2582
Polls, ballot measures, local moratoriums and protests show organised opposition to AI datacentres in the US and Europe. TRV-2026-0921TRV-2026-2965TRV-2026-1553TRV-2026-2407TRV-2026-1401TRV-2026-2632
AI models often beat or match conventional forecasts on temperature and storm tracks but lag on heavy rain and storm intensity. TRV-2026-2492TRV-2026-2612TRV-2026-2514
Lab and simulation tests show 4-bit models, cache resizing and cooling control cutting energy or carbon, while one rebound experiment found most expected savings eroded. TRV-2026-1741TRV-2026-2628TRV-2026-2624TRV-2026-2626
Irrigation shows large gains in a systematic review, while crop disease and pollution detection tools mostly report strong scores on datasets with little field testing. TRV-2026-2738TRV-2026-0262TRV-2026-1399TRV-2026-0894TRV-2026-1989TRV-2026-0926
Country and firm studies are observational and point both ways, and building control reviews report savings that rest largely on simulations. TRV-2026-0377TRV-2026-1231TRV-2026-2878TRV-2026-2379TRV-2026-2777TRV-2026-2116
Each record sits in one thread at most. A record with both a gain and a problem counts on both sides. 17 records fit none of these threads.